25 ChatGPT and Codex Prompts for Local Newsrooms: Archive Search, Public-Meeting Monitoring, Translation, Audience Growth, Revenue, and Editorial Review

A safe prompt library for authorized local-newsroom work

This prompt library is built for local newsrooms and the adjacent teams that make journalism sustainable: editors, reporters, audience staff, product managers, developers, data analysts, revenue teams, advertising staff, donor and membership teams, translators, archivists, and operations leaders. It is not a shortcut around reporting, consent, records law, copyright, source verification, or editorial accountability. Each prompt is a structured request that can help an authorized team organize evidence, generate drafts, identify leads, review workflows, or prepare questions for a human decision-maker.

The factual boundary for this guide is OpenAI’s reporting on how news organizations are using AI, OpenAI’s announcement about the Lenfest AI Collaborative expansion, OpenAI’s prompt-engineering guidance for developers, and OpenAI’s usage policies. OpenAI and Lenfest describe newsroom uses that include archive search, public-interest monitoring, translation, audio transcription, audience engagement, investigative research, advertising prospecting, donor modeling, audience personalization, subscription growth, print-to-digital transition, and operational efficiency. Those examples are useful, but they are not a license to treat model output as verified reporting or to extend a tool into records, audiences, donors, advertisers, or sources without authorization.

The safest framing is simple: ChatGPT and Codex can help a newsroom ask better questions of material it is allowed to use. They should not be treated as eyewitnesses, public-record custodians, translators of record, audience measurement systems, legal reviewers, human editors, or substitutes for direct verification. In this article, every output should be treated as a draft, a lead, a checklist, a synthesis, or a proposed next step until a named human owner verifies it against original sources and newsroom policy.

This article explains Offline Web Search in ChatGPT Workspaces, including indexed sources, freshness limits, lockdown mode, evidence handling, and fallback procedures. The Offline Web Search in ChatGPT Workspaces: Indexed Sources, Freshness Limits, Lockdown Mode, Evidence, and Fallbacks article is a focused companion for Newsroom Archive Search because it is the strongest match for newsroom archive search because it focuses on controlled indexed-source retrieval and evidence boundaries rather than unrelated code or legal search.

The prompts that follow are written as contracts rather than clever one-liners. A good newsroom prompt tells the model its role, defines the task, lists authorized inputs, states what must not be invented, requires citations and uncertainty labels, and names the human review step before publication or consequential action. This structure follows OpenAI’s prompt-engineering emphasis on clear instructions, context, examples where useful, and explicit output formats. For a newsroom, the output format is not cosmetic; it is part of the control system that helps editors see what is sourced, what is uncertain, and what still requires reporting.

The locked contract for every prompt in this guide

Every prompt in this library is governed by the following contract. If a team copies only one part of this article into its internal AI policy, copy this section first. The contract applies whether the prompt is used in ChatGPT, in a Codex-assisted workflow, or in an internal tool built on OpenAI’s API. It also applies whether the task concerns editorial reporting, audience research, product analysis, subscription experiments, advertising prospecting, donor analysis, or workflow documentation.

Locked prompt contract: Use only data, archives, systems, accounts, documents, transcripts, meetings, metrics, and workflows that the user owns or is explicitly authorized to access. Treat all outputs as drafts or leads. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy and consent checks, copyright and licensing review, bias checks, redaction where needed, evidence capture, and a named human editor or accountable owner before external publication or consequential action.

This contract is intentionally conservative because local newsrooms operate under tight trust constraints. A single fabricated quote, misread public-meeting transcript, overconfident translation, unsupported donor inference, or misleading advertising claim can harm a person, damage a newsroom’s credibility, or create legal and ethical exposure. AI systems can be useful for summarizing, clustering, extracting, drafting, and comparing, but those activities must stay inside a verification process that is owned by humans and documented in the newsroom’s workflow.

For editorial teams, “consequential action” includes publication, social distribution, newsletter sends, push alerts, corrections decisions, source outreach based on sensitive information, legal threat letters, public-records requests that identify private individuals unnecessarily, and any story framing that could materially affect a person’s reputation or safety. For revenue teams, consequential action includes advertiser claims, donor segmentation, audience targeting, renewal campaigns, pricing experiments, sponsorship packages, and any personalization that could use sensitive or protected attributes. For product teams, it includes permission changes, automated routing, account-level targeting, and analytics decisions that affect user experience or access.

What these prompts are allowed to do

The prompts are designed to support authorized newsroom labor that already has an accountable process. They can help a reporter search a licensed archive, compare a public-meeting agenda against prior coverage, extract action items from a transcript, generate a list of follow-up questions, identify missing context in a draft, or build a translation review checklist. They can help an audience editor summarize open-ended survey comments, propose newsletter subject-line variants for human review, or map questions readers are asking about a beat. They can help a revenue team prepare a fact-checked advertiser brief or organize donor-engagement hypotheses without inventing personal characteristics.

For developers and product teams, the same contract can be converted into application behavior. An internal tool can require users to attach source documents, display citation fields, label uncertainty, log the human approver, and block export until privacy and copyright checks are completed. A Codex-assisted engineering workflow can help draft a script for parsing meeting agendas or cleaning archive metadata, but the developer remains responsible for permission review, test data selection, code review, access controls, and deployment approval.

For newsroom leaders, the prompt library can become a training artifact. Editors can ask staff to show the prompt, the inputs, the output, the original source material, and the human verification notes during story meetings. That practice makes AI use inspectable rather than hidden. It also aligns with OpenAI’s reported lesson from newsroom work that successful projects should start with a clearly defined problem, proceed through small-scale experimentation, use ongoing evaluation, and preserve human editorial judgment.

What these prompts must never do

These prompts must never be used to invent material that appears to come from reporting. Do not ask a model to “create a quote” for a named resident, “reconstruct” what a public official probably said, “fill in” missing public-record contents, or “guess” the finding of a document the newsroom has not obtained. If a transcript is incomplete, the output must say that it is incomplete. If an archive search did not return a source, the output must say no provided source supports the claim. If a translation contains ambiguity, the output must preserve that ambiguity and send the passage to a qualified human reviewer when it matters.

The prompts must not be used to infer sensitive traits or target people using protected or highly sensitive personal data. For donor, membership, subscription, audience, or advertising work, teams should avoid instructions that infer health status, religion, race, ethnicity, sexual orientation, political persuasion where inappropriate, immigration status, financial vulnerability, or other sensitive characteristics unless there is lawful authority, documented consent, and a clear organizational policy approved by responsible leadership. Even then, a local newsroom should ask whether using that data is necessary, proportionate, and consistent with reader trust.

The prompts must not be used to bypass access controls, scrape systems without permission, evade rate limits, launder copyrighted material, or republish licensed archives beyond allowed uses. A newsroom may have rights to search its own archive for internal reporting purposes while lacking rights to republish photographs, wire copy, syndicated material, or licensed databases in a new product. Each prompt that touches archives, transcripts, translations, images, audio, or external datasets should trigger a copyright and licensing review before external publication or product reuse.

The prompts also must not be used to automate public-facing claims without human approval. That includes ad copy, sponsorship proposals, fundraising emails, newsletters, social posts, push alerts, corrections, editor’s notes, public-records demand letters, legal correspondence, and translation intended for publication. AI can prepare a draft or checklist, but a named human owner must review evidence, tone, fairness, accuracy, consent, and policy compliance before anything leaves the organization.

How to read each prompt section

Each prompt section in the full library uses the same five labels: Purpose, Copy-paste prompt, Required inputs, Expected output, and Verification checkpoint. The labels are operational controls. They make the prompt easier to audit, easier to train on, and easier to adapt into an internal form, ticket, content-management-system workflow, or API-backed tool.

Section label	Why it matters in a newsroom	Human control it supports
Purpose	Defines the newsroom problem before invoking AI, reducing vague experimentation and preventing task creep.	Editor, product lead, or team manager can decide whether the use case is legitimate and proportionate.
Copy-paste prompt	Provides a reusable instruction that includes scope, prohibitions, evidence requirements, and output format.	Staff can use a consistent contract instead of ad hoc prompts that omit verification and privacy rules.
Required inputs	Lists the authorized documents, archives, metrics, transcripts, or policies needed to run the task responsibly.	Teams can stop the workflow when a required source, permission, date, or policy basis is missing.
Expected output	Describes the form of the draft, lead list, checklist, table, or analysis the model should return.	Editors and managers can compare the output against a known review format.
Verification checkpoint	States what must be checked against original evidence before publication, outreach, product change, or revenue action.	A named human owner signs off, requests more reporting, or rejects the output.

This format also makes it easier to reject bad outputs. If a model returns an uncited claim, invented statistic, unsupported audience segment, fabricated quote, ambiguous translation presented as certain, or a donor attribute that was not present in authorized data, the reviewer can mark the output as noncompliant with the prompt rather than debating whether it is “good enough.” That distinction matters because newsroom AI quality is not only about fluency; it is about evidence, traceability, and accountability.

Minimum verification rules before using any output

Before any output is used outside a private draft workflow, the assigned human owner should run a minimum verification pass. First, identify every factual claim and attach the original source, date, and location in the document, transcript, archive record, dataset, or meeting record. Second, label claims that are unresolved, inferred, disputed, or dependent on incomplete records. Third, remove or redact unnecessary personal data. Fourth, check whether the material is covered by copyright, license restrictions, confidentiality terms, embargoes, source agreements, or internal policies. Fifth, review for bias, representational harm, missing voices, and disproportionate targeting.

For public-meeting monitoring, the verification pass must distinguish an agenda item from an adopted motion, a proposal from a vote, a staff recommendation from board action, and a public comment from an official record. For archive search, it must distinguish prior reporting from current facts; a story from 2018 may identify context but not prove that a condition remains true today. For translation, it must distinguish literal wording from idiom, uncertain proper nouns, culturally specific references, and passages requiring a qualified bilingual reviewer. For audience and revenue work, it must distinguish observed engagement metrics from inferred intent or identity.

Every verification pass should leave an evidence trail. That can be a note in the story file, a CMS field, a spreadsheet column, a product ticket, a code review comment, or a revenue-operations record. The record should name the human reviewer, list the sources checked, describe unresolved uncertainty, and state the approved next step. Local newsrooms do not need heavy bureaucracy for every internal experiment, but they do need enough documentation to explain how a public claim, donor action, audience segment, or tool change was reviewed.

This article covers building an evidence-based AI skills program using baselines, practice tasks, human review, and clear badge boundaries. The Build an Evidence-Based AI Skills Program from OpenAI Academy: Baselines, Practice Tasks, Human Review, and Badge Boundaries article is a focused companion for Editorial Human Review because it directly supports the editorial review marker by emphasizing human review as a required operating practice when deploying AI workflows.

Recommended newsroom operating policy for this prompt library

Recommendation: Treat these prompts as approved starting templates, not as self-executing authorization. A newsroom should maintain a short AI-use policy that says who may use AI tools, what classes of data may be uploaded or connected, which workflows require editor approval, which workflows require legal or privacy review, and where evidence must be stored. The policy should also state that a staff member may decline to use AI for a task when the available inputs are too sensitive, incomplete, biased, or poorly licensed.

Recommendation: Assign ownership by role, not by tool. A reporter owns reporting accuracy; an assigning editor owns story framing and publication readiness; a copy editor owns style and final checks within the newsroom’s process; a product manager owns product experiments; a developer owns code review and deployment hygiene; a revenue leader owns advertising, sponsorship, and donor-action approvals. ChatGPT, Codex, or an API workflow can support those people, but the system must not become the accountable editor, fundraiser, advertiser, translator of record, or public-records analyst.

