25 ChatGPT Prompts for Human-Led RFP Evaluation: Requirement Traceability, Evidence Matrices, Scoring Calibration, Conflict Flags, and Award-Memo Drafting

A bounded, human-led workflow for evaluating request for proposal (RFP)A formal solicitation asking suppliers to submit proposals against stated requirements and evaluation criteria. Open glossary entry responses

An RFP evaluation should begin with an authority boundary, not with document upload. The responsible procurement team must identify who may access proposals, who may evaluate them, which solicitation documents control, which communications are permitted, and who has authority to approve scores, exclusions, consensus findings and the award. ChatGPT can help organise authorised evidence and draft review materials, but it must not choose a supplier or make final determinations about responsiveness, responsibility, technical acceptability, value, conflicts, debarment or legal compliance.

Evidence checkpoints

Documented point: A ChatGPT Project can bound a solicitation, vendor responses, rubric, instructions, and saved review artefacts instead of scattering work across unrelated chats. Projects and sharing depend on plan/workspace settings and do not themselves establish a procurement record, retention schedule, or access authorisation. [official source 1 official source 2]

Documented point: ChatGPT, given suitable prompts, can compare submitted responses to an organisation-supplied rubric, extract criterion evidence and organise it in review tables. Supply the approved rubric, weights, rules and an authoritative source pack; each provisional finding should cite page or section evidence and flag absent, ambiguous or conflicting evidence. [official source 1 official source 2]

Documented point: ChatGPT can generate structured tables and some Python-backed calculations, but reviewers should inspect method, outputs, and assumptions. This does not validate procurement-policy compliance; exact extraction can fail with scanned, image-heavy, complex, large, or poorly structured files. [official source 1]

Documented point: For listed business offerings, OpenAI says business data is not used for training by default. Apply this statement only to the actual offering and configuration in use; it does not automatically extend to consumer accounts or third-party connected services. [official source 1 official source 2]

The workflow in this guide assumes that the organisation already has a final solicitation, approved addenda, mandatory requirements, an evaluation rubric, criterion weights where applicable, evaluation rules and authorised versions of submitted proposals. If any of those inputs remain disputed or incomplete, pause the evaluation and obtain direction from the named procurement official. A model should not infer which draft controls, repair missing rules or convert informal evaluator preferences into new criteria.

Establish authority, confidentiality and records controls first

Before uploading proposal material, document the authorised purpose, approved users and applicable handling restrictions. Proposals may contain confidential commercial information, personal data, export-controlled material, security-sensitive architecture or competitively sensitive pricing. Confirm the actual ChatGPT account and workspace, user permissions, sharing configuration, connected-app access, organisational approval and applicable retention requirements. Do not assume that creating a Project supplies any of those approvals.

OpenAI describes Projects as workspaces that can hold chats, files and project instructions for ongoing work. This can help keep a solicitation, rubric, authorised proposal set and review artefacts together rather than distributing them across unrelated chats. However, availability and sharing depend on the plan and workspace configuration, and a Project does not itself establish procurement authorisation, an official record, a retention schedule or access rights.

OpenAI states that business data in its listed business offerings is not used for training by default. Apply that statement only to the actual offering and configuration in use. Do not extend it to a consumer account or to third-party connected services. The procurement, privacy, information-security and records teams should determine whether the proposed use is authorised before any restricted file is uploaded.

Non-selection boundary: every extraction, comparison, flag and score produced through these prompts is provisional working material. The evaluation committee and authorised procurement official remain responsible for source checking, individual scores, consensus decisions, conflict handling, clarifications, exclusions, communications, award recommendations and the final procurement record.

Define the controlled source pack

Create a source register before extracting requirements or reviewing proposal evidence. Give every authorised file a stable source identifier, record its title, version, date, status and confidentiality classification, and state whether it is controlling, superseded, proposal-specific or reference-only. Preserve corrections and version history according to organisational policy rather than silently replacing earlier files.

A practical source hierarchy normally distinguishes the final solicitation and approved addenda from evaluator guidance, proposal submissions and administrative records. The procurement official must confirm the actual order of precedence. ChatGPT should report a conflict between sources rather than deciding which document prevails unless the approved rules expressly resolve that conflict.

Source class	Minimum register fields	Human control
Solicitation and addenda	Source identifier (ID)A value used to distinguish one record, task, source or object from another. Open glossary entry, title, version, issue date, effective date, page count, status	Procurement official confirms the controlling version and precedence
Evaluation rubric and rules	Criterion IDs, weights, scale, scoring anchors, mandatory rules, exceptions	Committee uses only pre-approved criteria and interpretations
Submitted proposal	Bidder-neutral ID, authorised version, receipt date, file list, page or section scheme	Procurement staff confirm that the submission is complete and permitted for review
Clarification or approved amendment	Issuer, recipient, date, authority, affected requirement, permitted use	Only authorised communications enter the evaluation record
Administrative record	Conflict declarations, attendance, approvals, correction log, access restrictions	Handled under applicable confidentiality and records procedures

For source-controlled deep-research workflows with plan review and citation checks, see 25 ChatGPT-5.5 Prompts for Source-Controlled Deep Research: Plan Review, Domain Filters, Live Steering, Citation Checks, and Decision Artifacts.

Account for extraction and calculation risk

OpenAI documents file-upload uses that include applying an organisation-supplied framework or rubric, extracting information and comparing documents. OpenAI also documents structured tables and calculations with Python in data analysis. These capabilities can support evidence organisation, but they do not validate procurement-policy compliance or establish that every source was extracted correctly.

Scanned pages, image-based PDFs, complex tables, embedded diagrams and poorly structured documents require direct human checking where exact wording affects a mandatory requirement or score. Record unreadable pages and extraction failures explicitly. Do not convert “not located” into “not provided” until an authorised reviewer has checked the relevant proposal source, including permitted appendices and cross-references.

Keep price and non-price evaluation separate wherever the approved process requires separation. For later calculations, provide the authorised formula, source inputs, rounding method, treatment of blanks and an exceptions list. A human reviewer should independently reconcile totals and verify that no unstated weight, normalisation rule or assumption has entered the calculation.

For supply-chain vendor-risk and operations prompts, see 35 ChatGPT-5.5 Prompts for Supply Chain Managers: Demand Forecasting, Inventory Optimization, Logistics Planning, and Vendor Risk Assessment.

Prompts 1–6: establish the evaluation controls

Prompt 1: Confirm authority and evaluation boundaries

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Use this prompt to convert the procurement official’s instructions into a reviewable operating boundary. It should identify missing approvals without interpreting law or policy, and it should prevent the model from beginning substantive evaluation before the source and authority conditions are satisfied.

Copy-paste prompt

You are supporting a human-led RFP evaluation. Do not select a supplier, recommend an award, determine responsiveness or responsibility, make a legal or policy conclusion, or communicate with bidders.

Use only the authorised inputs listed below. Do not fabricate missing approvals, facts, criteria, dates, source text or proposal evidence. Do not use external knowledge or research bidders.

Create an evaluation-boundary checklist with:
1. authorised purpose and procurement reference;
2. named human procurement official with final review authority;
3. authorised evaluators and permitted access;
4. controlling solicitation, addenda, rubric and evaluation rules;
5. authorised proposal versions;
6. permitted and prohibited uses of ChatGPT;
7. confidentiality, rights, privacy, records, retention and connected-service restrictions;
8. conflict-of-interest and recusal procedure supplied by the organisation;
9. price/non-price separation requirements;
10. unresolved prerequisites that require human direction.

For every factual statement, cite the supplied source identifier and page, section or record reference. Include the source date or document date where available. If pagination is absent, cite a stable heading and file name. Mark each item Confirmed, Unconfirmed, Conflicting or Not supplied. Express uncertainty explicitly and do not infer approval from silence.

Treat all rights, confidentiality and privacy labels as handling constraints. Do not reproduce restricted content beyond what is necessary for the checklist. End with a gate stating either “Ready for named human procurement review” or “Pause pending named human procurement review”, followed by the reasons. The named human procurement reviewer is [NAME AND ROLE].

Required inputs

• Procurement reference, authorised purpose and named procurement official.
• Approved evaluation procedure, evaluator list and role assignments.
• Account, workspace, access, retention and connected-app approval records.
• Confidentiality, privacy, records and information-security handling instructions.
• Conflict declaration and recusal procedure.
• List of controlling solicitation and proposal sources, with dates and identifiers.

Expected output

A status-controlled checklist that distinguishes confirmed authority from missing or conflicting instructions. The output should expose blockers without attempting to resolve them and should end with a clear human-review gate.

Verification checkpoint

The named procurement official checks every “Confirmed” item against its cited source and resolves all material “Unconfirmed” or “Conflicting” entries. Do not proceed merely because the model labels the package ready. Verify that every evaluator has appropriate access and that recused personnel cannot view or influence material prohibited by the approved process.

Prompt 2: Build the authoritative source register

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

This prompt creates the inventory needed for traceability. It separates controlling documents from superseded drafts and reference material while preserving uncertainty about precedence for human resolution.

Copy-paste prompt

Build a source register for this human-led RFP evaluation using only the authorised files and metadata supplied. Do not fabricate titles, versions, dates, page counts, classifications, precedence rules or missing files. Do not access external sources or research any bidder.

Assign or preserve a unique source identifier for each item. Produce columns for:
- source ID;
- document title;
- source class;
- bidder-neutral proposal ID, if applicable;
- file name;
- version or revision;
- document date and received date;
- page or section reference scheme;
- status: controlling, authorised proposal, approved clarification, superseded, reference-only, or unresolved;
- rights, confidentiality and privacy handling label;
- authorised audience;
- retention or record instruction supplied;
- relationship to another source;
- extraction issue;
- uncertainty or conflict;
- required human action.

Cite the source identifier and page, section or record reference for each derived entry, and preserve all supplied dates. If metadata and document content disagree, report both values and flag the conflict. Do not decide precedence unless an authorised rule expressly states it; cite that rule if used.

Minimise reproduction of confidential or personal content. Record handling labels without weakening them. Identify image-based, scanned, unreadable, incomplete or complex files that need manual checking.