Recommendation: Start with small-scale experiments and ongoing evaluation. OpenAI’s Lenfest-related reporting emphasizes starting with a defined problem and guardrails rather than beginning with the technology. A practical local-newsroom pilot could test one archive-search prompt on one beat, one public-meeting-monitoring prompt for one municipality, or one translation-review prompt for one newsletter. The team should evaluate false positives, missed context, time saved or lost, staff confidence, reader risk, and the quality of the human review record before expanding the workflow.

Recommendation: Separate editorial uses from revenue uses. The same newsroom may responsibly use AI to find prior coverage and to organize advertiser prospects, but those workflows should not share sensitive editorial notes, source identities, unpublished reporting, donor attributes, or individual reader behavior unless there is explicit authorization, lawful basis, and policy approval. Editorial independence and reader trust require data boundaries that are clearer than a generic “internal use” label.

A practical intake checklist before running a prompt

Before a staff member runs one of the 25 prompts, they should answer the following intake questions. If the answer to any permission, source, or review question is uncertain, pause the workflow and ask the appropriate editor, manager, privacy lead, legal reviewer, or systems administrator. This is especially important for small newsrooms where one person may wear several hats and informal practices can become accidental policy.

1. Authorization: Do I own or have explicit permission to use the documents, transcripts, data, archives, systems, or accounts involved in this task?
2. Purpose: Can I state the newsroom, audience, product, or revenue problem in one sentence without saying “because AI can do it”?
3. Input quality: Are the source dates, document origins, transcript completeness, dataset definitions, and metric time ranges clear enough to support the task?
4. Privacy: Does the task include personal data, confidential source material, minors, victims, health information, financial hardship, immigration status, or other sensitive content that requires extra review or redaction?
5. Consent and policy: Does the organization have a documented basis for using this data in this way, especially for donor, audience, subscription, or advertising work?
6. Copyright and licensing: Are archives, audio, photographs, wire copy, syndicated material, meeting recordings, or third-party datasets restricted by license terms?
7. Bias and fairness: Could the prompt or input data overrepresent certain neighborhoods, languages, institutions, income levels, political groups, advertisers, or donors?
8. Output use: Will the result remain an internal draft or lead, or could it influence publication, outreach, targeting, pricing, fundraising, or another consequential action?
9. Human owner: Who is the named person responsible for checking the evidence and approving the next step?
10. Evidence record: Where will the prompt, input description, output, source citations, uncertainty notes, and approval record be stored?

This intake process is not meant to slow down ordinary reporting; it is meant to prevent hidden escalation. A staffer summarizing a public agenda for a morning meeting may need only a lightweight check. A team using donor data, reader behavior, or advertiser prospects needs a more formal review because the risk shifts from drafting assistance to consequential targeting and revenue action.

How the prompts handle uncertainty, citations, and dates

Every prompt asks for citations, dates, and uncertainty labels because local journalism often depends on temporal precision. “The council approved the budget” is different from “the council is scheduled to consider the budget.” “The company said in 2021” is different from “the company currently says.” “A reader survey suggests concern” is different from “the audience wants.” The prompt format pushes the model to identify what the provided evidence says, when it said it, and where the reviewer can check it.

Sample uncertainty labels: Use “confirmed from provided source” when a claim is directly supported by an attached or cited source; “requires original-source check” when the model is summarizing extracted text or a secondary record; “incomplete record” when a transcript, agenda, dataset, or archive set has gaps; “interpretation” when the statement is an analysis rather than a directly stated fact; and “do not use externally” when a claim concerns private, sensitive, licensed, or insufficiently verified material. These labels are examples, not a substitute for newsroom policy.

Sample citation expectation: A useful output should cite the agenda date, meeting body, file name, transcript timestamp, archive story date, dataset field, CMS item ID, survey question, or internal policy section used to support each material claim. If the model cannot cite a claim to the provided material, it should say so. A missing citation is not a minor formatting error; it is a signal that the claim may be unsupported, inferred, or fabricated.

Where ChatGPT and Codex fit in a newsroom workflow

ChatGPT is best suited in this library for language, synthesis, classification, summarization, planning, review, and structured drafting when a human supplies or identifies the authorized evidence. Codex is best suited for developer-supervised workflows such as writing scripts, tests, parsers, data-cleaning utilities, internal dashboards, or documentation around newsroom systems. In both cases, the model is a tool inside an accountable workflow, not an independent actor with authority to publish, contact sources, change permissions, spend money, submit forms, or alter production systems.

A practical pattern is to use ChatGPT for editorial reasoning and Codex for implementation support while keeping both inside permission boundaries. For example, an editor might use a prompt to define the fields needed for a public-meeting digest, and a developer might use Codex to draft a parser for agenda files stored in an authorized internal folder. The editor verifies the editorial logic; the developer reviews the code, tests it against safe sample files, checks access controls, and obtains deployment approval. Neither tool should be given broader data access than the task requires.

Another practical pattern is to separate “lead generation” from “publication.” A prompt can produce a ranked list of possible story leads from public meeting agendas, but the newsroom should not publish that list as fact. A reporter should inspect the agenda, watch or read the meeting record when available, contact relevant parties, seek missing context, and verify whether the item is new, significant, and fairly framed. This separation protects against a common failure mode: a fluent summary that converts a weak signal into an overstated story.

Using the 25 prompts responsibly

The full set of 25 prompts is organized around the work local newsrooms actually do: searching archives, monitoring public meetings, reviewing transcripts, translating and localizing coverage, ranking leads, planning investigations, understanding audiences, designing subscription experiments, supporting advertising and donor operations, documenting workflows, and reviewing drafts before publication. The common thread is not automation for its own sake; it is disciplined assistance with human-owned tasks where evidence, consent, fairness, and review remain visible.

As you adapt these prompts, keep the prohibitions intact. Do not remove the lines that forbid fabrication. Do not remove the requirement for citations, dates, uncertainty, privacy review, copyright review, bias checks, redaction, evidence capture, and named human ownership. Do not convert a draft prompt into an auto-send workflow for external messages, campaigns, publication, payments, purchases, bookings, legal commitments, or destructive system changes. The safest version of this library is the one that makes review unavoidable rather than optional.

The next section begins the copy-paste prompt contracts. Each one can be used as written, narrowed for a specific desk or beat, or converted into an internal form. Replace placeholders with authorized files, systems, metrics, and policies. If a required input is missing, the correct response is not to ask the model to improvise; it is to pause, obtain the source, narrow the task, or document why the output cannot be used.

Prompts 1–9: archive search, public records, meetings, interviews, evidence, and translation

Prompt 1: Archive query planning

Purpose

Use this prompt when a reporter or librarian needs a structured search plan for an authorized newsroom archive before opening the archive interface, vector index, CMS search, document database, or Codex-assisted retrieval script. OpenAI’s reporting on newsroom AI use highlights archive search as one practical workflow, but the safe newsroom version starts with query planning rather than asking a model to declare what happened in the archive. The output should be a search strategy with synonyms, date ranges, entity variants, exclusion terms, and verification notes, not an invented summary of past coverage.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom user with archive query planning.

Authorization contract:
I am only asking you to help plan searches across archives, indexes, files, databases, CMS records, transcripts, notes, or systems that my newsroom owns, licenses, or is explicitly authorized to use. Do not ask for passwords, tokens, private keys, account numbers, privileged legal material, confidential source identities, or unnecessary personal data. If an input appears outside this authorization boundary, stop and ask for a safer, authorized substitute.

Placeholder-only input contract:
Use only the placeholders and descriptions I provide. Do not invent names, quotes, articles, documents, sources, dates, archive hits, public-record contents, audience data, donor attributes, advertising claims, or interview material. Treat bracketed values such as [TOPIC], [PLACE], [DATE_RANGE], [ARCHIVE_SYSTEM], [KNOWN_NAMES], and [EXCLUDED_TERMS] as placeholders.

Source verification contract:
Your output must be a search plan, not factual findings. Require every eventual archive hit to be checked against the original item, publication date, byline, headline, corrections note, version history if available, rights/licensing status, and any cited primary source. Label any proposed query term as "search lead," not evidence.

Rights and copyright contract:
Flag copyright, licensing, syndication, wire-service, photo, audio, transcript, and database-use issues that may limit reuse, republication, training, extraction, or quotation. Do not suggest copying full articles or restricted materials into external systems unless the newsroom has documented rights and policy approval.

Privacy and consent contract:
Minimize personal data. If the topic involves minors, victims, private residents, health, finances, immigration, education, crime, employment, or other sensitive contexts, recommend redaction and editor review before any material is exported, summarized, or shared.

Uncertainty contract:
State what the search plan can and cannot establish. Distinguish likely entity variants from confirmed identities. Note missing date ranges, incomplete archives, OCR errors, spelling variants, paywall gaps, and deleted or corrected items as uncertainty sources.

Human-editor contract:
Outputs are drafts or leads, not publish-ready facts. A named human editor, [EDITOR_NAME], must review the search plan, approve the archive systems to be queried, verify sources and rights, check bias and privacy risks, and approve any external publication or consequential action.

Task:
Create an archive query plan for:
Topic: [TOPIC]
Geography or community: [PLACE]
Date range: [DATE_RANGE]
Archive system or source list: [ARCHIVE_SYSTEMS]
Known people, agencies, organizations, projects, addresses, spellings, or aliases: [KNOWN_NAMES]
Terms to avoid or exclude: [EXCLUDED_TERMS]
Editorial angle or unanswered question: [ANGLE_OR_QUESTION]
Known sensitivities: [SENSITIVITIES]
Named human editor: [EDITOR_NAME]

Return:
1. A short restatement of the authorized task.
2. Primary search strings.
3. Alternative spellings, aliases, acronyms, and historical names.
4. Date-range strategy and why each window matters.
5. Exclusion terms to reduce false positives.
6. Source systems to search first, second, and last.
7. Rights, privacy, and consent cautions.
8. Bias and coverage-gap cautions.
9. Verification checklist for each archive hit.
10. Questions for [EDITOR_NAME] before any findings are used.

Required inputs

• [TOPIC]: the issue, beat, project, person, institution, phrase, or event being researched.
• [PLACE]: the local geography, neighborhood, school district, county, agency jurisdiction, or service area.
• [DATE_RANGE]: the known window and any suspected earlier or later periods worth checking.
• [ARCHIVE_SYSTEMS]: authorized archives such as the newsroom CMS, print archive, clipping library, audio archive, photo archive, newsletter archive, or licensed database.
• [KNOWN_NAMES]: entity variants supplied by the reporter, not generated as assumed facts.
• [EDITOR_NAME]: the accountable human editor who will decide whether to run the plan.

Expected output

The model should produce a practical search worksheet that a reporter can test in authorized tools. Good output separates broad discovery queries from precision queries, includes false-positive controls, and warns that OCR, headline changes, archival gaps, and correction histories can distort results. The most useful plans include local-government acronyms, neighborhood names, former agency names, common misspellings, and date windows tied to elections, budget cycles, development approvals, disasters, lawsuits, school-board votes, or public-health events when those have been provided by the user.

Verification checkpoint

Before relying on any archive result, open the original item and capture the headline, date, byline, publication, URL or archive identifier, version or correction note, quoted sources, and rights status. Do not treat a search snippet, embedding match, OCR fragment, or model summary as evidence. If the archive contains syndicated, wire, third-party, freelance, audio, photo, or user-generated material, require a licensing review before reuse.

Prompt 2: Archive result verification

Purpose

Use this prompt after a reporter has gathered candidate archive hits and needs a disciplined verification worksheet. The goal is to prevent a common newsroom failure mode: treating search results as facts. This prompt asks the model to organize source metadata, identify contradictions, flag missing originals, and build a human verification queue without making factual claims beyond the supplied placeholders.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom user with archive result verification.

Authorization contract:
I am only asking you to help verify candidate materials from archives, indexes, files, databases, CMS records, transcripts, notes, or systems that my newsroom owns, licenses, or is explicitly authorized to use. Do not ask for passwords, tokens, private keys, account numbers, privileged legal material, confidential source identities, or unnecessary personal data. If an input appears outside this authorization boundary, stop and ask for a safer, authorized substitute.

Placeholder-only input contract:
Use only the placeholders and metadata I provide. Do not invent names, quotes, articles, documents, sources, dates, archive hits, public-record contents, audience data, donor attributes, advertising claims, or interview material. If a field is missing, label it "missing from supplied input."

Source verification contract:
Treat each candidate result as unverified until checked against the original item. Require verification of original text or recording, publication date, byline, headline, correction note, version history, cited primary sources, and archive identifier. Do not convert snippets, summaries, OCR text, transcript fragments, or embedding matches into confirmed facts.

Rights and copyright contract:
Flag copyright, licensing, syndication, wire-service, freelance, photo, audio, transcript, and database-use issues that may limit reuse, republication, training, extraction, quotation, or display. Do not recommend reproducing restricted material without documented newsroom rights and policy approval.

Privacy and consent contract:
Minimize personal data. Flag records involving minors, victims, private residents, health, finances, immigration, education, crime, employment, or other sensitive contexts for redaction and editor review before any export, summary, publication, or contact.

Uncertainty contract:
Use uncertainty labels: Confirmed from supplied original, Needs original, Metadata conflict, Rights unclear, Privacy-sensitive, Out of scope, or Not enough information. Explain the reason for each label.

Human-editor contract:
Outputs are drafts or leads, not publish-ready facts. A named human editor, [EDITOR_NAME], must review the verification table, approve source use, verify rights, check bias and privacy risks, and approve any external publication or consequential action.

Task:
Review these candidate archive results:
Archive system: [ARCHIVE_SYSTEM]
Search query used: [SEARCH_QUERY]
Candidate results pasted as metadata only: [CANDIDATE_RESULT_METADATA]
Available originals or excerpts: [AVAILABLE_ORIGINAL_EXCERPTS]
Known corrections, takedowns, or version notes: [CORRECTION_VERSION_NOTES]
Editorial question: [EDITORIAL_QUESTION]
Known sensitivities: [SENSITIVITIES]
Named human editor: [EDITOR_NAME]

Return:
1. A verification table with one row per candidate result.
2. The current verification status using the allowed uncertainty labels.
3. Missing metadata or source material needed.
4. Rights and licensing concerns.
5. Privacy and consent concerns.
6. Contradictions among candidates.
7. Recommended next verification action.
8. A short note to [EDITOR_NAME] listing what cannot be claimed yet.

Required inputs

• [CANDIDATE_RESULT_METADATA]: metadata such as headline, date, byline, archive ID, URL, content type, and snippet, excluding unnecessary personal data.
• [AVAILABLE_ORIGINAL_EXCERPTS]: short excerpts only when the newsroom has the right to process them in the chosen tool.
• [CORRECTION_VERSION_NOTES]: known corrections, updates, takedowns, retractions, archive migrations, or duplicate records.
• [EDITORIAL_QUESTION]: the claim or background issue the reporter hopes the archive can help answer.

Expected output

The expected result is a verification matrix, not a story memo. The best output assigns each candidate to a status such as “Needs original” or “Metadata conflict,” then identifies the next action: retrieve the print page, check the correction tag, compare the web and e-edition versions, inspect the audio transcript against the recording, or ask the archive manager for licensing status. This helps small newsrooms preserve evidence discipline even when staffing is thin.

Verification checkpoint

A human must confirm every item against the original record before it informs a published claim. If the only evidence is a snippet, OCR line, model-generated summary, or database match score, the item remains a lead. If the result mentions a private individual in a sensitive context, the editor should decide whether the public-interest value outweighs privacy risk before further processing.

Prompt 3: Public-meeting agenda monitoring

Purpose

Use this prompt to turn authorized public-meeting agendas into a story-lead digest. OpenAI’s Lenfest-related source notes that newsroom projects have included public-interest monitoring and public-meeting monitoring; the safe implementation is narrow, auditable, and human-reviewed. The model should identify agenda items that may merit coverage, but it must not assert what an agency will decide or imply wrongdoing from agenda language alone.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom user with public-meeting agenda monitoring.

Authorization contract:
I am only asking you to review public agendas, notices, packets, calendars, minutes, or documents that my newsroom owns, lawfully obtained, or is explicitly authorized to use. Do not access restricted systems, bypass paywalls or access controls, scrape in violation of terms or law, or ask for passwords, tokens, private keys, account numbers, privileged legal material, confidential source identities, or unnecessary personal data. If an input appears outside this authorization boundary, stop and ask for a safer, authorized substitute.

Placeholder-only input contract:
Use only the placeholders and documents I provide. Do not invent agencies, meetings, agenda items, votes, quotes, sources, documents, public-record contents, public comments, conflicts, audience data, donor attributes, or advertising claims. If a detail is absent, say it is absent from supplied input.

Source verification contract:
Treat agenda items as leads, not facts about outcomes. Require citations to the supplied agenda, packet page, notice date, meeting date, agency name, item number, and source URL or file identifier. Require later verification against the recording, minutes, staff report, adopted resolution, contract, budget document, or direct agency confirmation.

Rights and copyright contract:
Flag copyright, licensing, terms-of-use, packet reuse, map, photo, consultant report, and attachment restrictions before quoting, republishing, uploading, or extracting materials. Do not suggest bulk reuse unless the newsroom has documented rights and policy approval.

Privacy and consent contract:
Minimize personal data. Flag items involving minors, victims, home addresses, health, disability, immigration, education records, personnel matters, discipline, public benefits, crime, or other sensitive contexts. Recommend redaction and editor review before any summary, alert, contact, or publication.

Uncertainty contract:
Use uncertainty labels: Clear agenda lead, Needs packet review, Needs meeting attendance, Needs records request, Privacy-sensitive, Rights unclear, or Not enough information. Do not infer motives, legality, corruption, intent, or impact beyond the supplied documents.

Human-editor contract:
Outputs are drafts or leads, not publish-ready facts. A named human editor, [EDITOR_NAME], must review the digest, approve source use, verify citations, check rights, privacy, fairness, and bias, and approve any external publication, alert, assignment, or consequential action.

Task:
Review these public-meeting materials:
Agency or body: [AGENCY_NAME]
Meeting date and time: [MEETING_DATE_TIME]
Agenda or packet source: [AGENDA_SOURCE]
Agenda text or item list: [AGENDA_TEXT]
Community or beat priorities: [BEAT_PRIORITIES]
Known watch terms: [WATCH_TERMS]
Known sensitivities: [SENSITIVITIES]
Named human editor: [EDITOR_NAME]

Return:
1. A ranked list of potential story leads.
2. For each lead, cite the agenda item number, page, date, and supplied source identifier.
3. Explain why it may matter to the community.
4. Label uncertainty using the allowed labels.
5. Identify missing documents or confirmations.
6. Suggest non-accusatory reporter questions.
7. Flag privacy, rights, consent, and fairness risks.
8. List what [EDITOR_NAME] must approve before assignment or publication.

Required inputs

• [AGENCY_NAME]: the city council, school board, zoning board, county commission, transit authority, water district, police oversight board, or other body.
• [AGENDA_TEXT]: agenda items, staff report summaries, packet excerpts, or a structured list prepared from authorized public materials.
• [BEAT_PRIORITIES]: local issues the newsroom has decided to monitor, such as housing, taxes, school closures, procurement, land use, infrastructure, policing, environmental permits, or public health.
• [WATCH_TERMS]: supplied terms such as “emergency contract,” “executive session,” “rate increase,” “variance,” “bond,” “settlement,” or “memorandum of understanding.”

Expected output

The model should produce a ranked agenda digest that helps an editor decide whether to assign a reporter. Each item should include a citation back to the supplied agenda or packet, a community-relevance explanation, and a missing-evidence list. Useful results avoid loaded language; for example, “Needs packet review: proposed sole-source contract listed on page [X]” is safer than “possible corruption,” unless verified evidence supports that framing.

Verification checkpoint

Before publication or an external alert, confirm the agenda source, meeting date, item number, packet attachment, and any later changes. A reporter should attend or review the meeting, check minutes and recordings when available, and contact the agency for clarification when the stakes are high. If the item involves identifiable private residents or students, escalate privacy review before naming anyone.

Prompt 4: Meeting transcript triage

Purpose

Use this prompt after a newsroom has an authorized meeting transcript, caption file, or reporter-prepared notes and needs to identify segments worth human review. The output should help find timestamps, motions, votes, public comments, commitments, and unanswered questions, but it must treat automated transcripts as error-prone. This is especially important for names, dollar figures, addresses, legal terms, and public-comment statements.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom user with meeting transcript triage.

Authorization contract:
I am only asking you to review transcripts, captions, notes, recordings, minutes, or public-meeting materials that my newsroom owns, lawfully obtained, or is explicitly authorized to use. Do not access restricted systems, bypass controls, or ask for passwords, tokens, private keys, account numbers, privileged legal material, confidential source identities, or unnecessary personal data. If an input appears outside this authorization boundary, stop and ask for a safer, authorized substitute.

Placeholder-only input contract:
Use only the placeholders and transcript excerpts I provide. Do not invent speakers, quotes, votes, motions, agencies, documents, public-record contents, conflicts, audience data, donor attributes, or advertising claims. If a detail is unclear, mark it as unclear rather than guessing.

Source verification contract:
Treat the transcript as a navigation aid, not a definitive record. Require citation to timestamp, speaker label if supplied, transcript source, meeting date, and agenda item. Require verification against the original audio or video, official minutes, adopted motions, resolutions, contracts, staff reports, or agency confirmation before any factual claim is published.

Rights and copyright contract:
Flag copyright, licensing, caption-source, recording-use, packet-use, and platform terms issues before quoting, republishing, uploading, or extracting substantial material. Do not recommend bulk reuse unless the newsroom has documented rights and policy approval.

Privacy and consent contract:
Minimize personal data. Flag public comments or segments involving minors, victims, home addresses, health, disability, immigration, education records, personnel matters, discipline, benefits, crime, or other sensitive contexts. Recommend redaction and editor review before any summary, alert, contact, or publication.

Uncertainty contract:
Use uncertainty labels: Needs audio check, Needs official minutes, Needs document confirmation, Speaker unclear, Transcript likely error, Privacy-sensitive, Rights unclear, or Not enough information. Do not infer motives, legality, intent, or impact beyond the supplied transcript.

Human-editor contract:
Outputs are drafts or leads, not publish-ready facts. A named human editor, [EDITOR_NAME], must review the triage notes, approve source use, verify timestamps and quotes, check rights, privacy, fairness, and bias, and approve any external publication, assignment, or consequential action.

Task:
Triage this meeting material:
Agency or body: [AGENCY_NAME]
Meeting date: [MEETING_DATE]
Agenda source or item list: [AGENDA_SOURCE_OR_ITEMS]
Transcript, captions, or reporter notes: [TRANSCRIPT_OR_NOTES]
Known watch topics: [WATCH_TOPICS]
Known sensitivities: [SENSITIVITIES]
Named human editor: [EDITOR_NAME]

Return:
1. A timestamped triage table.
2. Potentially newsworthy segments and why they may matter.
3. Motions, votes, commitments, dollar figures, deadlines, and named documents mentioned, each labeled as unverified unless confirmed by supplied official material.
4. Segments that need audio/video review.
5. Non-accusatory follow-up questions for officials, speakers, or affected people.
6. Privacy, consent, rights, fairness, and bias risks.
7. A list of statements that must not be quoted until checked against the recording.
8. What [EDITOR_NAME] must approve before publication or outreach.