Conclude with:
1. controlling-source candidates;
2. superseded or duplicate candidates;
3. missing expected items;
4. version conflicts;
5. files unsuitable for reliable automated extraction;
6. questions for named human procurement review.

The named human procurement reviewer is [NAME AND ROLE]. No source becomes authoritative until that reviewer confirms it.

Required inputs

• File manifest and authorised document set.
• Existing source identifiers, if the organisation already uses them.
• Document metadata, receipt records and version history.
• Approved source-precedence rule, if one exists.
• Handling classifications and permitted audience.

Expected output

A source register with stable identifiers, version relationships, extraction warnings and unresolved authority questions. It should be suitable for later citation in requirement and evidence matrices, subject to human correction.

Verification checkpoint

Procurement staff compare the register with the official receipt log and document repository. Manually open each file marked scanned, incomplete or unreadable. Confirm that bidder files have not been mixed, that superseded versions remain identifiable as records where required, and that the register’s “controlling” status matches a human-approved authority decision.

Prompt 3: Reconcile solicitation versions and addenda

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Use this prompt to identify where approved addenda change dates, requirements, submission instructions or evaluation provisions. The output is a change map, not an interpretation of bidder compliance.

Copy-paste prompt

Compare the authorised solicitation versions and approved addenda listed in the confirmed source register. Use no other material. Do not fabricate changes, effective dates, replacement text, precedence or legal effects.

Create an addendum reconciliation table containing:
- change ID;
- earlier source ID and page/section;
- later source ID and page/section;
- document dates;
- affected requirement or instruction ID, if supplied;
- exact limited quotation needed to locate the change;
- change type: added, deleted, replaced, clarified, renumbered, date changed, weight changed, or unresolved;
- practical evaluation impact stated only from the approved text;
- uncertainty or conflict;
- required human procurement decision.

Every entry must cite source identifiers and page or section references. Preserve dates exactly. If text appears similar but its effect is uncertain, label it “Possible change—human interpretation required”. Do not infer that later-dated text prevails unless the authorised precedence rule says so.

Respect all rights, confidentiality and privacy restrictions. Quote only the minimum text necessary for traceability. Do not include proposal content in this task.

Finish with a consolidated list of provisions that appear current, but label it provisional pending named human procurement review. Identify any changed mandatory requirement, evaluation criterion, criterion weight, deadline or price instruction prominently without deciding its validity.

The named human procurement reviewer is [NAME AND ROLE], who must approve the reconciliation before requirement extraction or scoring begins.

Required inputs

• Human-confirmed source register.
• Final solicitation and each approved addendum.
• Approved precedence and effective-date rules.
• Existing requirement identifiers or amendment logs, where available.

Expected output

A source-cited change table and a provisional consolidated provision list. Material changes to mandatory requirements, weights, dates and pricing instructions should be easy for the procurement reviewer to locate.

Verification checkpoint

The procurement reviewer checks every reported change against both source passages and confirms the authorised interpretation. Pay particular attention to renumbered sections and tables whose headings continue on another page. No evaluator should use the consolidated list until the responsible official has approved it or supplied a corrected version.

Prompt 4: Extract the approved requirement catalogue

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

This prompt turns the confirmed solicitation into a traceable requirement catalogue. It distinguishes mandatory requirements, scored criteria, submission instructions and contractual information without deciding whether any proposal satisfies them.

Copy-paste prompt

Extract a provisional requirement catalogue from the authorised, human-confirmed solicitation and addenda only. Do not use proposal content, external facts or assumed procurement practice. Do not fabricate requirement IDs, obligations, thresholds, dates, evidence expectations or consequences.

For each requirement, provide:
- requirement ID;
- requirement type: mandatory, scored criterion, submission instruction, informational, contractual, pricing-related, or unresolved;
- concise requirement text;
- actor responsible;
- required action, deliverable or condition;
- stated threshold or deadline;
- stated evidence or response location, if any;
- source ID;
- page and section reference;
- source date;
- relationship to an addendum;
- ambiguity, internal cross-reference or conflict;
- rights, confidentiality or privacy handling note;
- named human review status.

Use minimal quotations and preserve controlling defined terms. If a requirement is embedded in a table, footnote, attachment or cross-reference, cite every necessary source location. If exact extraction is uncertain because of a scan, image, layout or missing page, mark “Manual source check required”.

Do not classify a provision as mandatory merely because its wording appears important. Use only the approved organisation-supplied classification rule; otherwise mark the type unresolved. Do not score, rank or assess bidders.

End with lists of duplicate candidates, compound requirements that may need human splitting, unresolved cross-references and possible unstated assumptions. All findings remain provisional until review by [NAME AND ROLE], the named human procurement reviewer.

Required inputs

• Approved solicitation and addendum reconciliation.
• Organisation-supplied rules for classifying mandatory and scored provisions.
• Defined terms, attachments and incorporated schedules permitted for use.
• Confirmed source identifiers and page-reference scheme.

Expected output

A requirement catalogue in which every row traces to controlling text and no bidder has yet been assessed. Ambiguous classifications, compound clauses and extraction risks should remain visible rather than being silently normalised.

Verification checkpoint

Two authorised reviewers should check high-consequence mandatory requirements and any provision associated with rejection, pass/fail treatment or a scoring threshold. Confirm each citation directly in the source. The procurement official should approve any splitting, merging or reclassification because those changes can alter how proposal evidence is evaluated.

Prompt 5: Convert the approved rubric into a criterion dictionary

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

This prompt converts the approved evaluation rubric into a structured dictionary of criteria, weights, scoring anchors and evidence boundaries. It must not invent an interpretation where the approved rubric is silent.

Copy-paste prompt

Transform the authorised evaluation rubric and rules into a provisional criterion dictionary. Use only the supplied, approved sources. Do not invent criteria, subcriteria, weights, scoring anchors, rounding rules, pass marks, tie-breakers, price formulas or evaluator discretion.

For each criterion, show:
- criterion ID and approved name;
- criterion type: price, non-price, pass/fail, or unresolved;
- approved weight or points;
- scoring scale;
- verbatim or minimally quoted scoring anchors;
- evidence evaluators are instructed to consider;
- evidence evaluators are prohibited from considering;
- relationship to requirement IDs;
- blank, not-applicable and exception handling;
- calculation or rounding rule;
- source ID, page/section reference and document date;
- ambiguity, inconsistency or missing rule;
- required named human procurement review.

Check arithmetic only using the supplied figures and formula. Show the calculation method, inputs and exceptions. If weights do not reconcile, report the difference without correcting it. Keep price and non-price criteria separate when the authorised process requires it.

Do not add external bidder information or convert desirable features into criteria. Do not infer that a proposal deserves a score. Preserve rights, confidentiality and privacy restrictions, and minimise copied text.

Finish with a rubric-integrity report covering total available points, weight reconciliation, duplicate criteria, gaps between requirements and criteria, conflicting anchors and questions requiring [NAME AND ROLE], the named human procurement reviewer.

Required inputs

• Final approved rubric, scoring instructions and evaluator guidance.
• Criterion weights, scales, anchors and authorised calculation formulae.
• Approved rules for blanks, exceptions, rounding and price separation.
• Human-approved requirement catalogue.

Expected output

A criterion dictionary and rubric-integrity report that preserve the approved scoring design. Any unreconciled total, missing anchor or mismatch between requirements and criteria should be escalated rather than repaired.

Verification checkpoint

The procurement lead independently reconciles points and weights, preferably outside the model-generated table. Confirm that each scoring anchor is quoted or paraphrased accurately, price treatment remains separate where required, and no new criterion has been introduced through a column label, summary or suggested interpretation.

Prompt 6: Create the proposal intake and citation map

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Use this prompt to map each authorised proposal’s files, sections, appendices and internal cross-references before extracting evidence. The map helps reviewers locate material while avoiding premature conclusions about completeness or compliance.

Copy-paste prompt

Create a proposal intake and citation map for one authorised proposal version. Use only the proposal files, receipt metadata and confirmed source register supplied for this bidder-neutral proposal ID. Do not compare bidders, score content, determine compliance, use external information or fabricate missing sections.

Produce:
1. a file inventory with source ID, file name, version, document date, received date, page count if reliably available, and rights/confidentiality/privacy label;
2. a section map with headings, page ranges and stable citation references;
3. an appendix and attachment map;
4. a list of internal cross-references and whether the referenced location was found;
5. a table of unreadable, image-based, complex, password-protected, corrupted, truncated or apparently missing material;
6. a list of duplicate or conflicting files;
7. a list of proposal statements whose dates or source locations are uncertain.

Cite every mapped item with source identifier and page, section or record reference. Preserve dates exactly. Do not state “not provided” merely because a search did not find text; use “not located in extracted text—manual review required”. Do not infer the bidder’s intent.

Apply supplied rights, confidentiality, privacy and access restrictions. Minimise reproduction of proposal content and do not expose personal or restricted data in summaries. Keep this proposal isolated from other bidders’ files.

Conclude with an intake status of Ready for evidence extraction, Ready with manual checks, or Pause for named human procurement review. Explain uncertainty and list the required checks. The named human procurement reviewer is [NAME AND ROLE], and only that reviewer may confirm the authorised proposal set.

Required inputs

• One bidder-neutral proposal ID and its authorised file set.
• Official receipt record and confirmed proposal version.
• Source register entries and handling restrictions.
• Permitted citation convention for pages, sections, worksheets and attachments.
• Known decryption or accessibility procedure handled through authorised organisational channels.

Expected output

A proposal-specific navigation and citation map, plus a list of files or pages requiring direct inspection. The map should isolate bidders, preserve dates and provide stable references for later evidence extraction without asserting that absent search results prove missing evidence.

Verification checkpoint

An authorised reviewer compares the map with the official receipt package, checks page ranges and opens every flagged file. Confirm that appendices, spreadsheets and cross-referenced attachments are associated with the correct proposal. The reviewer must approve the intake status before any evidence matrix, provisional score or comparative analysis is produced.

Prompts 7–13: trace requirements, organise evidence, flag contradictions and reconcile calculations

This stage converts the controlled source pack into review material for evaluators. OpenAI documents that uploaded files can be compared with an organisation-supplied framework or rubric and that ChatGPT can organise extracted material into tables. Those capabilities do not establish whether a proposal is responsive, technically acceptable or eligible for award. Treat every output below as a source-location aid that the responsible evaluators must verify against the approved solicitation, addenda, evaluation rules and authorised proposal version.