Required inputs

• [TRANSCRIPT_OR_NOTES]: meeting transcript, captions, or reporter notes that the newsroom is authorized to process.
• [AGENDA_SOURCE_OR_ITEMS]: agenda item list or source identifiers to connect transcript segments to official agenda structure.
• [WATCH_TOPICS]: newsroom priorities such as budget amendments, land-use decisions, school policies, contracts, public safety, utilities, or litigation updates.

Expected output

The expected output is a timestamped review queue. A strong result flags uncertain speaker labels, transcript artifacts, ambiguous votes, and moments where an official refers to a document not included in the packet. It should also produce neutral follow-up questions, such as “What document was referenced at [timestamp]?” or “Was the motion amended before the vote?” rather than accusatory claims.

Verification checkpoint

Before quoting or summarizing a speaker, check the original audio or video at the timestamp and compare it with official minutes or the adopted motion. If a transcript contains a potentially damaging allegation, a private resident’s address, a student’s name, or a medical or personnel detail, do not publish from the transcript alone. Require editor review and, when appropriate, additional reporting and right-of-reply procedures.

Prompt 5: Public-record source inventory

Purpose

Use this prompt at the beginning of a public-records project to map what sources might exist, who likely holds them, what authorization or request path may apply, and what privacy or legal limits may affect use. The output is not legal advice and should not draft an aggressive request designed to obtain protected information. It is a planning inventory for a reporter and editor to discuss with counsel or a records specialist when needed.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom user with a public-record source inventory.

Authorization contract:
I am only asking you to help inventory possible public, newsroom-owned, licensed, lawfully obtained, or explicitly authorized records and source paths. Do not access restricted systems, bypass controls, scrape in violation of law or terms, impersonate anyone, or ask for passwords, tokens, private keys, account numbers, privileged legal material, confidential source identities, or unnecessary personal data. If an input appears outside this authorization boundary, stop and ask for a safer, authorized substitute.

Placeholder-only input contract:
Use only the placeholders and descriptions I provide. Do not invent records, agencies, statutes, exemptions, documents, quotes, sources, public-record contents, audience data, donor attributes, advertising claims, or interview material. If jurisdiction-specific law is needed, say that a qualified human must verify it.

Source verification contract:
Treat all possible records as leads. Require verification of the record’s existence, custodian, date range, retention schedule if known, request pathway, fee rules if applicable, exemptions or confidentiality limits, and authenticity after receipt. Require citation to supplied sources only; do not fabricate legal citations.

Rights and copyright contract:
Flag copyright, database licensing, vendor terms, GIS/map restrictions, court-record access rules, transcript restrictions, photo/video rights, and reuse limits before publication, republication, upload, or extraction.

Privacy and consent contract:
Minimize personal data. Flag records involving minors, victims, home addresses, health, disability, immigration, education, personnel files, discipline, benefits, crime, protected traits, or other sensitive contexts. Recommend redaction, consent review where relevant, and editor/legal review before requesting, processing, contacting, or publishing.

Uncertainty contract:
Use uncertainty labels: Likely public-record lead, Custodian uncertain, Legal review needed, Privacy-sensitive, Rights unclear, Authentication needed, Not enough information, or Out of scope. Do not state that a record is public, disclosable, exempt, complete, or accurate unless the supplied input establishes that.

Human-editor contract:
Outputs are drafts or leads, not publish-ready facts or legal advice. A named human editor, [EDITOR_NAME], must review the inventory, approve request strategy, verify law and policy with qualified support when needed, check rights, privacy, fairness, and bias, and approve any external request, publication, or consequential action.

Task:
Create a public-record source inventory for:
Reporting question: [REPORTING_QUESTION]
Jurisdiction or agencies involved: [JURISDICTION_OR_AGENCIES]
Known time period: [DATE_RANGE]
Known documents or data fields: [KNOWN_DOCUMENTS_OR_FIELDS]
Records already obtained: [OBTAINED_RECORDS]
Sensitive populations or privacy risks: [SENSITIVITIES]
Newsroom policy constraints: [NEWSROOM_POLICY]
Named human editor: [EDITOR_NAME]

Return:
1. A table of possible record categories.
2. Possible custodians or source paths, labeled as unverified leads.
3. What each record could help establish if obtained and authenticated.
4. Privacy, consent, rights, and fairness risks.
5. Authentication steps after receipt.
6. Fields or details to avoid requesting unless clearly necessary and lawful.
7. Questions for [EDITOR_NAME] and legal/records review.
8. What cannot be concluded from the inventory alone.

Required inputs

• [REPORTING_QUESTION]: the accountability question, such as spending, inspections, discipline, procurement, environmental enforcement, school policy, zoning, or public safety.
• [JURISDICTION_OR_AGENCIES]: agencies, courts, boards, districts, or offices that may hold records.
• [KNOWN_DOCUMENTS_OR_FIELDS]: supplied document names or data fields already known to the reporter.
• [NEWSROOM_POLICY]: any internal rule about minors, crime victims, addresses, data retention, anonymous sources, or legal review.

Expected output

The model should return an inventory of record categories and likely custodians, clearly labeled as unverified leads. A useful output also identifies fields that may create privacy or fairness risks, such as home addresses, dates of birth, student identifiers, medical information, immigration details, and personnel records. The prompt intentionally asks what not to request because overcollection can create legal, ethical, security, and trust problems.

Verification checkpoint

A qualified human must confirm the applicable records law, request procedure, exemptions, fees, deadlines, and appeal path for the relevant jurisdiction. After records arrive, authenticate them by checking source, date, completeness, metadata where appropriate, chain of custody, and consistency with other evidence. Do not publish raw datasets containing unnecessary personal information.

This article provides ChatGPT prompts for AI-powered meeting notes, including summaries, action items, follow-ups, and decision tracking. The 20 ChatGPT Prompts for AI-Powered Meeting Notes: Summaries, Action Items, Follow-Ups, and Decision Tracking article is a focused companion for Public Meeting Monitoring because it is semantically appropriate because public-meeting monitoring in a newsroom relies on extracting summaries, decisions, and follow-up items from meeting records.

Prompt 6: Interview preparation

Purpose

Prepare evidence-led questions without inventing claims about a source.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Use only interview materials, documents, and systems the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, or facts. Cite each source identifier and date, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. A named human editor must review and approve the questions before outreach. Using [SOURCE_PACKET], draft questions grouped by confirmed facts, gaps, contradictions, and follow-ups. Do not contact anyone or publish anything.

Required inputs

A redacted source packet, interview purpose, known facts, disputed claims, deadline, and named editor.

Expected output

A question plan with source references, dates, uncertainty notes, and no assumed answers.

Verification checkpoint

The named human editor verifies every premise against the source packet, removes unsafe or unsupported questions, confirms consent and privacy boundaries, and approves outreach.

Prompt 7: Claim-evidence mapping

Purpose

Map each draft claim to inspectable evidence before editing or publication.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Use only drafts, records, and systems the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, dates, or facts. Cite the source identifier and date for every claim, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. A named human editor must review and approve the result. From [DRAFT] and [SOURCE_PACKET], create a table with claim, evidence, source date, strength, contradiction, missing proof, and required editor action. Do not publish or contact sources.

Required inputs

The draft, cleared source packet, correction history, publication standard, and named editor.

Expected output

A claim ledger that separates verified, disputed, unsupported, and context-dependent statements.

Verification checkpoint

The named human editor opens every cited source, checks the date and context, confirms rights and privacy, and removes or qualifies unsupported claims.

Prompt 8: Bilingual translation draft

Purpose

Create a reviewable translation draft while preserving names, figures, quotations, and uncertainty.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Translate only material the newsroom owns or is explicitly authorized to use. Do not fabricate quotations, sources, records, documents, cultural context, or official instructions. Preserve source identifiers and dates, label uncertain terms, check copyright and licensing, privacy and consent, and redact unnecessary personal data. A named human editor and qualified language reviewer must approve the translation before publication. Translate [SOURCE_TEXT] from [SOURCE_LANGUAGE] to [TARGET_LANGUAGE], keeping names, dates, amounts, quotations, caveats, and links unchanged unless a documented house rule applies.

Required inputs

Cleared source text, language pair, audience, glossary, style guide, rights status, and named reviewers.

Expected output

A parallel translation draft, uncertainty log, glossary conflicts, and review questions.

Verification checkpoint

A qualified human language reviewer compares the draft line by line with the dated source, checks rights, privacy, consent, quotations, and cultural context, then the named editor approves publication.

Prompt 9: Translation verification

Purpose

Audit a translation for factual, linguistic, legal, and accessibility drift.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Review only source and translated material the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, corrections, or certainty. Cite source identifiers and dates, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. A named human editor and qualified language reviewer must approve the findings. Compare [SOURCE_TEXT] with [TRANSLATION] and flag changed facts, missing caveats, altered quotations, ambiguous terms, accessibility issues, and items requiring specialist review.

Required inputs

The dated source, translation, glossary, house style, legal or safety terminology, and reviewer names.

Expected output

A severity-ranked discrepancy table and corrected draft segments, each tied to the original source.

Verification checkpoint

The language reviewer verifies every correction against the original, and the named human editor confirms rights, privacy, consent, uncertainty labels, and publication readiness.

Prompts 10–18: audio archives, story selection, investigative planning, audience learning, accessibility, newsletters, and subscription tests

OpenAI’s Lenfest AI Collaborative materials describe newsroom uses that include archive search, public-interest monitoring, translation, audio transcription, audience engagement, investigative research, audience personalization, subscription growth, and operational efficiency. The prompts in this section convert those broad use cases into bounded newsroom contracts: the model may help organize evidence, identify gaps, draft alternatives, and propose experiments, but it must not create facts, infer sensitive traits, publish material, contact people, change offers, or launch campaigns without named human approval.

For teams using these prompts with ChatGPT, Codex, or an internal tool built on the API, follow OpenAI’s prompt-engineering guidance in practical terms: give the model a clear task, provide the relevant context, specify the output format, and define what evidence it must use. In newsroom work, the most important “context” is not just the text of an article or transcript; it is the authorization boundary, source provenance, publication status, consent status, copyright status, audience risk, and the identity of the editor who owns the next decision.

Prompt 10: Audio archive transcription review

Purpose

Use this prompt to review a machine transcript of archival audio that your newsroom owns, has licensed, or is explicitly authorized to use. The goal is to flag likely transcription errors, uncertain names, inaudible segments, date references, quote-risk areas, and follow-up checks before a reporter relies on the transcript for search, research, captions, excerpts, or story development. This prompt is especially useful for local stations and community newsrooms digitizing older audio, but it must be treated as a review aid rather than a replacement for listening to the original recording.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom with audio archive transcription review.

Authorization and scope:
- I am using only audio, transcripts, metadata, archives, systems, accounts, and documents that my newsroom owns, licenses, or is explicitly authorized to use.
- Do not request credentials, private keys, account numbers, confidential source identities, personal identifiers that are not necessary for this review, or privileged material.
- Do not publish, contact anyone, upload to outside services, change metadata, or make rights decisions.

Evidence and nonfabrication rules:
- Use only the transcript text, timecodes, notes, and metadata I provide.
- Do not invent speakers, interviews, quotes, documents, dates, locations, or missing audio.
- If a word, name, title, quotation, date, or claim is uncertain, label it as “uncertain” and explain what evidence is missing.
- Keep every suggested correction tied to a timecode, line number, or provided excerpt.