Run these prompts separately for each proposal unless the approved process expressly permits cross-proposal analysis at that stage. Keeping proposal workspaces or chats distinct reduces the risk of attributing one bidder’s statement to another. If a shared ChatGPT Project is used, its files, instructions and permissions can help contain the work, but the Project does not create an official procurement record, confer access authority or determine the applicable retention schedule.

For OpenAI OneGov 2.0 access, pricing and procurement boundaries, see OpenAI OneGov 2.0 procurement guide.

Recommended operating rule: require page, section, table or worksheet citations for every extracted statement. Where a source is scanned, image-heavy, poorly structured or contains complex tables, compare the cited passage directly with the authoritative original. OpenAI documents that exact extraction can be affected by source format, so an apparently precise table should not be accepted without source checking when it could affect evaluation.

Prompt 7: Build the mandatory-requirement evidence worksheet

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Create a proposal-specific worksheet that maps every approved mandatory requirement to the bidder’s submitted evidence without making the final responsiveness decision.

Required inputs

Provide the final mandatory-requirement catalogue from Prompt 4, the authorised proposal version, applicable addenda, the approved rules for acceptable evidence, and any permitted clarification record already incorporated into the evaluation file. Exclude external bidder research, prior dealings and facts not authorised by the evaluation rules.

Copy-paste prompt

You are supporting a human-led Request for Proposal evaluation. Build a mandatory-requirement compliance worksheet for one proposal. Do not decide responsiveness, eligibility, responsibility, legal compliance, technical acceptability or exclusion.

Authoritative inputs:
1. Mandatory-requirement catalogue: [PASTE OR IDENTIFY FILE]
2. Final solicitation and applicable addenda: [IDENTIFY FILES]
3. Authorised proposal version: [IDENTIFY BIDDER AND FILE VERSION]
4. Approved evidence and evaluation rules: [PASTE OR IDENTIFY FILE]
5. Authorised clarification material, if any: [IDENTIFY FILE OR STATE NONE]

For every mandatory requirement, return one row with:
- requirement ID;
- exact or faithfully condensed requirement text;
- controlling solicitation or addendum citation;
- bidder evidence citation, including file, page, section, table or worksheet cell where available;
- short neutral extract or paraphrase of the bidder’s statement;
- evidence status: located, partially located, not located, ambiguous, or conflicting;
- condition, exception, dependency or qualification stated by the bidder;
- extraction warning where the evidence comes from a scan, image, complex table or unclear layout;
- reviewer action required.

Apply these rules:
- Preserve the requirement catalogue’s IDs and order.
- Do not reinterpret “shall”, “must”, deadlines, signatures, certifications, formats or minimum thresholds.
- Do not treat marketing language, an implied capability or evidence for another requirement as proof unless the approved rules permit that inference.
- Do not convert “not located” into “non-compliant”.
- Do not repair omissions, assume intent or invent evidence.
- If proposal evidence appears inconsistent with an addendum or another proposal section, flag it without resolving it.
- Quote only enough text to support source checking.
- End with a count by evidence status and an exceptions list. Label both as extraction aids, not findings.
- State: “An authorised human reviewer must verify every citation and make any responsiveness or exclusion decision under the controlling rules.”

Expected output

A requirement-by-requirement worksheet in the same order as the approved catalogue, followed by status counts and an exceptions list. The useful distinction is between evidence location and procurement judgement: “located” means that potentially relevant text was found, not that the requirement has been satisfied.

Verification checkpoint

Check every requirement classified as mandatory, particularly signatures, dates, licences, certifications, minimum experience, submission forms and threshold values. Verify whether the controlling rules permit clarification, waiver or cure before communicating with a bidder; the model must not suggest that an omission can be repaired. Record corrections in the official evaluation file according to policy rather than silently editing the generated worksheet.

Prompt 8: Assemble a criterion-level evidence matrix

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Organise proposal evidence under the approved non-price evaluation criteria so that evaluators can find relevant passages without introducing new criteria, weights or comparative claims.

Required inputs

Supply the criterion dictionary created from the approved rubric, the authorised proposal version, evaluator instructions, and the solicitation sections defining the requested response. Include criterion weights only if evaluators are authorised to see and use them at this stage.

Copy-paste prompt

Create a criterion-level evidence matrix for human review of this proposal. This is evidence organisation, not scoring or supplier selection.

Use only:
- approved criterion dictionary: [PASTE OR IDENTIFY FILE];
- authorised proposal version: [IDENTIFY FILE];
- final solicitation and addenda: [IDENTIFY FILES];
- approved evaluator instructions: [PASTE OR IDENTIFY FILES].

For each approved criterion and subcriterion, provide:
1. criterion ID and title;
2. approved criterion description;
3. approved weight or points, only if supplied and permitted;
4. response locations requested by the solicitation;
5. proposal evidence citations;
6. concise evidence summary using neutral language;
7. bidder-stated deliverable, method, resource, timetable or control;
8. assumptions, dependencies, exclusions or qualifications attached to that evidence;
9. evidence type: commitment, description, example, reference, plan, schedule, certification, quantitative claim, or other;
10. evidence-location status: direct, distributed across sections, ambiguous, or not located;
11. source-quality warning for scanned pages, image-based content, complex tables or unclear cross-references;
12. questions for the evaluator to verify.

Rules:
- Preserve the approved criterion structure and ordering.
- Do not create an evaluation criterion from a persuasive proposal feature.
- Do not award points or use adjectives such as excellent, weak, superior, credible or high risk.
- Distinguish a future commitment from evidence of previous performance.
- Distinguish a bidder assertion from independently verified information.
- Do not research the bidder or supplement the record with external facts.
- If one passage may apply to several criteria, cite it in each relevant row but identify the duplication.
- If evidence is spread across files, list every source needed to understand it.
- End with a coverage map showing which criteria have direct, distributed, ambiguous or unlocated evidence.
- State that evaluators remain responsible for relevance, interpretation and scoring.

Expected output

A traceable matrix that separates what the bidder says from how evaluators may later judge it. The evidence-type field helps prevent a case study, aspiration or general company description from being treated automatically as a binding delivery commitment.

Verification checkpoint

Confirm that every criterion came from the approved rubric and that no attractive proposal feature has become an unstated criterion. Review duplicated evidence carefully: the same paragraph may be relevant to several criteria, but it should not receive unintended multiple credit where the scoring rules prohibit that treatment. Evaluators should add their own notes and scores only in the approved record or scoring tool.

For website quality assurance, privacy, accessibility and stakeholder-approval prompts, see 25 ChatGPT-5.5 Prompts for Website QA, Privacy Reviews, Accessibility Checks, and Stakeholder Approval.

Prompt 9: Produce a missing, ambiguous and insufficient-evidence register

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Isolate evidence gaps that require evaluator attention while avoiding the unsupported conclusion that a bidder failed a requirement or deserves a lower score.

Required inputs

Use the completed mandatory-requirement worksheet, criterion evidence matrix, approved rules governing clarification and the authorised proposal files. If clarification rules are not supplied, instruct ChatGPT to mark the procedural route as unknown rather than proposing contact with the bidder.

Copy-paste prompt

Review the supplied mandatory-requirement worksheet and criterion evidence matrix for evidence gaps. Produce an issue register, not a compliance decision, score adjustment or bidder communication.

Inputs:
- mandatory-requirement worksheet: [PASTE OR IDENTIFY FILE];
- criterion evidence matrix: [PASTE OR IDENTIFY FILE];
- authorised proposal files: [IDENTIFY FILES];
- approved clarification and evaluation rules: [PASTE OR IDENTIFY FILE, OR STATE NOT SUPPLIED].

Classify each issue as one of:
- not located: no relevant evidence found in the authorised proposal after checking stated cross-references;
- incomplete: some requested elements are present, but one or more expressly requested elements were not located;
- ambiguous: text was located but supports more than one reasonable reading;
- inaccessible or unreadable: the content cannot be reliably interpreted from the supplied file;
- citation failure: a proposal cross-reference points to a missing, incorrect or unclear location;
- rule-dependent: significance depends on an evaluation, waiver, cure or clarification rule that has not been supplied.

For every issue, provide:
1. issue ID;
2. linked requirement or criterion ID;
3. exact requested element;
4. locations checked;
5. what was located;
6. what was not located or remains unclear;
7. any relevant bidder qualification or cross-reference;
8. potential evaluation dependency stated neutrally;
9. governing rule needed;
10. authorised human action: source check, evaluator interpretation, procurement review, legal review where organisational policy requires it, or no action pending;
11. confidence in the extraction only: high, medium or low, with a short reason.

Constraints:
- Do not call an issue a deficiency, disqualification, non-compliance, material omission or clarification opportunity unless an authorised human has already made and documented that determination.
- Do not draft a bidder question.
- Do not recommend waiving, curing or clarifying an issue.
- Do not infer evidence from company reputation, external websites or another bidder’s proposal.
- Recheck cited appendices and cross-references before using “not located”.
- Label extraction confidence as confidence in finding and reading text, not confidence in the procurement outcome.
- End with a human-verification queue ordered by requirement or criterion ID, not by predicted severity.

Expected output

An issue register that distinguishes absence from ambiguity, file-access problems and rule-dependent questions. That distinction matters operationally because each category may require a different authorised response, and some processes restrict evaluator communications or clarification activity.

Verification checkpoint

Re-run searches using requirement IDs, defined terms and likely synonyms before accepting “not located”. Inspect appendices, embedded worksheets and proposal cross-references manually. Only the authorised procurement official should determine whether an issue may be clarified, waived, cured, scored as submitted or escalated, and any bidder communication must follow the controlling process.

Prompt 10: Detect internal contradictions and qualification conflicts

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Find potentially inconsistent statements within one authorised proposal and between that proposal and permitted clarification material, without deciding which statement governs or whether the conflict is material.

Required inputs

Provide the proposal, its schedules and appendices, permitted clarifications, and the approved hierarchy or precedence rules if they exist. Keep other bidders’ responses outside this analysis.

Copy-paste prompt

Identify potential internal contradictions, conflicting commitments and unresolved qualifications within this bidder’s authorised submission. Do not determine materiality, compliance, contractual interpretation or the governing statement.

Source boundary:
- proposal and appendices: [IDENTIFY FILES];
- pricing files only if evaluators are authorised to inspect them at this stage: [IDENTIFY OR STATE EXCLUDED];
- authorised clarification records: [IDENTIFY OR STATE NONE];
- approved document-precedence rules: [PASTE OR STATE NOT SUPPLIED].

Check for conflicts involving:
- scope and exclusions;
- delivery dates, milestones and implementation duration;
- staffing levels, named personnel and availability;
- technical specifications, integrations and dependencies;
- service levels, support hours and response commitments;
- security, privacy, hosting or data-location statements;
- subcontractors and delivery responsibility;
- acceptance assumptions;
- contract exceptions;
- quantities, units and totals;
- statements changed by an authorised clarification.

Return a conflict table containing:
1. conflict ID;
2. topic;
3. first statement with exact citation and concise extract;
4. second statement with exact citation and concise extract;
5. conflict type: direct contradiction, numerical mismatch, scope qualification, timing mismatch, responsibility ambiguity, cross-reference inconsistency, or possible conflict;
6. explanation of why both statements may not be simultaneously true;
7. applicable precedence rule, if supplied;
8. unresolved question for human review;
9. source-extraction warning;
10. status: confirmed textual mismatch, possible interpretive conflict, or resolved by supplied precedence rule.

Rules:
- Compare exact units, dates, versions, definitions and conditions.
- Do not call normal elaboration a contradiction.
- Do not resolve a conflict using assumptions or external facts.
- Do not decide that a contract exception is acceptable or unacceptable.
- Do not infer intent.
- Keep pricing-derived conflicts in a separate appendix if price separation is required.
- End with a list of statements requiring direct source verification.
- State that authorised reviewers determine significance and any permitted procedural response.

Expected output

A paired-citation conflict table. Each flag should show the two passages that created the concern, rather than offering an unsupported conclusion. “Possible interpretive conflict” is appropriate when two passages could coexist under an unstated condition.

Verification checkpoint

Confirm definitions, document precedence and the effect of addenda before treating a mismatch as unresolved. A later schedule may intentionally refine an earlier narrative, and a conditional statement may not contradict an unconditional-looking summary once its context is read. Procurement, technical and legal reviewers should address only the questions assigned to them under organisational policy.

Prompt 11: Prepare a technical-evaluation evidence packet

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Convert technical proposal material into a source-cited packet for subject-matter experts while keeping evaluation criteria, technical claims and human judgements distinct.

Required inputs

Include only approved technical criteria, required specifications, authorised proposal content, applicable addenda and any approved technical-evaluation instructions. Do not add generic industry practices, preferred architectures or evaluator preferences unless they appear in the approved criteria.

Copy-paste prompt

Prepare a technical-evaluation evidence packet for authorised subject-matter experts. Use only the supplied solicitation, rubric and proposal. Do not score, rank, select or declare technical acceptability.

Inputs:
- approved technical criteria and subcriteria: [IDENTIFY FILE];
- technical requirements and specifications: [IDENTIFY FILES];
- applicable addenda: [IDENTIFY FILES];
- authorised proposal and technical appendices: [IDENTIFY FILES];
- approved evaluator instructions: [IDENTIFY FILE].

For each technical criterion, produce:
A. criterion and requirement IDs;
B. controlling source citation;
C. bidder response citations;
D. bidder-stated approach;
E. bidder-stated deliverables and acceptance dependencies;
F. stated architecture, components, interfaces or operating model;
G. stated implementation assumptions and customer responsibilities;
H. stated standards, certifications or test evidence, clearly labelled as bidder assertions unless the evaluation record includes authorised verification;
I. exceptions, alternatives and dependencies;
J. quantitative claims with units, conditions and source locations;
K. missing, ambiguous or conflicting evidence references from the issue registers;
L. neutral verification questions for the evaluator.

Then create:
1. a requirement-to-component traceability table;
2. a dependency and customer-responsibility table;
3. a list of claimed tests, demonstrations, certifications or references and what each actually evidences;
4. a source-check queue for diagrams, scanned pages and complex technical tables.

Constraints:
- Do not introduce unstated technical preferences.
- Do not describe a solution as secure, compliant, scalable, resilient, proven or compatible unless reporting the bidder’s exact claim; label it “bidder-stated”.
- Do not infer successful integration from a product name or interface reference.
- Distinguish existing capability, configuration, custom development, roadmap commitment and third-party dependency.
- Do not use external product knowledge.
- Do not compare this proposal with another bidder.
- Conclude with: “Subject-matter experts must verify the cited evidence and apply the approved scoring standard.”

Expected output

A technical packet that gives evaluators a consistent route from each approved criterion to the bidder’s architecture, commitments, dependencies and evidence. The distinction between existing functionality, configuration, custom work, roadmap items and third-party dependencies prevents materially different delivery propositions from being collapsed into one generic capability statement.

Verification checkpoint

Assign the packet to evaluators with the authorised expertise and conflict clearances. Verify technical diagrams and tables against the original files, particularly where text extraction may omit labels, legends or relationships. Do not convert bidder-stated certifications, security controls, test results or compatibility claims into verified facts unless the approved process contains and permits the necessary verification evidence.

Prompt 12: Enforce price and non-price evaluation separation

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Create a procedural separation check that identifies possible price leakage into technical evaluation and non-price judgements appearing in the commercial workbook.

Required inputs

Supply the approved evaluation sequence, evaluator access rules, redacted or unredacted file sets as authorised, and the current technical and commercial artefacts. Do not upload pricing to a technical-evaluation workspace if the rules prohibit technical evaluators from seeing it.

Copy-paste prompt

Audit the supplied evaluation artefacts for separation between price and non-price evaluation. This is a procedural review aid, not a determination of compliance, bias, value for money or procurement validity.

Approved process:
- evaluation sequence: [PASTE];
- evaluator access rules: [PASTE];
- price-opening or release condition: [PASTE];
- permitted shared fields, if any: [PASTE];
- current technical artefacts: [IDENTIFY FILES];
- current commercial artefacts: [IDENTIFY FILES];
- redaction rules: [PASTE OR STATE NOT SUPPLIED].

Perform these checks:
1. identify prices, rates, totals, discounts, commercial rankings or price-derived commentary appearing in technical artefacts;
2. identify technical scores, qualitative judgements or non-price rankings appearing in commercial artefacts;
3. identify filenames, metadata, cover pages, appendices or cross-references that may expose restricted information;
4. identify calculations that combine price and non-price results before the approved stage;
5. identify evaluator notes that appear to use cost as an unstated technical criterion or technical preference as an unstated price adjustment;
6. identify missing process information that prevents a separation check.

Return:
- artefact name and version;
- location;
- information type;
- expected audience under the supplied rules;
- observed audience or placement;
- potential separation issue;
- exact supporting citation;
- recommended human control action, such as access review, authorised redaction, artefact replacement, audit-log review or procurement-official decision.

Constraints:
- Do not reveal restricted pricing in the narrative; use “[PRICE CONTENT PRESENT]” and cite its location.
- Do not calculate rankings or combined results.
- Do not recommend a bidder.
- Do not assume that separation is required where no rule has been supplied.
- Do not perform unauthorised redaction; specify what an authorised person should review.
- Keep findings provisional until the procurement official confirms the governing rule and access history.

Expected output

A separation-control register that can be reviewed without unnecessarily reproducing confidential price data. It should identify where restricted information appears, who was expected to access it and which human-led control may be needed.

Verification checkpoint

Check actual permissions, sharing history and authorised evaluator access rather than relying only on document contents. If price information has reached an unauthorised evaluator, follow the organisation’s incident, integrity and procurement procedures; do not ask ChatGPT to decide whether the process remains valid. Retain original files and authorised redacted derivatives according to records policy.

Prompt 13: Recalculate and reconcile commercial worksheets

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Verify arithmetic using the organisation-supplied formula, inputs, blank-handling rules and exceptions, then produce a reconciliation log for independent human checking.

Required inputs

Provide the approved pricing formula, authorised pricing workbook, evaluation rules, units, currencies, rounding conventions, treatment of blanks and zeroes, optional items, taxes, escalation, discounts and documented exceptions. If any rule is absent, require an unresolved-input flag instead of allowing an assumption.

Copy-paste prompt

Recalculate the supplied commercial worksheet as a verification exercise. Do not decide price acceptability, affordability, value, ranking, abnormally low pricing, award or contract interpretation.

Authoritative materials:
- bidder pricing workbook and version: [IDENTIFY FILE];
- approved pricing formula: [PASTE OR IDENTIFY FILE];
- quantities and evaluation scenario: [PASTE OR IDENTIFY FILE];
- currency and unit rules: [PASTE];
- rounding rule and stage: [PASTE];
- blank, zero, “not applicable” and missing-value handling: [PASTE];
- discount, tax, escalation, optional-item and exception rules: [PASTE];
- approved price-evaluation instructions: [IDENTIFY FILE].

Procedure:
1. List every supplied input used, with workbook sheet, row, column or cell citation.
2. List excluded inputs and the rule supporting each exclusion.
3. Restate the supplied formula exactly before calculating.
4. Identify missing, text-formatted, duplicated, inconsistent-unit or apparently hard-coded values.
5. Calculate line extensions and subtotals using the supplied rules.
6. Apply discounts, taxes, escalation, options and rounding only where explicitly instructed.
7. Compare recalculated figures with bidder-submitted totals.
8. Report each variance as submitted amount, recalculated amount, absolute variance and formula-level explanation.
9. Produce an exceptions list for any calculation that required interpretation or could not be completed.
10. If data-analysis tools or Python are used, show the method or code and provide an input-to-output reconciliation table.

Constraints:
- Never replace a blank with zero unless the supplied rule requires it.
- Never infer a currency, quantity, period, unit, tax treatment or discount.
- Preserve full precision until the specified rounding stage.
- Keep one-time, recurring, optional and evaluated costs separate unless the approved formula combines them.
- Do not correct the bidder’s workbook or overwrite source values.
- Do not compare bidders or generate a price rank.
- Mark every result “provisional calculation for independent reconciliation”.
- End with a reviewer checklist covering formulas, inputs, units, blanks, exceptions, rounding and source cells.