Privacy, fairness, rights, and human-editor contract:
- Flag personal information, minors, health information, immigration status, allegations of wrongdoing, trauma, private grief, confidential-source risk, or other sensitive material for editor review.
- Do not infer sensitive traits or protected characteristics unless they are explicitly relevant, sourced, and necessary.
- Identify copyright, licensing, consent, reuse, and archival-rights questions before excerpting or republication.
- Check for bias in speaker labeling, dialect handling, translation assumptions, accents, and automated confidence scores.
- The output is a draft review memo, not a publish-ready transcript.
- A named human editor, [EDITOR_NAME], must verify the audio against the original recording before publication, external sharing, captioning, legal reliance, or consequential action.

Task:
Review the transcript and metadata below. Produce:
1. A concise archive item summary.
2. A table of likely transcription issues with timecode, transcript text, suspected issue, confidence level, and verification action.
3. A table of quote-risk passages that require human listening before use.
4. A list of names, places, organizations, dates, and titles that need independent verification.
5. Privacy, consent, rights, and sensitivity flags.
6. Search keywords and tags, clearly labeled as draft metadata suggestions.
7. Questions for the reporter or archivist.

Transcript and metadata:
[PASTE_TRANSCRIPT_WITH_TIMECODES]
[PASTE_AUDIO_METADATA]
[PASTE_RIGHTS_OR_CONSENT_NOTES]
[PASTE_STORY_CONTEXT_IF_ANY]

Required inputs

• Machine transcript with timecodes, speaker labels if available, and any confidence indicators generated by your authorized transcription workflow.
• Audio metadata such as approximate recording date, program name, collection name, file identifier, rights status, and known participants.
• Any newsroom notes about consent, licensing, embargoes, donor restrictions, archival restrictions, or whether the audio involves minors, private grief, allegations, or vulnerable communities.
• The intended use: internal search, reporter research, caption preparation, excerpt selection, social clip review, or story publication.

Expected output

The expected result is a structured review memo, not a corrected transcript. It should separate high-confidence transcription cleanup from uncertainty, list the timecodes where human listening is mandatory, and identify rights or privacy issues before anyone republishes archival material. A useful answer will say, for example, that a mayor’s name at 00:13:42 appears uncertain because the transcript conflicts with the archive metadata, rather than silently “fixing” the name.

Verification checkpoint

Before using the output externally, the named editor or archivist must listen to the original audio at every flagged timecode, confirm the exact wording of any quotation, verify names and dates against independent records, and review reuse rights. If the transcript contains allegations, private personal details, children’s voices, medical information, or confidential-source risk, require a higher editorial review before excerpting, indexing, or republishing.

Prompt 11: Story-lead scoring from monitored materials

Purpose

Use this prompt to rank possible story leads from monitored public documents, meeting notes, newsroom tips, authorized scraping outputs, or internal digests. OpenAI’s Lenfest/OpenAI reporting describes public-interest monitoring and daily lead digests as a newsroom use case, but the safe editorial posture is to treat AI-ranked leads as triage signals, not as fact determinations or assignment orders. The model can help sort urgency, public impact, novelty, and evidence gaps, while editors retain news judgment.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local newsroom with story-lead scoring.

Authorization and scope:
- I am using only monitored materials, public documents, tips, digests, archives, systems, accounts, and workflows that my newsroom owns or is explicitly authorized to use.
- Do not scrape, access, bypass, or infer from restricted systems.
- Do not contact sources, publish, assign reporters, send alerts, or take consequential action.

Evidence and nonfabrication rules:
- Use only the lead items, citations, dates, URLs or document identifiers, meeting names, and notes I provide.
- Do not invent facts, public-record contents, quotes, interviews, named sources, audience data, legal conclusions, or wrongdoing.
- If evidence is missing, label the lead as unverified and list the missing documents or people to contact through normal reporting channels.
- Preserve uncertainty labels and distinguish allegation, proposal, vote, staff recommendation, public comment, and confirmed action.

Privacy, fairness, rights, and human-editor contract:
- Flag personal data, minors, crime victims, health information, immigration status, housing insecurity, allegations, employment discipline, and other sensitive material.
- Do not infer protected traits or target communities unfairly.
- Include privacy, consent, copyright, licensing, and public-record-use checks where relevant.
- Evaluate whether coverage could amplify unverified claims or disproportionately burden vulnerable people.
- The output is a draft lead-ranking memo, not a publish-ready story budget.
- A named human editor, [EDITOR_NAME], must approve any assignment, source contact, publication, alert, or external message.

Task:
Score the following possible leads using the rubric below. Return a ranked table and a short editor memo.

Rubric:
- Public impact: 1–5
- Timeliness: 1–5
- Evidence strength: 1–5
- Novelty compared with prior coverage: 1–5
- Accountability value: 1–5
- Harm and fairness risk: low/medium/high
- Verification difficulty: low/medium/high
- Rights or privacy concern: low/medium/high

For each lead, include:
1. Rank.
2. One-sentence lead description.
3. Evidence cited from provided material.
4. Scores and short rationale.
5. Missing verification.
6. Fairness, privacy, rights, and safety flags.
7. Recommended next reporting step for editor approval.

Lead items:
[PASTE_AUTHORIZED_LEAD_DIGEST_OR_NOTES]
Prior coverage summary:
[PASTE_ARCHIVE_OR_PRIOR_COVERAGE_NOTES]
Editor priorities:
[PASTE_NEWSROOM_PRIORITIES]

Required inputs

• A lead digest, meeting-monitoring output, public-record summary, tip log excerpt, or reporter notes that your newsroom is authorized to review.
• Dates, source identifiers, meeting names, agenda items, document links or internal IDs, and any known status such as proposed, pending, approved, denied, withdrawn, or disputed.
• A short description of current editorial priorities, such as schools, housing, courts, environment, public spending, public health, local elections, or service journalism.
• Prior coverage notes so the model can avoid treating an already-covered item as new.

Expected output

The prompt should produce a ranked story-lead table with evidence citations and missing-verification notes. The best outputs will downgrade leads that sound dramatic but lack documents, distinguish official action from public comment, and call out leads that may require trauma-informed reporting or legal review. The ranking is useful only if the newsroom can trace each score to provided evidence.

Verification checkpoint

An editor must review the source documents and score rationales before assigning work. Any lead involving allegations, private individuals, minors, legal exposure, or sensitive personal information should be escalated to a senior editor or counsel according to newsroom policy. Do not use the output to contact sources, publish push alerts, or imply wrongdoing until a human has verified the record.

This article explains Chain-of-Verification prompting as a method for reducing AI hallucinations through structured checking templates and examples. The Chain-of-Verification Prompting: The Advanced Technique That Eliminates AI Hallucinations in 2026 article is a focused companion for Translation Verification because it fits translation verification because the marker requires a verification workflow, and this post specifically addresses structured checking of AI outputs.

Prompt 12: Investigative hypothesis planning

Purpose

Use this prompt to turn a reporting question into a testable investigative plan with hypotheses, evidence needs, public-record requests, interview categories, counterevidence, and ethical risk checks. It is designed for authorized newsroom planning, not for claiming misconduct or building a predetermined narrative. The safest use is to force the investigation to look for disconfirming evidence as actively as confirming evidence.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom with investigative hypothesis planning.

Authorization and scope:
- I am using only information, documents, systems, archives, accounts, tips, and workflows that my newsroom owns, has received lawfully, or is explicitly authorized to use.
- Do not request credentials, private data, hacked material, sealed records, confidential-source identities, or privileged material.
- Do not contact sources, file records requests, accuse anyone, publish, or take consequential action.

Evidence and nonfabrication rules:
- Use only the facts, documents, tips, prior coverage, and constraints I provide.
- Do not invent wrongdoing, quotes, sources, documents, agency responses, public-record contents, data fields, or expert opinions.
- Separate confirmed facts, allegations, hypotheses, unknowns, and possible explanations.
- Include counter-hypotheses and evidence that could disprove the story premise.

Privacy, fairness, rights, and human-editor contract:
- Flag risks involving private individuals, minors, victims, health, immigration, employment, housing, finances, protected traits, or allegations of misconduct.
- Do not infer sensitive traits or suggest targeting people based on protected or highly sensitive personal data.
- Include copyright, licensing, consent, public-record, source-protection, and data-minimization checks.
- Evaluate fairness: right of response, proportionality, context, error risk, community impact, and whether the plan overweights one source type.
- The output is a draft reporting plan, not a conclusion or publish-ready investigation.
- A named human editor, [EDITOR_NAME], must approve the plan before records requests, source outreach, publication, legal review submissions, or consequential action.

Task:
Build an investigative plan from the material below.

Return:
1. A one-paragraph working question.
2. A table separating confirmed facts, allegations, hypotheses, unknowns, and possible benign explanations.
3. Three to five testable hypotheses with what evidence would support or weaken each.
4. A records and data plan using only lawful, authorized channels.
5. Interview categories, not invented names, unless names are provided.
6. A counterevidence plan.
7. Privacy, fairness, safety, copyright, consent, and source-protection risks.
8. A staged reporting timeline with editor approval gates.
9. Questions for legal or standards review.

Material:
[PASTE_CONFIRMED_FACTS]
[PASTE_TIPS_OR_ALLEGATIONS_WITH_SOURCE_STATUS]
[PASTE_PRIOR_COVERAGE]
[PASTE_AVAILABLE_DOCUMENTS]
[PASTE_NEWSROOM_CONSTRAINTS]

Required inputs

• A concise statement of the reporting concern, with each fact labeled as confirmed, alleged, unknown, or based on a tip.
• Prior coverage, public documents, data inventories, and known access limits.
• Newsroom standards, legal review triggers, source-protection requirements, and deadlines.
• Any known conflicts of interest, community harm concerns, or reasons the subject matter requires special care.

Expected output

The model should produce a hypothesis matrix and reporting plan that helps the newsroom avoid confirmation bias. A good plan will include the possibility that the tip is wrong, stale, incomplete, or explained by a lawful administrative process. It should also identify what the newsroom must not claim until documents, interviews, and subject responses are gathered through authorized reporting methods.

Verification checkpoint

The named editor must approve the hypothesis plan before reporters pursue records, interviews, data analysis, or publication. For high-risk investigations, the newsroom should require standards, legal, and security review before handling sensitive data or contacting vulnerable sources. Every factual claim must remain tied to evidence, and any adverse claim must receive appropriate right-of-response review under newsroom policy.

Prompt 13: Source diversity review

Purpose

Use this prompt after drafting or outlining a story to evaluate whether the sourcing pattern is too narrow, overdependent on officials, missing affected people, or lacking relevant expertise. This is not a demographic guessing tool. It should not infer race, ethnicity, religion, disability, sexuality, immigration status, income, health status, or other sensitive traits from names, locations, voices, photos, or writing style. Instead, it reviews source roles, perspectives, documentation types, and gaps explicitly present in the story file.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom with source diversity and fairness review.

Authorization and scope:
- I am using only story drafts, outlines, source lists, interview notes, documents, and workflows that my newsroom owns or is explicitly authorized to use.
- Do not contact sources, add names, invent interviews, or publish.
- Do not infer demographics or sensitive traits from names, addresses, accents, photos, neighborhoods, or other proxies.

Evidence and nonfabrication rules:
- Use only the source information I provide.
- Do not invent quotes, viewpoints, documents, affiliations, expertise, identities, or community reaction.
- If a perspective is missing, describe the category of source to seek, not a fabricated person.
- Distinguish quoted sources, background sources, documents, data, officials, advocates, affected people, independent experts, and opposing views.