Expected output

A transparent calculation workbook or table containing the source inputs, supplied formula, intermediate steps, provisional results, variances and unresolved exceptions. OpenAI documents that ChatGPT can perform some calculations with Python and expose methods, but reviewers must inspect the method, outputs and assumptions rather than treating tool use as validation.

Verification checkpoint

Reconcile the result independently in the approved spreadsheet or financial system. Check source cells, units, quantities, time periods, currency, formula order, precision and rounding. Pay particular attention to blank cells, hidden rows, merged cells, formula overrides and complex tables because extraction or interpretation errors can propagate through otherwise correct arithmetic. Only authorised officials may determine how an arithmetic discrepancy is handled or communicated under the solicitation rules.

Prompts 14–19: score provisionally, calibrate evaluators and prepare controlled clarifications

These six prompts convert the source-cited evidence assembled earlier into review materials for authorised evaluators. They do not determine responsiveness, technical acceptability, responsibility, value, conflicts of interest or the award decision. Every score remains provisional until the evaluator applies the approved rubric, verifies the cited proposal passages and records the reasoning required by the organisation’s evaluation rules.

Keep price and non-price material separate wherever the solicitation or governing procedure requires that separation. If ChatGPT performs arithmetic, provide the approved formula, weights, rounding rule, treatment of blanks and exceptions; then independently reconcile the result. OpenAI documents that ChatGPT can create tables and perform some calculations with Python, but reviewers still need to inspect the method, inputs, assumptions and outputs. Extraction from scanned, image-heavy or structurally complex files can also be incomplete, so a page citation is a retrieval aid rather than proof that the underlying passage was captured correctly.

For enterprise procurement considerations for Claude Code and the GPT-5.4 family, see enterprise AI coding-agent procurement comparison.

Prompt 14: Generate criterion-by-criterion provisional scoring worksheets

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Prepare a separate scoring worksheet for each proposal by mapping verified evidence to the approved scoring descriptors. The model may suggest a provisional score or score range only as a review aid. It must not rank suppliers, select a preferred proposal or turn an evidence summary into a final committee judgement.

Required inputs

Run this prompt after the requirement catalogue, criterion dictionary, mandatory-requirement worksheet and criterion-level evidence matrix have been reviewed for the proposal concerned. Do not run it against draft solicitation language, superseded addenda, unapproved weights or proposal versions that are not in the source register.

Copy-paste prompt

You are preparing a provisional evaluator worksheet for one submitted proposal. This is evidence organisation and rubric application support, not an award decision.

Use only:
1. the final approved solicitation and incorporated addenda;
2. the approved evaluation rubric, criterion weights and scoring descriptors;
3. the authorised proposal version identified below;
4. the reviewed evidence matrix and correction log.

Proposal identifier: [PROPOSAL_ID]
Authorised proposal version: [FILE_NAME_AND_VERSION]
Rubric version: [RUBRIC_VERSION]
Criteria in scope: [CRITERION_IDS]
Criteria excluded from this pass: [EXCLUDED_CRITERIA]
Approved scoring scale: [SCALE]
Approved weighting rule: [WEIGHTING_RULE]
Approved rounding rule: [ROUNDING_RULE]
Blank or not-applicable treatment: [BLANK_RULE]
Separation rule for price and non-price material: [SEPARATION_RULE]

For each criterion:
- reproduce the criterion ID, approved wording, weight and score descriptors without rewriting their meaning;
- list only proposal evidence relevant to that criterion;
- cite every evidence item by authorised file name, page and section or table reference;
- distinguish an explicit statement from an inference;
- identify missing, ambiguous, conditional or internally conflicting evidence;
- compare the cited evidence with each applicable rubric descriptor;
- suggest either a provisional score or, where the evidence does not clearly support one descriptor, a provisional score range;
- explain the suggestion in no more than three source-cited sentences;
- state what an evaluator must verify before adopting or changing it.

Do not:
- introduce criteria, subcriteria, weights, thresholds or external facts;
- infer supplier capability from reputation, incumbency, marketing language or information outside the authorised source pack;
- resolve a mandatory-requirement failure;
- combine price and non-price analysis if the supplied rule separates them;
- rank proposals or recommend an award.

Return a table with these columns:
Criterion ID | Approved criterion | Weight | Cited evidence | Evidence status | Rubric comparison | Provisional score or range | Provisional weighted result | Evaluator verification | Evaluator-entered final score | Evaluator rationale

Leave the final-score and evaluator-rationale columns blank. After the table, provide a calculation ledger showing every supplied input, formula step, rounding operation, blank treatment and exception. Mark all model-generated scores and calculations “PROVISIONAL — HUMAN APPROVAL REQUIRED”.

Expected output

The result should be a proposal-specific worksheet rather than a cross-supplier ranking table. Each provisional score must be traceable to both an approved descriptor and cited proposal evidence. A calculation ledger should make it possible for a reviewer to reproduce every weighted result without relying on the model’s arithmetic.

Verification checkpoint

The assigned evaluator checks the cited pages in the original files, confirms that no criterion has been silently reinterpreted and independently recalculates weighted values. The evaluator then enters the final individual score and rationale under the applicable rules. A low provisional score must not be treated as a determination that a proposal is non-responsive, unacceptable or excludable.

Prompt 15: Attach confidence and uncertainty annotations to provisional findings

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Separate confidence in evidence retrieval from confidence in a scoring suggestion. A clearly located sentence can still be ambiguous under the rubric, while a plausible interpretation can rest on poor extraction from a scanned table. This prompt makes those different uncertainty sources visible without presenting model confidence as a statistical probability.

Required inputs

Apply this prompt to the provisional worksheet before evaluators compare notes. It is especially useful where proposals contain cross-references, conditional commitments, diagrams, image-based schedules, dense appendices or qualifications that may alter an apparently direct answer.

Copy-paste prompt

Review the provisional scoring worksheet for [PROPOSAL_ID] and annotate uncertainty. Do not change any score, calculate a probability of correctness or use confidence as an unstated scoring factor.

For every criterion, assess these dimensions separately:
A. Source-location confidence: whether the cited passage can be located in the authorised file and appears to support the quoted or paraphrased text.
B. Extraction confidence: whether formatting, scanning, images, merged cells, footnotes or document structure may have caused content loss or misreading.
C. Interpretation confidence: whether the relationship between the evidence and the approved rubric descriptor is direct or requires judgement.
D. Completeness confidence: whether all identified proposal sections relevant to the criterion have been checked.
E. Calculation confidence: whether the result follows the supplied formula, inputs, blank rule, exceptions and rounding rule.

Use only these labels:
- High: direct, clearly located and independently checkable from the cited source;
- Medium: usable but dependent on interpretation, cross-reference or a manageable extraction limitation;
- Low: missing, visually complex, ambiguous, contradictory, conditional or not reproducible from the supplied material;
- Not applicable: the dimension does not apply.

For every Medium or Low label:
- give the specific reason;
- cite the affected source location where one exists;
- state the required human verification step;
- identify whether the uncertainty could affect evidence sufficiency, descriptor matching or arithmetic.

Do not average the labels. Do not convert them into points, percentages or a supplier risk rating. Do not imply that High confidence proves the claim is true, compliant or acceptable.

Return:
1. an uncertainty table with Criterion ID | Source location | Extraction | Interpretation | Completeness | Calculation | Reason | Human verification;
2. a short list of criteria that should not be finalised until the specified check is completed;
3. a statement confirming that confidence labels have not altered the approved score or weight.

Expected output

The annotation should expose why a finding needs attention. “Low” must be accompanied by an operational reason such as an unreadable table, unresolved cross-reference, conflicting statement or missing formula input. It must not become a hidden penalty or an unsupported numerical confidence score.

Verification checkpoint

Evaluators inspect every medium- and low-confidence item against the source file. Where exact wording affects a score, they should review the original page or native document rather than relying on extracted text. Any correction should be entered in the correction log and propagated to the evidence matrix and provisional worksheet before consensus activity begins.

Prompt 16: Run cross-document conflict and evaluator-conflict checks

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Produce two strictly separated registers: one for conflicts within the authorised evaluation materials, and one for potential evaluator conflicts or process restrictions disclosed in approved records. The model only flags conditions for authorised review; it cannot decide that a legal or policy conflict exists.

Required inputs

Run the documentary check after provisional worksheets exist and before evaluator materials are circulated for consensus. Run the evaluator check only if the authorised procurement lead has supplied the relevant declarations, assignments and policy rules and has approved their use in the workspace.

Copy-paste prompt

Perform two separate conflict checks. Do not merge them and do not make a legal, ethics, recusal, responsiveness or exclusion determination.

PART A — SOURCE AND PROPOSAL CONFLICTS
Compare only the authorised solicitation, addenda, rubric, evaluation instructions, proposal and reviewed evidence matrix.

Flag:
- a proposal statement contradicted by another proposal statement;
- a qualification, assumption or exception that may modify an answer elsewhere;
- inconsistent quantities, dates, deliverables, staffing, dependencies or calculation inputs;
- a source citation that does not support the worksheet summary;
- apparent tension between solicitation instructions and the supplied rubric;
- conflicting addendum treatment or unresolved version references.

For each flag, quote or closely paraphrase both sides, cite both locations, explain the narrow inconsistency and state “authorised human determination required”. Do not decide which statement controls unless an approved source expressly resolves it.

PART B — EVALUATOR OR PROCESS CONFLICT FLAGS
Use only the supplied declarations, assignment register and approved conflict policy:
[AUTHORISED_DECLARATION_SET]
[ASSIGNMENT_REGISTER]
[APPROVED_POLICY_RULES]

Identify only:
- an explicit disclosure matching a supplied policy trigger;
- an incomplete or internally inconsistent declaration;
- a prohibited assignment expressly identified by the supplied rule;
- a missing approval or disposition field required by that rule.

Do not investigate people or suppliers, search outside the source pack, infer relationships, label misconduct or recommend a recusal outcome.

Return two tables.

Table A:
Flag ID | Proposal or source | Issue type | Source 1 citation | Source 2 citation | Neutral description | Potentially affected criterion | Human owner | Disposition

Table B:
Flag ID | Evaluator or role identifier | Supplied policy trigger | Authorised record citation | Missing information | Authorised decision-maker | Status

Leave disposition and status blank. End with a list of materials you could not assess and why.

Expected output

Two review queues should emerge without conflating proposal inconsistency with evaluator impartiality. Neutral descriptions should identify the source pair and affected criterion while avoiding conclusions such as “material deviation”, “conflicted evaluator” or “disqualifying exception”.

Verification checkpoint

The procurement official or other authorised decision-maker resolves source precedence, conflict handling, evaluator assignments and any communication required by policy. Potential evaluator conflicts should not be shared more widely than necessary. Record the official disposition in the organisation’s controlled system rather than treating the chat output as the authoritative register.

Prompt 17: Create an anonymised evaluator-calibration exercise

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Test whether evaluators apply an approved scoring scale consistently before discussing live proposal scores. Calibration should align interpretation of the rubric, not force identical judgements or expose one evaluator’s live scoring to another prematurely.

Required inputs

Use an authorised training example, synthetic passage or formally approved historical extract. Do not use a live proposal excerpt if evaluation rules require independent scoring or prohibit pre-consensus discussion. The procurement lead must decide what material may be used.

Copy-paste prompt

Create an evaluator-calibration exercise using only the approved rubric and the authorised calibration material below.

Rubric version: [RUBRIC_VERSION]
Criteria in scope: [CRITERION_IDS]
Authorised calibration material: [CALIBRATION_MATERIAL]
Permitted source type: [SYNTHETIC_OR_APPROVED_HISTORICAL]
Discussion restrictions: [RULES]
Facilitator: [ROLE]

Prepare:
1. a neutral evidence excerpt with supplier and evaluator identifiers removed where required;
2. an individual worksheet asking each evaluator to identify evidence, select the applicable descriptor, propose a score and write a source-based rationale;
3. a facilitator key that maps the excerpt to the approved descriptors without declaring one mandatory score unless an authorised answer key supplies it;
4. discussion questions focused on descriptor interpretation, evidence sufficiency, treatment of ambiguity and prohibited unstated criteria;
5. a variance table for recording evaluator scores and reasons.

Do not:
- use live proposal scores unless expressly authorised;
- invent an “expert” answer;
- change rubric language to make it easier;
- average scores as a substitute for discussion;
- identify dissent as an evaluator performance problem;
- recommend changing weights or criteria during evaluation.

If a numeric variance calculation is requested, use this supplied method only:
[APPROVED_VARIANCE_METHOD]
Show inputs and workings. Otherwise, list differences descriptively.

End with a calibration record containing:
Criterion | Interpretation issue | Agreed application note | Rubric text supporting note | Unresolved issue | Authorised owner

State that an application note may clarify consistent use but cannot amend the solicitation or rubric.

Expected output

The exercise should reveal differences such as whether evaluators are scoring the presence of a claim or the strength of supporting evidence. It should also distinguish legitimate judgement from accidental use of outside knowledge, unapproved subcriteria or inconsistent treatment of missing material.

Verification checkpoint

The facilitator verifies that the exercise is permitted and that any agreed application note stays within the existing rubric. If discussion reveals a genuine defect or ambiguity in the approved evaluation scheme, escalate it under the governing procedure rather than asking the model or evaluators to repair the rubric informally.

Prompt 18: Draft a structured consensus-meeting agenda

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Turn verified individual worksheets, uncertainty notes and unresolved flags into a controlled agenda. The agenda should help the committee discuss evidence and scoring differences without using averaging, majority pressure or model-generated recommendations as substitutes for the approved consensus method.

Required inputs

Use only after required independent evaluations are complete and the procurement lead confirms that consensus discussion may begin. If price must remain sealed or separately evaluated, exclude it completely from this agenda and identify the later authorised stage.

Copy-paste prompt

Draft a consensus-evaluation meeting agenda from the authorised individual worksheets and issue registers.

Inputs:
[APPROVED_MEETING_RULES]
[PARTICIPANT_ROLES]
[CRITERIA_IN_SCOPE]
[INDIVIDUAL_WORKSHEETS]
[UNCERTAINTY_REGISTER]
[SOURCE_CONFLICT_REGISTER]
[CLARIFICATION_STATUS]
[PRICE_SEPARATION_RULE]
[REQUIRED_RECORD_FIELDS]

For each criterion:
- show each evaluator’s entered score without calculating an average unless the approved method expressly requires one;
- summarise the source-cited rationale for material differences;
- identify unresolved extraction, interpretation or contradiction issues;
- list the exact source pages the committee should inspect;
- provide neutral discussion questions tied to rubric descriptors;
- leave consensus score and consensus rationale blank.

Order the agenda according to the approved meeting rules. If no ordering rule is supplied, preserve criterion order and do not prioritise suppliers.

Include:
1. opening checks for attendance, authority, declarations and confidentiality;
2. confirmation that authorised source versions are being used;
3. unresolved procedural issues requiring the chair’s direction;
4. criterion-by-criterion discussion blocks;
5. a separate deferred-items register for matters outside the rubric;
6. a decision log for corrections, approved clarifications and deferred items;
7. closing checks for complete rationales, signatures or approvals required by policy.

Do not recommend compromise scores, preferred suppliers, eliminations or an award. Do not treat score convergence as the objective. The objective is a documented application of the approved rubric to verified evidence.

Return an agenda and a blank consensus record with:
Criterion ID | Individual scores | Evidence citations reviewed | Discussion summary | Consensus score | Consensus rationale | Dissent or unresolved matter | Decision owner | Approval status

Expected output

The agenda should direct attention to material interpretive differences and unresolved evidence rather than merely highlighting the largest numerical spread. It should preserve a place for dissent or unresolved matters if the organisation’s rules require them to be documented.

Verification checkpoint

The chair confirms meeting sequence, participants, conflict dispositions and the permitted consensus method. Only authorised committee members enter consensus scores and rationales. The official meeting record must be retained through the organisation’s approved records process; a ChatGPT Project does not by itself establish access authority, a procurement record or a retention schedule.

Prompt 19: Draft neutral clarification questions for human approval

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Convert unresolved, material ambiguities into neutral draft questions. ChatGPT must not contact suppliers, decide whether clarification is legally or procedurally permissible, request an improved offer or draft language that gives one supplier an unequal opportunity.

Required inputs

Use after reviewers have confirmed that the answer is not already available in the authorised proposal and when the procurement official has supplied the governing clarification rules. Keep this distinct from negotiation, discussion, proposal revision, best and final offer procedures or cure processes unless the applicable rules expressly define and authorise those steps.

Copy-paste prompt

Draft candidate clarification questions for review by the authorised procurement official. Do not send, address or simulate sending any communication to a supplier.

Use only:
[AUTHORISED_PROPOSAL]
[VERIFIED_EVIDENCE_MATRIX]
[UNRESOLVED_ISSUE_REGISTER]
[APPROVED_CLARIFICATION_RULES]
[COMMUNICATION_TEMPLATE_IF_ANY]
[DEADLINE_AND_CHANNEL_RULES_IF_APPROVED]

For each unresolved issue:
1. identify the exact proposal passage, page and criterion;
2. state why the existing text is ambiguous, incomplete or internally inconsistent;
3. state whether the requested fact appears necessary to apply an existing requirement or rubric descriptor;
4. draft one narrow, neutral question that asks the supplier to identify or explain existing submitted information;
5. identify any risk that the question could invite a proposal change, new evidence, negotiation, unequal treatment or disclosure of another proposal;
6. suggest a narrower alternative where that risk exists.

Do not:
- ask the supplier to improve, replace or supplement its offer unless the supplied rule expressly permits it;
- disclose another supplier’s content, score, standing or approach;
- reveal evaluator identities or deliberations;
- coach the supplier towards a higher score;
- assert non-compliance, default, misrepresentation or disqualification;
- create a deadline, channel, authority or consequence not present in the supplied rules;
- contact the supplier or produce an email marked ready to send.

Return:
Question ID | Proposal citation | Criterion | Verified ambiguity | Materiality to existing criterion | Draft neutral question | Proposal-change risk | Equal-treatment check | Rule citation | Procurement-official decision | Final approved wording

Leave the last two columns blank. Group materially equivalent issues so the procurement official can check whether comparable questions should be considered consistently across proposals. End with a list of issues for which no safe clarification question can be drafted under the supplied rules.

Expected output

Each draft should be anchored to an existing proposal passage and approved criterion. The question should seek interpretation or location of submitted information, not encourage a supplier to repair a weakness. The equal-treatment field should expose comparable issues without assuming that identical wording is always required.

Verification checkpoint

The authorised procurement official decides whether a clarification is permitted, necessary and appropriately framed; determines whether comparable treatment is required; approves the final wording, recipient, channel and deadline; and controls the communication record. No supplier contact should occur from ChatGPT or an automated workflow. Responses must be added to the source register, versioned and evaluated only under the approved rules.

Prompts 20–25: document consensus, residual risk and the authorised decision record

The final six prompts convert reviewed evaluator material into controlled records for human approval. They do not resolve disagreements, determine supplier eligibility, select a preferred proposal or authorise an award. The evaluation committee and authorised procurement official must verify every source citation, score, exception, risk treatment and decision statement against the approved solicitation, addenda, evaluation rubric and applicable procedure.

For organising conversations, files and instructions in ChatGPT Projects, see ChatGPT Projects Feature Complete Guide: How to Organize Conversations, Files, and Custom Instructions for Complex Workflows.

Prompt 20: Produce the human-approved consensus evaluation record

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Convert the committee’s recorded discussion and approved consensus outcomes into a criterion-by-criterion record. Use this only after individual evaluators have completed their reviews and the committee has followed the prescribed consensus procedure.

Required inputs

Provide the final approved rubric, criterion definitions and weights; the proposal identifier; individual evaluator worksheets; the meeting agenda or minutes; the chair’s record of agreed scores and reasons; approved corrections; and the authoritative source register. If price must remain separate, omit commercial material and state that restriction explicitly.

Copy-paste prompt

You are preparing a draft consensus evaluation record for human review. You do not decide scores, average scores unless the approved rules explicitly require it, resolve disagreements, determine responsiveness or recommend an award.