Privacy, fairness, rights, and human-editor contract:
- Protect confidential sources, vulnerable people, minors, victims, and people facing retaliation.
- Do not infer protected characteristics or recommend targeting people using protected or highly sensitive personal data.
- Flag consent, attribution, anonymity, copyright, licensing, and document-use concerns.
- Evaluate fairness: overreliance on institutions, missing right of response, lack of community context, tokenism risk, and disproportionate harm.
- The output is a draft sourcing audit, not a publication decision.
- A named human editor, [EDITOR_NAME], must approve any additional outreach, attribution change, anonymity decision, publication, or consequential action.

Task:
Review the story materials below for sourcing breadth and fairness.

Return:
1. Source map by role and evidence type.
2. Missing perspective categories.
3. Overrepresented source categories.
4. Right-of-response and adverse-claim gaps.
5. Confidentiality, consent, privacy, and retaliation risks.
6. Copyright and licensing issues for documents, photos, audio, or user-generated material.
7. Bias and fairness notes.
8. Specific reporting questions to close gaps, phrased as categories rather than invented sources.

Story materials:
[PASTE_DRAFT_OR_OUTLINE]
[PASTE_SOURCE_LIST_WITH_ROLES]
[PASTE_INTERVIEW_STATUS_NOTES]
[PASTE_DOCUMENT_AND_MEDIA_INVENTORY]
[PASTE_ATTRIBUTION_OR_ANONYMITY_NOTES]

Required inputs

• Story draft or outline, source list, source roles, document list, and interview status.
• Attribution notes such as on the record, background, not for attribution, anonymous, embargoed, or off limits.
• Any standards notes about vulnerable sources, minors, retaliation risk, or right of response.
• Known publication constraints, including whether the story is breaking news, enterprise, investigative, explanatory, or service journalism.

Expected output

The output should show whether the story relies mainly on officials, advocates, affected residents, documents, data, or experts. It should recommend categories such as “tenant affected by the policy,” “independent public-finance expert,” or “agency spokesperson with authority to respond,” rather than inventing people. It should also warn if the current draft includes an adverse claim without a documented response opportunity.

Verification checkpoint

The human editor must compare the audit with the story file and assignment notes. Do not add demographic labels unless the source self-identifies, the information is relevant, and newsroom policy supports inclusion. Do not publish anonymous-source material, sensitive details, or adverse claims without editor approval and applicable standards review.

Prompt 14: Correction-risk audit before publication

Purpose

Use this prompt for pre-publication quality control when a draft contains names, titles, numbers, dates, locations, links, legal terminology, translations, quotations, archive references, or sensitive claims. The purpose is to reduce preventable corrections by producing a checkable risk register. It does not decide whether a story is legally safe or ethically ready; it helps editors see what still needs verification.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom with a correction-risk audit.

Authorization and scope:
- I am using only drafts, notes, documents, data, images, audio, transcripts, and editorial workflows that my newsroom owns or is explicitly authorized to use.
- Do not publish, update the CMS, contact sources, change headlines, or make legal decisions.
- Do not request confidential credentials, privileged legal advice, or unnecessary personal data.

Evidence and nonfabrication rules:
- Use only the provided draft and supporting materials.
- Do not invent facts, corrections, source responses, quotes, documents, links, data, or legal conclusions.
- Identify claims that lack support in the supplied evidence.
- Label each risk as high, medium, or low and explain the verification step.

Privacy, fairness, rights, and human-editor contract:
- Flag personal data, minors, private addresses, health information, immigration status, allegations, crime-victim details, confidential sources, and other sensitive content.
- Do not infer sensitive traits or recommend disclosure of unnecessary private details.
- Include copyright, licensing, consent, attribution, quotation, image, audio, translation, and archive-reuse checks.
- Evaluate fairness: right of response, loaded language, unsupported implications, missing context, and disproportionate harm.
- The output is a draft correction-risk memo, not legal advice or final publication approval.
- A named human editor, [EDITOR_NAME], must approve publication, correction language, legal escalation, takedown decisions, or any consequential action.

Task:
Audit the draft below for correction risk.

Return:
1. Top ten correction risks in priority order.
2. A claim-by-claim table with claim, evidence supplied, missing evidence, risk level, and verification action.
3. Names, titles, dates, numbers, locations, links, and quotes requiring confirmation.
4. Sensitive personal information and fairness flags.
5. Copyright, licensing, consent, attribution, image, audio, translation, and archive-reuse issues.
6. Right-of-response gaps and adverse-claim warnings.
7. A final “do not publish until checked” list.

Draft:
[PASTE_DRAFT]
Supporting evidence:
[PASTE_NOTES_DOCUMENTS_DATA_TRANSCRIPTS_LINKS]
Publication context:
[PASTE_DEADLINE_SECTION_STORY_TYPE]
Known standards or legal triggers:
[PASTE_NEWSROOM_POLICY_NOTES]

Required inputs

• The current draft, including headline, deck, captions, graphics text, newsletter blurb, social copy, and push language if available.
• Supporting notes, documents, datasets, transcripts, archive references, translation notes, and prior versions that reporters used to support the story.
• Any known standards triggers, such as allegations, minors, suicide, sexual assault, medical claims, confidential sources, court records, or election information.
• The name of the editor responsible for final publication approval.

Expected output

The output should be a practical pre-publication risk memo. It should not merely say “verify facts”; it should list concrete items such as “confirm the council vote date,” “verify spelling and current title,” “match the quoted sentence to the transcript timecode,” or “confirm the chart denominator.” High-risk issues should be those that could materially mislead readers, harm a person, trigger a correction, or require legal review.

Verification checkpoint

Before publication, the editor must confirm that every item on the “do not publish until checked” list has an owner and a resolution. If the story includes adverse claims, private information, translated quotations, archival audio, or complex data, require standards review before publication. This prompt must never be used to avoid a correction; if a published error is found, follow the newsroom’s correction policy.

Prompt 15: Audience-needs research synthesis

Purpose

Use this prompt to synthesize authorized audience research such as surveys, reader interviews, callouts, listening sessions, analytics summaries, customer-support themes, comment moderation summaries, or membership feedback. The model can help cluster needs and identify editorial opportunities, but it must not invent audience data, infer protected traits, or convert sensitive personal information into targeting rules. Treat the result as qualitative research support, not a definitive market model.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom with audience-needs research synthesis.

Authorization and scope:
- I am using only audience research, analytics summaries, survey responses, interview notes, callouts, support themes, comments, and workflows that my organization owns, collected lawfully, or is explicitly authorized to use.
- Do not access outside accounts, identify anonymous users, deanonymize people, contact readers, publish, or launch campaigns.
- Do not use sensitive personal data for targeting, donor scoring, subscription decisions, or personalization without lawful authority, documented consent, and organization policy.

Evidence and nonfabrication rules:
- Use only the audience inputs I provide.
- Do not invent audience segments, survey results, quotes, demographics, behavioral data, donor attributes, subscription likelihood, or reader needs.
- Label small-sample, anecdotal, stale, biased, or unrepresentative evidence.
- Separate observed reader statements from interpretation and recommendations.

Privacy, fairness, rights, and human-editor contract:
- Redact or minimize personal identifiers and sensitive details.
- Do not infer protected traits, health status, financial status, political persuasion, religion, immigration status, or other sensitive attributes.
- Include privacy, consent, retention, copyright, licensing, and research-ethics checks.
- Evaluate fairness: who is missing, who may be overrepresented, language access, disability access, digital divide, and risk of reinforcing stereotypes.
- The output is a draft research synthesis, not a final audience strategy or targeting plan.
- A named human editor or audience lead, [EDITOR_NAME], must approve any publication, product change, personalization rule, subscription campaign, or consequential action.

Task:
Synthesize the audience research below.

Return:
1. Key audience needs, each tied to evidence excerpts or aggregate notes.
2. Evidence limitations and sampling caveats.
3. Missing audience groups or perspectives, without inferring protected traits.
4. Editorial opportunities framed as service, accountability, explanatory, community, or utility journalism.
5. Privacy, consent, fairness, accessibility, copyright, and retention concerns.
6. Questions for follow-up research.
7. A recommended next small-scale experiment requiring human approval.

Audience research:
[PASTE_REDACTED_SURVEY_OR_INTERVIEW_NOTES]
[PASTE_ANALYTICS_SUMMARY_IF_AUTHORIZED]
[PASTE_CALLOUT_OR_SUPPORT_THEMES]
[PASTE_RESEARCH_METHOD_AND_SAMPLE_NOTES]
[PASTE_ORGANIZATION_POLICY_CONSTRAINTS]

Required inputs

• Redacted audience research materials and the method used to collect them, including survey wording, interview protocol, callout prompt, or analytics summary source.
• Sampling notes such as number of responses, time period, channel, geography, language, and known limitations.
• Consent and privacy notes explaining whether responses can be quoted internally, externally, anonymously, or only in aggregate.
• Editorial or product goals, such as improving election guides, schools coverage, weather alerts, civic explainers, newsletters, or membership onboarding.

Expected output

The output should cluster audience needs into evidence-backed themes and plainly state limits. A useful synthesis may say that a survey overrepresents newsletter subscribers and therefore cannot describe the entire county. It should recommend follow-up research for missing communities rather than pretending the provided dataset is complete.

Verification checkpoint

A human audience lead or editor must review the synthesis against the raw research and collection consent. Do not use the output to create targeted advertising, donor scoring, subscription suppression, or personalization rules based on sensitive traits. Any external quotation from a reader requires consent and normal editorial verification.

This sponsored-agent and ChatGPT Ads governance playbook covers disclosure, claim evidence, human creative review, CRM and ecommerce data boundaries, and escalation, offering concrete safeguards for audience research that could otherwise drift into sensitive targeting or unsupported assumptions. The Sponsored Agent and ChatGPT Ads Governance Playbook: Disclosure, Claim Evidence, Human Creative Review, CRM and Ecommerce Data Boundaries, and Escalation article is a focused companion for Audience Research Ethics because the target provides explicit data-boundary, disclosure, and human-review controls, making it more ethically relevant than a general content-strategy prompt collection.

Prompt 16: Accessibility rewrite for public-service journalism

Purpose

Create a plain-language accessibility draft without changing verified facts or obligations.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Use only materials the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, eligibility rules, deadlines, or official instructions. Cite each source identifier and date, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. A named human editor must review and approve the draft. Rewrite [SOURCE_TEXT] for [AUDIENCE] in plain language while preserving names, dates, amounts, caveats, quotations, and legal or safety meaning. Flag anything that requires specialist review.

Required inputs

Source text, audience needs, glossary, accessibility standard, cleared visuals, rights status, and named editor.

Expected output

A plain-language draft, glossary, alt-text suggestions, and a list of meaning-sensitive passages.

Verification checkpoint

The named human editor checks every fact and date, and appropriate accessibility, language, legal, or subject-matter reviewers approve the final wording.

Prompt 17: Newsletter experiment design

Purpose

Design a bounded newsletter test without inventing audience intent or using sensitive targeting.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Use only aggregate audience data, content, and systems the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, audience motives, performance claims, or results. Cite data sources and dates, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. Do not infer protected or sensitive traits. A named human editor and product owner must approve the experiment. Design an A/B test for [NEWSLETTER_GOAL] with one variable, an ethical audience definition, success and harm metrics, a stop rule, and rollback.

Required inputs

Goal, approved aggregate baseline, audience policy, editorial standard, test duration, delivery constraints, and named owners.

Expected output

A preregistered experiment plan with hypotheses, metrics, sample limitations, fairness checks, and decision rules.

Verification checkpoint

The named human editor and product owner verify source dates, consent, privacy, rights, fairness, statistical limits, stop criteria, and that no result is treated as causal without adequate evidence.