Use only:
1. the final approved solicitation, addenda, rubric and evaluation rules;
2. the authorised proposal version;
3. the evaluators’ completed worksheets;
4. the chair’s approved record of the consensus meeting; and
5. the supplied source register and correction log.

Prepare one row for each approved criterion with these columns:
- criterion ID and exact criterion title;
- approved weight or maximum points;
- final human-approved consensus score, copied exactly from the chair’s record;
- concise consensus rationale;
- proposal evidence citations, including document name, section and page where available;
- solicitation or rubric citation governing the assessment;
- material strengths accepted by the committee;
- material weaknesses or limitations accepted by the committee;
- unresolved evidence issue, if any;
- approved clarification response relied upon, if any;
- difference from individual evaluator scores, described without implying that a minority view was invalid;
- verification status: verified, citation needs checking, or decision record incomplete.

Do not calculate a consensus score from individual scores unless the supplied rules and chair’s record expressly instruct that calculation. Do not add a strength, weakness, criterion, subcriterion, weight or external fact. Do not convert an evaluator comment into committee consensus unless the chair’s record confirms adoption.

After the table, produce:
A. a reconciliation list of any mismatch between the approved consensus score and other supplied records;
B. a list of citations that require manual page-level verification;
C. a list of incomplete decision-record fields;
D. a short approval block with placeholders for committee chair, authorised procurement official, approval date and record version.

If evidence conflicts, quote or closely paraphrase both relevant passages with citations and label the matter “human resolution required”. If a field is unavailable, write “not recorded”; do not infer it.

Expected output

A consensus table followed by reconciliation, citation-check and approval sections. The record should distinguish proposal evidence from the committee’s judgement and should make absent documentation visible rather than filling gaps with plausible language.

Verification checkpoint

The chair should compare every consensus score and rationale with the meeting record. Evaluators should confirm that adopted strengths and weaknesses accurately represent the discussion. The procurement official should verify that the required consensus method was followed and that price information has not entered a non-price record where separation is required.

Prompt 21: Preserve a minority or dissenting evaluator view

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Draft a neutral minority-view annex where the governing procedure permits or requires one. The aim is to preserve a material disagreement without allowing the model to adjudicate it, discredit an evaluator or rewrite a dissent as consensus.

Required inputs

Supply the relevant criterion, approved consensus record, evaluator’s submitted minority statement or meeting notes, cited proposal passages, applicable procedure for recording dissent, and any approved redactions. Do not ask ChatGPT to infer which evaluator dissented from score patterns.

Copy-paste prompt

Draft a minority-view annex for human approval. Preserve the evaluator’s documented position accurately and neutrally. Do not decide whether the consensus or minority view is correct, reasonable, compliant or better supported.

Use only the supplied records. Structure the annex as follows:
1. proposal identifier;
2. criterion ID and title;
3. committee’s human-approved consensus score and concise rationale, copied from the approved record;
4. minority score or position, only if explicitly recorded;
5. minority rationale, retaining the evaluator’s material reasoning;
6. evidence cited for the minority view, with document, section and page references;
7. evidence cited for the consensus view, with document, section and page references;
8. exact point of disagreement;
9. factual matters agreed by both positions;
10. unresolved interpretation or judgement issue;
11. procedural disposition recorded by the chair;
12. confirmation that the authorised consensus result remains unchanged unless the official record says otherwise.

Use neutral attribution such as “the minority view records” and “the consensus record states”. Do not use loaded descriptions such as unreasonable, mistaken, biased, obstructive or unsupported unless that exact characterisation is part of an authorised finding and may lawfully be repeated.

Do not identify an evaluator unless the supplied procedure requires identification and the input expressly authorises it. Do not invent a procedural disposition. Where meeting notes and the evaluator’s written statement differ, show the discrepancy and label it for human reconciliation.

Finish with:
- source citations needing verification;
- omissions in the supplied record;
- a sign-off block appropriate to the supplied procedure.

Expected output

A self-contained annex that places the consensus and minority positions side by side, identifies their precise evidential or interpretive difference and records the procedural disposition without attempting to settle the disagreement.

Verification checkpoint

The dissenting evaluator should be offered the review or acknowledgement process required by policy. The chair should confirm that the annex does not alter the consensus outcome. An authorised procurement or legal reviewer should determine whether names, deliberative material or other information must be included, withheld or handled differently; ChatGPT cannot make that legal or records determination.

Prompt 22: Build the post-evaluation risk and issue register

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Consolidate unresolved evaluation issues into a register for human ownership. This prompt records matters such as unverified citations, inconsistent proposal statements, calculation exceptions and pending approvals; it must not transform signals into final findings about responsibility, compliance, conflicts or supplier risk.

Required inputs

Provide the missing-evidence register, contradiction log, calculation reconciliation, clarification log, consensus record, minority-view annex if applicable, source-check notes and the organisation’s approved risk categories or issue taxonomy. Exclude external bidder research unless the solicitation and governing procedure expressly authorise it.

Copy-paste prompt

Create a post-evaluation risk and issue register from the supplied authorised records. Treat each entry as a review item, not a final legal, compliance, responsibility, security, conflict, debarment or award determination.

For each item, include:
- unique issue ID;
- proposal identifier;
- related criterion or mandatory requirement;
- issue category selected only from the supplied taxonomy;
- factual description;
- originating record;
- proposal citation;
- solicitation, addendum, rubric or rule citation;
- current status;
- potential evaluation effect stated conditionally;
- existing control or completed action;
- required next human action;
- authorised owner or “owner not assigned”;
- target date, only if supplied;
- escalation route, only if supplied;
- decision or closure evidence required;
- residual issue after the proposed action;
- confidentiality or handling note, only where supplied;
- last-updated date and version.

Separate the register into:
A. open items that could affect a recorded evaluation result;
B. administrative or citation-quality items;
C. closed items with documented closure evidence;
D. matters outside the evaluation team’s authority.

Do not create probability or impact ratings unless the approved method and definitions are supplied. Do not label a bidder high risk, non-responsive, non-responsible, conflicted, debarred or non-compliant. Do not recommend exclusion or award. Do not infer deadlines, owners or closure from silence.

Deduplicate entries only when they concern the same underlying issue and preserve all original record references. If two records describe the same matter differently, retain the discrepancy in a “record conflict” field.

Conclude with a control summary listing:
- open items without owners;
- open items without source citations;
- issues whose stated resolution would change a score or rationale;
- issues outside committee authority;
- records that require an updated version after closure.

Expected output

A sortable issue register with explicit ownership, evidence and closure requirements. Conditional phrasing matters: “could affect criterion C3 if confirmed” is appropriate; “supplier fails C3” is not, unless that is already an authorised and documented human decision.

Verification checkpoint

The chair and procurement official should classify each issue under the approved taxonomy, assign owners and determine whether it can be closed, escalated or carried into later due diligence. Specialists should review security, financial, legal, privacy or export-control matters within their authority. Any score change must return through the approved evaluation process rather than being silently applied in the register.

Prompt 23: Design a controlled reference-check plan

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Prepare a consistent reference-check script and evidence form after the procurement official confirms that reference checks are authorised, correctly timed and within scope. The prompt does not contact references, investigate bidders or decide how reference evidence changes an evaluation.

Required inputs

Supply the solicitation’s reference-check provisions, approved timing and scope, authorised proposal claims to verify, reference contact details only where approved for use, standard questions, prohibited topics, evaluator roles, documentation requirements and rules for applying findings. Personal data should be minimised and handled under organisational policy.

Copy-paste prompt

Prepare a controlled reference-check plan using only the supplied solicitation provisions, approved procedure and proposal claims. Do not add external research, infer contact details, draft deceptive outreach or expand the check beyond the authorised scope.

Create five sections.

Section 1 — Authority and scope
- governing solicitation clause and procedure citation;
- proposals or stage to which checks apply;
- authorised checker;
- approved contact method;
- permitted subjects;
- prohibited subjects;
- required notices or consent, copied only from supplied materials;
- rule for documenting unsuccessful contact attempts.

Section 2 — Claim-to-question map
For each proposal claim approved for verification, list:
- claim ID;
- exact or faithful proposal statement;
- proposal citation;
- related criterion, if authorised;
- neutral verification question;
- acceptable follow-up prompts;
- prohibited inference;
- rule governing use of the response.

Section 3 — Standard interview script
Draft the same core introduction and questions for all similarly situated references. Questions must seek factual examples, dates, scope and the referee’s basis of knowledge. Avoid leading questions and requests for confidential, protected or unrelated personal information.

Section 4 — Evidence capture form
Include:
- date and method;
- checker;
- referee identity and stated relationship, using only authorised fields;
- question asked;
- response recorded accurately;
- direct observation versus second-hand information;
- documents offered;
- follow-up required;
- disputed or ambiguous point;
- checker notes clearly separated from the reference’s statements;
- acknowledgement or verification method required by policy.

Section 5 — Human disposition table
Include:
- finding requiring no action;
- finding requiring clarification;
- finding potentially relevant to an approved criterion;
- matter outside authorised scope;
- escalation owner;
- decision record required.

Do not score responses, rank proposals, recommend a supplier or state that a claim is false merely because a reference cannot confirm it. If the rules for using reference evidence are missing, identify that gap and stop before proposing an evaluative treatment.

Expected output

A uniform plan connecting each permitted question to a proposal claim and governing provision. It should separate what the reference said, what the checker observed and what an authorised evaluator later decides.

Verification checkpoint

Procurement counsel or the designated policy owner should confirm authority, timing, consent, permitted topics and treatment of results. The checker should use the approved script consistently and create a contemporaneous record. Any adverse, inconsistent or unverified information should be handled through the prescribed clarification and evaluation process, not treated as an automatic score reduction or exclusion.

Prompt 24: Draft the award memorandum without making the award decision

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Assemble a source-cited award memorandum draft from decisions already made by authorised humans. This prompt must not identify a winner from raw scores, resolve open issues, determine best value or supply missing justifications.

Required inputs

Provide the approved memorandum template; solicitation and addenda; procurement chronology; final consensus records; approved commercial evaluation; final calculation reconciliation; clarification and reference-check records; conflict-management records permitted for inclusion; approval history; and the authorised selection decision. If no selection has been approved, request a neutral evaluation-summary draft instead.