Prompt 18: Subscription-offer testing

Purpose

Plan a reversible subscription-message test with claim and consumer-protection review.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom. Use only approved offer terms, aggregate metrics, and systems the newsroom owns or is explicitly authorized to use. Do not fabricate sources, quotations, records, documents, discounts, scarcity, benefits, deadlines, audience intent, or projected results. Cite each offer source and date, label uncertainty, check copyright and licensing, privacy and consent, and redact unnecessary personal data. Do not infer sensitive traits or change prices, accounts, or payment settings. A named human editor, subscription owner, and legal reviewer must approve the test. Draft variants for [APPROVED_OFFER] and a measurement plan with fairness, accessibility, stop, and rollback rules.

Required inputs

Approved offer terms, eligible audience rules, baseline metrics, legal requirements, accessibility standard, and named approvers.

Expected output

Claim-safe variants, a source-backed offer table, risk checklist, test design, and rollback plan.

Verification checkpoint

The named human editor and authorized owners verify every price, term, date, claim, consent rule, privacy condition, accessibility requirement, and approval before launch.

Prompt 19: Advertising prospect research from authorized public and first-party materials

Purpose

This prompt helps a local newsroom’s advertising or revenue team prepare a prospect-research brief using only authorized materials: the newsroom’s own CRM exports, media kit, published rate-card language, public business websites, public filings, publicly available sponsorship history, and first-party performance summaries that the organization is permitted to use. Lenfest/OpenAI’s reporting on newsroom AI projects includes advertising prospecting as one of the operational areas explored by participating organizations, but any output from this prompt is a draft lead file, not a sales claim or a targeting decision.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized local-newsroom revenue team. Work only with the materials I provide or explicitly describe as public and authorized. Do not invent business facts, decision-makers, campaign history, budget, audience fit, or advertising claims.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Task: Create an advertising prospect research brief for [PROSPECT_OR_CATEGORY] using the following authorized inputs:
- Public business materials: [PASTE LINKS/TITLES/SUMMARIES OR ATTACH FILE NAMES]
- First-party newsroom materials: [MEDIA_KIT_FILE], [AUDIENCE_SUMMARY_FILE], [AD_PRODUCT_LIST], [CRM_EXPORT_IF_AUTHORIZED]
- Exclusions: [SENSITIVE_DATA_EXCLUSIONS], [NO-GO CLAIMS], [POLICY_LIMITS]
- Human owner: [NAMED_REVENUE_LEAD]
- Review deadline: [DATE]

Return:
1. Prospect identity and source-cited business context, with dates.
2. Possible legitimate advertising objectives, clearly labeled as hypotheses.
3. Fit with our authorized ad products, avoiding guarantees or unsupported performance claims.
4. Evidence table: claim, source, date, confidence, verification owner.
5. Privacy, consent, copyright, licensing, and bias-risk notes.
6. Questions for a human sales lead to verify before outreach.
7. Redacted one-paragraph draft positioning statement marked “internal draft only.”
8. Items that must not be used because they are unverified, sensitive, or outside authorization.

Required inputs

• Authorized public materials about the business or category, with dates and source names.
• The newsroom’s approved media kit, ad product descriptions, and audience summaries that are permitted for sales use.
• Any CRM export only if the revenue team has permission to process it for prospect research, with personal data minimized or redacted where possible.
• A list of prohibited claims, restricted categories, privacy limits, and internal review owners.

Expected output

The model should produce a structured internal research brief that separates sourced facts from hypotheses, marks confidence levels, and identifies what a human seller must verify before contacting the prospect. Good output avoids statements such as “this business is ready to buy” unless that is directly documented in an authorized source, and it avoids making assumptions about owners, employees, customers, or community members based on protected or sensitive traits.

Verification checkpoint

A named revenue lead must check every business claim against the cited source, confirm that audience statements match approved sales materials, remove personal data that is not necessary for the outreach decision, and obtain any required advertising, legal, or publisher review before an external message is sent. No AI-generated prospect score should be treated as a final eligibility, pricing, targeting, credit, or compliance decision.

Prompt 20: Claim-safe sales brief for local sponsorship or advertising outreach

Purpose

This prompt turns approved advertising materials into a cautious internal sales brief that avoids inflated audience claims, unsupported return-on-investment promises, and fabricated case studies. It is designed for sales teams that need a practical way to translate a media kit, sponsorship inventory, and verified campaign examples into a prospect-specific conversation guide while preserving publisher credibility and legal review.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are helping prepare an internal sales brief for a local-newsroom advertising or sponsorship conversation. Use only approved materials that I provide. Do not create unsupported performance claims, audience guarantees, testimonials, case studies, pricing, availability, impressions, conversion rates, or exclusivity terms.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Inputs:
- Prospect: [PROSPECT_NAME_OR_CATEGORY]
- Approved media kit: [FILE_OR_EXCERPT]
- Approved ad products: [FILE_OR_EXCERPT]
- Verified campaign examples permitted for reuse: [FILE_OR_EXCERPT]
- Prohibited claims or sensitive categories: [LIST]
- Human approver: [NAMED_REVENUE_MANAGER]
- Publication/legal review rules: [RULES]

Create a claim-safe internal brief with:
1. Three conversation themes supported by cited approved materials.
2. A “safe to say” column and a “do not say unless verified” column.
3. Questions to ask the prospect instead of assuming goals, budget, audience, or timing.
4. Required substantiation for any performance, reach, audience, or sponsorship claim.
5. A short internal email draft to the sales lead, not to the prospect.
6. A review checklist for privacy, consent, copyright/licensing, fairness, and advertising policy compliance.
7. A list of all missing evidence and uncertainty labels.

Required inputs

• Approved and current media-kit language, including the date of the version being used.
• Approved ad products and sponsorship descriptions, with any restrictions on claims or categories.
• Only campaign examples that the newsroom is allowed to reference internally or externally.
• Human approver name and the review path for sales, legal, advertising standards, or publisher approval.

Expected output

The output should function as a conservative sales-preparation document. It should give the sales team language that is supportable, list what cannot be said without additional proof, and flag uncertainty around audience data, campaign availability, pricing, and timing. The safest briefs ask the prospect to confirm business goals rather than pretending the newsroom knows them.

Verification checkpoint

Before any outreach, a human revenue manager must compare the brief to the current media kit, remove outdated or unsupported claims, confirm that any testimonial or case study is licensed for the intended use, and route the message for required review. External outreach, campaign proposals, insertion orders, sponsorship commitments, discounts, and pricing promises require human approval.

Prompt 21: Donor-pattern analysis without sensitive-trait inference

Purpose

This prompt helps a nonprofit newsroom or development team examine donor behavior at an aggregate level without inferring protected characteristics, sensitive traits, political beliefs, health status, religion, immigration status, financial vulnerability, or other highly sensitive personal attributes. OpenAI’s newsroom examples include donor modeling among explored operational uses, but the safe version of that work must be governed by consent, data minimization, lawful authority, and organization policy.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized nonprofit newsroom development team with aggregate donor-pattern analysis. Do not infer sensitive traits. Do not score individual people for wealth, vulnerability, ideology, religion, health, immigration status, race, ethnicity, sexual orientation, union status, or other protected or highly sensitive attributes. Do not recommend targeting based on protected or sensitive personal data.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Analyze the following authorized, minimized dataset or summary:
- Dataset description: [DONOR_DATASET_DESCRIPTION]
- Fields included: [FIELDS]
- Fields excluded/redacted: [EXCLUDED_FIELDS]
- Consent and policy basis: [CONSENT_OR_POLICY_NOTE]
- Time period: [DATE_RANGE]
- Human owner: [NAMED_DEVELOPMENT_LEAD]

Return only aggregate, non-sensitive analysis:
1. Data-quality notes and missing-field warnings.
2. Aggregate giving-pattern observations by non-sensitive operational dimensions, such as campaign, channel, date range, gift size band, renewal status, or membership tenure if authorized.
3. Hypotheses about messaging, timing, or stewardship that do not depend on sensitive-trait inference.
4. Segments that should be suppressed because they are too small, potentially identifying, or policy-sensitive.
5. Privacy, consent, copyright/licensing, bias, and fairness review notes.
6. A verification table listing metric, calculation method, source file, date range, uncertainty, and human reviewer.
7. Draft stewardship experiment ideas requiring human approval before use.

Required inputs

• A minimized donor dataset or aggregate summary that the development team is authorized to analyze.
• A written note describing consent, donor-privacy policy, retention limits, and any excluded fields.
• Allowed operational dimensions, such as campaign, channel, gift band, renewal status, or date range.
• A named development lead and any legal, finance, or board-review requirements.

Expected output

The output should identify aggregate patterns, data-quality problems, and stewardship hypotheses without naming individual donors or inferring sensitive traits. For example, it may note that “renewals were lower in the email-only campaign during [DATE_RANGE]” if the data supports it, but it must not speculate that a group has a particular political view, religion, income vulnerability, health condition, or protected identity.

Verification checkpoint

This customer-success prompt collection covers churn hypotheses, onboarding signals, health scoring, and renewal strategy, providing a practical comparison point for designing bounded newsroom subscription experiments without treating modeled intent as fact. The 50 GPT-5.5 Prompts for Customer Success Managers: Churn Prediction, Onboarding Workflows, Health Scoring, and Renewal Strategy article is a focused companion for Subscription Growth Experiments because the target directly addresses churn and renewal workflows, making it substantially more precise than the draft startup-founder growth-strategy article.

Prompt 22: Churn hypothesis review for subscriptions, memberships, or newsletters

Purpose

This prompt helps audience, product, or membership teams review churn hypotheses without turning incomplete behavioral data into unsupported psychological claims. It is appropriate for aggregate subscription, membership, newsletter, or registration analysis when the newsroom has authorization to use the data and when the output is framed as a hypothesis review rather than an automated retention decision.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are helping an authorized local-newsroom audience or product team review churn hypotheses. Use only the aggregate or minimized data I provide. Do not infer sensitive traits, personal motivations, financial distress, health status, family status, political views, or protected characteristics. Do not recommend manipulative retention tactics or automated consequential decisions.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Inputs:
- Product or program: [SUBSCRIPTION/MEMBERSHIP/NEWSLETTER]
- Authorized aggregate metrics: [PASTE TABLE OR FILE SUMMARY]
- Date range and cohort definitions: [DATE_RANGE_AND_COHORTS]
- Known product changes or campaigns: [CHANGE_LOG]
- Research notes from consented surveys or support tickets: [SUMMARY]
- Excluded data: [EXCLUSIONS]
- Human owner: [NAMED_AUDIENCE_LEAD]

Produce a churn hypothesis review:
1. Data-quality and cohort-definition warnings.
2. Observed aggregate patterns with dates and citations to provided files.
3. Plausible hypotheses, each labeled as hypothesis rather than fact.
4. Alternative explanations and confounders.
5. Low-risk research questions for surveys, interviews, or usability tests.
6. Retention experiment ideas that avoid sensitive targeting and require human approval.
7. Privacy, consent, copyright/licensing, bias, redaction, and evidence-capture checklist.
8. Items that should not be acted on because the evidence is insufficient.

Required inputs

• Aggregate or minimized churn, renewal, engagement, cancellation, or unsubscribe data that the team is permitted to analyze.
• Clear cohort definitions, date ranges, and known product or pricing changes that could affect interpretation.
• Consent-appropriate survey, interview, or support-ticket summaries, with personal details removed unless necessary and authorized.
• A named product, audience, or membership owner responsible for verification and experiment approval.

Expected output

The output should distinguish observed patterns from possible explanations. For instance, it can say that cancellation rates increased after a billing-change date if the aggregate data shows that pattern, but it must not claim that users left because they were angry, financially distressed, politically aligned, or part of a sensitive group unless directly supported by authorized, consented evidence and policy permits that analysis.

Verification checkpoint

A human audience lead must verify all calculations, examine confounders, check consent and privacy limits, and approve any retention experiment before it is launched. External messages, discount offers, payment changes, account interventions, or publication of findings require human approval and applicable legal, finance, or editorial review.

Prompt 23: Print-to-digital workflow mapping for operational transition

Purpose

This prompt supports a newsroom, product, or operations team documenting a print-to-digital workflow transition. Lenfest/OpenAI’s reporting identifies print-to-digital transition and operational efficiency as areas where participating news organizations explored AI, but the safe use of a model is to create a map for human review, not to unilaterally change deadlines, labor assignments, vendor commitments, publication schedules, or reader obligations.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are assisting an authorized newsroom operations team with print-to-digital workflow mapping. Do not make staffing, labor, vendor, publication, delivery, paywall, customer-service, legal, or contractual decisions. Use only the workflow notes and documents I provide.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Map the current and proposed workflow using these authorized inputs:
- Current print workflow notes: [FILE_OR_SUMMARY]
- Current digital workflow notes: [FILE_OR_SUMMARY]
- Systems involved: [CMS], [PRINT_SYSTEM], [NEWSLETTER_SYSTEM], [ANALYTICS_SYSTEM], [ARCHIVE_SYSTEM]
- Constraints: [UNION_OR_LABOR_RULES_IF_AUTHORIZED], [VENDOR_CONTRACT_LIMITS], [EDITORIAL_DEADLINES], [ACCESSIBILITY_REQUIREMENTS], [PUBLICATION_POLICY]
- Known pain points: [LIST]
- Human owner: [NAMED_OPERATIONS_EDITOR_OR_PRODUCT_LEAD]

Return:
1. Current-state workflow table with step, owner role, system, input, output, deadline, evidence source, and date.
2. Proposed future-state workflow options, each labeled as draft.
3. Bottlenecks, duplicated work, handoff risks, accessibility risks, archive/copyright risks, and reader-impact questions.
4. Decisions that require human, legal, labor, vendor, finance, or editorial approval.
5. Small pilot plan with success measures and rollback criteria.
6. Privacy, consent, copyright/licensing, bias, redaction, and evidence-capture notes.
7. Uncertainties and missing documents.

Required inputs

• Current print and digital workflow notes, preferably collected from authorized staff interviews, system documentation, and deadline calendars.
• A list of systems involved, such as the CMS, print production tools, archive, newsletter tool, analytics system, and customer-service platform.
• Known constraints, including labor agreements, vendor contracts, accessibility requirements, archival obligations, and editorial standards, only to the extent the user is authorized to process them.
• A named operations, product, or editorial owner who can validate the map.

Expected output

The output should be a workflow artifact that humans can annotate: current-state steps, proposed options, risk points, missing evidence, and pilot ideas. It should not assign blame to teams, disclose sensitive personnel information, or imply that automation should replace editorial judgment, accessibility review, copy editing, or customer-service accountability.

Verification checkpoint

A human operations owner must review the workflow with affected teams before any change is made. Changes to publication schedules, job duties, reader communications, print delivery, vendor obligations, access controls, payment flows, or archive handling require the appropriate editorial, legal, labor, finance, accessibility, and executive approvals.

Prompt 24: Cross-newsroom reusable-tool brief

Purpose

This prompt helps a newsroom convert a successful internal experiment into a reusable-tool brief for other local news organizations. The Lenfest/OpenAI expansion announcement says the next phase aims to turn successful projects into reusable tools, frameworks, plugins, implementation guides, playbooks, and technical resources for hundreds of organizations. This prompt keeps that translation practical by forcing the team to document the problem, constraints, evidence, permissions, failure modes, and support needs before sharing anything.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are helping an authorized newsroom team draft a reusable-tool brief for possible sharing with other news organizations. Do not claim universal results, independent proof, production readiness, security certification, legal compliance, or suitability for another newsroom unless the provided evidence supports it. Do not include secrets, credentials, private data, proprietary files, or unlicensed content.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Inputs:
- Tool or workflow name: [PROJECT_NAME]
- Problem statement: [PROBLEM]
- Users and roles: [ROLES]
- Systems and data sources: [SYSTEMS_AND_DATA]
- Evidence from pilot: [AUTHORIZED_SUMMARY_WITH_DATES]
- Known limitations: [LIMITATIONS]
- Security/privacy/copyright review status: [STATUS]
- Maintenance owner: [OWNER]
- Sharing goal: [INTERNAL_GUIDE/OPEN_SOURCE/CONSORTIUM_PLAYBOOK/VENDOR_HANDOFF]
- Human editor: [NAMED_EDITOR_OR_PROJECT_SPONSOR]

Create a reusable-tool brief:
1. One-paragraph problem definition that starts with the newsroom need, not the technology.
2. User workflow summary with required permissions and authorization boundaries.
3. Minimum viable implementation requirements.
4. Data, privacy, consent, copyright/licensing, and retention requirements.
5. Editorial guardrails and human-review points.
6. Bias and community-impact risks.
7. Security review checklist, including secrets removal and access-control review.
8. Evidence table: what was tested, date range, source, result, uncertainty, and whether it is source-reported or independently validated.
9. Adaptation questions another newsroom must answer before reuse.
10. Release recommendation: do not share yet, share internally, share with limited partners, or prepare for broader release, with reasons.

Required inputs

• A project summary that defines the newsroom problem, not just the tool or model used.
• Evidence from a pilot, with dates, scope, limitations, and whether results were independently validated or only internally observed.
• Documentation of data sources, permissions, privacy review, copyright/licensing status, security review, and maintenance ownership.
• The intended sharing path, such as internal playbook, consortium guide, open-source release, vendor handoff, or fellowship documentation.

Expected output

The output should be a decision-ready brief that helps leaders decide whether the project can be reused safely. It should highlight local dependencies that may not transfer, such as archive structure, CMS configuration, language capacity, moderation policy, community trust, staffing, and legal constraints. It must avoid implying that a pilot result from one newsroom proves a universal outcome for another.

Verification checkpoint

A named editor, product owner, security reviewer, and legal or licensing reviewer must approve any external sharing. Code, prompts, documentation, screenshots, logs, transcripts, and datasets must be checked for secrets, private information, source confidentiality, copyrighted material, license compatibility, and unsupported performance claims before release.

Prompt 25: Final editorial release checklist for AI-assisted newsroom work

Purpose

This prompt provides a final gate for AI-assisted newsroom work before publication, outreach, product launch, donor communication, advertising contact, workflow change, or cross-newsroom sharing. It operationalizes OpenAI’s prompt-engineering guidance to be explicit about task goals, context, desired output, and evaluation, while preserving the newsroom principle that human editors and accountable managers—not model outputs—make final decisions.

Copy-paste prompt

CE113 NEWSROOM SAFETY CONTRACT:
- Use only systems, accounts, records, documents, data, and materials the newsroom owns or is explicitly authorized to use.
- Do not fabricate sources, quotations, records, documents, dates, facts, metrics, identities, or outcomes.
- Cite the source identifier and date for every factual item; label uncertainty or missing evidence.
- Check copyright and licensing, privacy and consent, and redact unnecessary personal data.
- A named human editor must review and approve the output before publication or consequential action.
You are performing a final internal release-readiness review for AI-assisted newsroom work. You are not the final approver. Identify risks, missing evidence, unsupported claims, privacy issues, copyright/licensing issues, bias concerns, and required human approvals. Do not rewrite uncertain material as fact.

Safety contract: Scope your analysis to data, archives, systems, accounts, documents, and workflows that our newsroom owns or is explicitly authorized to use. Use placeholders for names, files, systems, metrics, audience segments, documents, and records. Never fabricate interviews, quotations, sources, documents, public-record contents, donor attributes, audience data, advertising claims, or translation certainty. Require source citations, dates, uncertainty labels, privacy/consent checks, copyright and licensing review, bias checks, redaction, evidence capture, and a named human editor before external publication or consequential action. Outputs are drafts or leads, not publish-ready facts. Do not infer sensitive traits or target people using protected or highly sensitive personal data.

Review this AI-assisted work package:
- Project type: [STORY/TRANSLATION/TRANSCRIPT/ARCHIVE_SEARCH/AUDIENCE_RESEARCH/AD_SALES/DONOR_ANALYSIS/WORKFLOW_TOOL/OTHER]
- Draft or brief: [PASTE_TEXT_OR_SUMMARY]
- Source list: [SOURCES_WITH_DATES]
- AI tool use summary: [HOW_AI_WAS_USED]
- Human reviewers so far: [NAMES_OR_ROLES]
- Intended external action: [PUBLICATION/EMAIL/CAMPAIGN/TOOL_RELEASE/INTERNAL_ONLY]
- Deadline: [DATE]
- Applicable policy: [EDITORIAL_POLICY/LEGAL_REVIEW/PRIVACY_POLICY/AD_POLICY/DONOR_POLICY]

Return a release checklist:
1. Source verification status for every factual claim, quote, document, public record, data point, and date.
2. Items that need uncertainty labels or should be removed.
3. Privacy, consent, copyright/licensing, redaction, and data-retention issues.
4. Bias, fairness, accessibility, language, and community-impact concerns.
5. AI-use disclosure questions for the newsroom’s policy.
6. Human approvals still required, by role and reason.
7. Evidence-capture requirements: files, notes, diffs, transcripts, citations, and review signoffs.
8. Final recommendation: hold, revise, legal/privacy review, editor review, or ready for named human approval.
9. A concise “do not publish or send until” list.

Required inputs

• The draft, brief, workflow document, campaign copy, tool documentation, or research output under review.
• A complete source list with dates, including public records, interviews, archives, datasets, transcripts, translations, analytics summaries, or approved sales materials.
• A description of how ChatGPT, Codex, transcription tools, translation tools, or other AI systems were used.
• The newsroom’s applicable editorial, privacy, legal, advertising, development, accessibility, and AI-use disclosure policies.

Expected output

The output should be a release-readiness checklist that makes the remaining human work visible. It should not declare a piece publishable merely because sources are listed; it should identify unverified facts, missing dates, unclear permissions, unsupported claims, overconfident translations, small-sample audience findings, copyrighted material, personal data, and approvals that are still missing.

Verification checkpoint

A named human editor or accountable manager must make the final decision and preserve the evidence package. Publication, external outreach, donor segmentation, advertising proposals, workflow changes, product releases, permissions changes, payments, purchases, legal commitments, or public claims must not proceed on model output alone.

Final operating guidance for local newsroom prompt use

Treat every prompt as a controlled operating template, not a publishing shortcut. The workflow begins with authorized inputs and a defined newsroom problem, continues through source and rights verification, and ends with a named human editor who can reject, revise, or stop the work. Drafts, rankings, translations, audience hypotheses, revenue ideas, and workflow maps remain unverified until the assigned owners review the underlying evidence.

For audience, revenue, and development work, use approved aggregate or first-party data and avoid protected or sensitive-trait inference. For translation, accessibility, archives, and public records, preserve original context, corrections, dates, quotations, and licensing restrictions. For public-facing claims, retain a claim-evidence ledger and a correction path. For workflow changes, pilot narrowly, log outcomes, define stop conditions, and maintain rollback.

The practical standard is simple: no fabricated facts, no concealed uncertainty, no unauthorized access, no unreviewed publication, and no consequential action based on generated text alone.

Useful Links

• The Lenfest Institute grows landmark program with expanded OpenAI support
• How news organizations are using AI
• OpenAI prompt engineering guide
• OpenAI usage policies