Copy-paste prompt

Draft an award memorandum from the supplied approved records. The authorised selection decision has already been made by humans. Do not independently compare bidders, choose a supplier, determine best value, change a score or cure a missing justification.

Follow the supplied memorandum template. If no template is supplied, use these headings only where the governing procedure permits:
1. procurement identification and purpose;
2. authority and approved evaluation method;
3. solicitation and addendum chronology;
4. proposals received and evaluated, using only authorised facts;
5. evaluator appointment and conflict-management process;
6. mandatory-requirement process and approved outcomes;
7. non-price evaluation summary;
8. price or commercial evaluation summary, kept separate where required;
9. clarifications, references or due-diligence steps;
10. final approved scores or rankings;
11. source-based rationale for the authorised decision;
12. unresolved conditions, approvals or pre-award actions;
13. approval and signature block;
14. record index.

For every material statement, cite the originating record by document name, version, section and page or record ID where available. Distinguish:
- proposal assertions;
- evaluator findings;
- committee consensus;
- procurement-official decisions;
- calculations;
- pending conditions.

Use the exact approved score and price figures. Do not recompute them unless asked to perform a separate verification table using the supplied formula, inputs, blank-handling rules and exceptions. Never insert unstated criteria, weights, external reputation information or generalised claims about a bidder.

If the authorised rationale is incomplete, insert “[human justification required]” and list the missing supporting record. If records conflict, do not choose one silently; create a conflict note for resolution. If an open issue could affect the decision, display it prominently in a pre-approval exceptions section.

Finish with:
A. a claim-to-source verification table;
B. a figures-reconciliation table;
C. a list of unresolved placeholders;
D. a list of records referenced but not supplied;
E. a publication or disclosure review flag without making disclosure decisions.

Expected output

A draft memorandum whose narrative can be traced to approved records, plus verification tables that expose unsupported sentences, conflicting figures and unresolved placeholders. A fluent paragraph must never conceal a missing decision record.

Verification checkpoint

The committee chair should verify the evaluation account, the commercial lead should reconcile all figures, and the procurement official should confirm the decision and rationale. Legal, records, accessibility and disclosure reviewers should apply organisation-specific requirements. Only an authorised official may approve, sign, issue or publish the memorandum or award.

Prompt 25: Prepare the audit, records and retention handoff

Safety and verification contract: Use only authorised source material and the authorised proposal version. Do not fabricate facts, dates, versions, source identifiers, page references or citations. Preserve uncertainty explicitly. Protect confidential information and privacy, and redact personal data when the approved procedure requires it. A human procurement reviewer or human evaluator must verify the cited evidence and retain every final decision.

Purpose

Create a handoff inventory for transferring working materials and approved records into the organisation’s controlled procurement repository. ChatGPT retention, Project storage or chat history must not be treated as the official records system.

Required inputs

Supply the approved file plan, retention schedule, records taxonomy, repository requirements, access roles, naming convention, version history, source register, final evaluation records, approval evidence, correction log and instructions for handling duplicates, drafts, exports and deletion. If these controls are unavailable, the output should be a gap list rather than an invented retention plan.

Copy-paste prompt

Prepare an audit, records and retention handoff manifest from the supplied procurement file and records instructions. Do not decide legal record status, retention periods, disclosure status, privilege, deletion authority or access entitlement.

Build a manifest with one row per record:
- record ID;
- record title;
- record category from the supplied taxonomy;
- procurement stage;
- authoritative or working-copy status, only if supplied;
- creator or responsible role;
- creation date;
- approved version;
- file format;
- source-system or current location;
- target repository and folder;
- related proposal or criterion;
- confidentiality or sensitivity marking supplied by the organisation;
- access group;
- retention class and trigger copied from the approved schedule;
- legal hold or disposal restriction, if supplied;
- approval or signature status;
- checksum or integrity field, if organisational procedure requires it;
- related records;
- superseded versions;
- correction-log reference;
- migration status;
- human verifier;
- verification date;
- disposal authorisation field.

Then produce:
1. a completeness check against the supplied file plan;
2. a version-conflict report;
3. an orphan-record list for files not linked to a source-register entry;
4. a missing-approval list;
5. an access-review list;
6. an export-quality check for truncated tables, missing pages, broken citations or unreadable scans;
7. a list of ChatGPT chats, Project files or generated drafts that require an explicit human disposition under policy;
8. a handoff certificate template for the authorised records owner.

Do not recommend deleting a chat, file, proposal or draft unless the supplied schedule and authorised instruction expressly require that action. Do not assume that downloading a file, moving it to a repository or removing it from a Project satisfies retention, deletion or legal-hold obligations.

Where the official record and a generated draft differ, preserve the distinction and identify the approved authoritative version. Label any uncertain classification “records-owner decision required”.

Expected output

A record-level manifest and exception report suitable for verification by the procurement and records teams. It should identify what must be transferred, what remains incomplete and which working materials still need an authorised disposition.

Verification checkpoint

The records owner should assign classifications, retention triggers, holds and disposal instructions. The procurement official should confirm completeness and authoritative versions. Workspace administrators should address ChatGPT access and file disposition under approved policy, but should not assume product-level controls replace the organisation’s repository, audit trail or records obligations.

Practical deployment checklist

Use this checklist before turning the outputs of Prompts 20–25 into an approval packet. Each item is a human control rather than a claim that the model has validated the procurement.

• Freeze the source set: Confirm that the solicitation, all addenda, rubric, proposal versions, clarification responses and decision records are the final authorised versions. Record hashes or other integrity values only if organisational procedure requires them.
• Confirm the working environment: Verify the account, workspace, permissions, sharing settings, connected applications and organisational approval before uploading sensitive proposals. OpenAI states that listed business offerings do not use business data for training by default, but this statement must be tied to the actual offering and configuration; it must not be generalised to consumer accounts or third-party services.
• Apply least access: Limit project and repository access to authorised participants. Project sharing depends on plan and workspace settings and does not itself confer procurement-file authority.
• Separate proposal sets: Use unambiguous bidder identifiers and filenames. Check that evidence from one proposal has not entered another proposal’s worksheet or rationale.
• Maintain evaluation separation: Keep price and non-price records apart wherever the approved process requires it. Remove commercial information from calibration or technical packets when evaluators are not authorised to see it.
• Verify every citation: Open the cited source and inspect the relevant page or section. Give extra attention to scanned, image-heavy, poorly structured or visually complex PDFs because exact extraction can fail.
• Reconcile calculations independently: Check formulas, inputs, weights, rounding, blanks and exceptions outside the generated narrative. OpenAI documents Python-assisted data analysis, but reviewers still need to inspect the method, output and assumptions.
• Lock human decisions: Copy consensus scores, issue dispositions and award decisions only from authorised records. Do not let later drafting regenerate or reinterpret them.
• Resolve visible exceptions: Address citation gaps, record conflicts, missing approvals and open risk items before signature. If an issue remains open, state its owner and required disposition rather than hiding it in notes.
• Review sensitive statements: Route legal, financial, security, privacy, export-control, conflict, debarment and records matters to authorised specialists. A model-generated flag is not a determination.
• Preserve version history: Assign a version and date to each consensus record, annex, register, memorandum and handoff manifest. Record approved corrections without overwriting the history required by policy.
• Export and inspect: Check exported tables for clipped rows, changed pagination, broken citations, lost footnotes and inaccessible formatting before filing.
• Complete the repository handoff: Transfer authoritative records under the approved file plan and obtain human confirmation. Do not rely on a ChatGPT Project as the sole archive.

Failure modes that require correction or escalation

Failure mode	Why it matters	Required response
A consensus score is calculated from individual scores without authority.	The result may not reflect the approved consensus method or committee decision.	Remove the calculated value and copy the score from the chair’s authorised record. Escalate any missing decision.
A minority view is rewritten as an error or omitted.	This can distort the deliberative record and conceal a material disagreement.	Restore neutral attribution and follow the procedure governing dissenting views.
A review flag becomes a categorical supplier finding.	Evidence inconsistencies do not independently establish non-responsiveness, non-responsibility, conflict, debarment or non-compliance.	Recast it as an issue requiring authorised human determination and preserve its supporting citations.
Reference questions introduce new evaluation criteria.	Unstated criteria can change the basis of evaluation.	Delete the question unless the governing documents authorise it; have procurement counsel or the policy owner review the script.
The award memorandum fills a rationale gap with fluent drafting.	Fluent text can conceal the absence of an approved decision or supporting record.	Replace the passage with a visible human-justification placeholder and obtain the missing approval.
Evidence from different proposals is cross-contaminated.	The resulting record may attribute another bidder’s claim, weakness or price to the wrong proposal.	Stop drafting, verify document identity and rebuild affected outputs from separated source sets.
A scan or complex table is treated as perfectly extracted.	Missing text, merged columns or incorrect values can materially affect evaluation and calculation.	Compare the output with the original pages and manually transcribe critical values under a second-person check where policy requires it.
Chat history is treated as the authoritative record.	Product storage and sharing do not establish the organisation’s official file, retention class or approval trail.	Export verified records to the controlled repository and apply the approved file plan.
An unresolved issue silently changes a score.	Score changes must follow the authorised evaluation and approval process.	Reopen the prescribed decision process, document the reason and preserve the prior version.
A privacy statement is applied to the wrong account or connected service.	OpenAI’s business-data statement is offering-specific and does not establish the practices of consumer accounts or third parties.	Verify the actual offering, workspace configuration and connected-service terms before processing material.

Frequently asked questions

Can ChatGPT combine evaluator scores into the final consensus score?

Only as a mechanical calculation when the approved rules explicitly prescribe that method and humans provide the formula, inputs, blank treatment, rounding and exceptions. Many processes use deliberative consensus rather than an average. In those cases, ChatGPT should copy the human-approved consensus score and flag any mismatch; it should not calculate a score itself.

Should a minority view appear in the award memorandum?

That depends on the organisation’s procedure and the authorised memorandum format. Use Prompt 21 to preserve the view in a controlled annex, then let the procurement, legal or records authority decide whether and how it is referenced. Do not omit or publish deliberative material based on model judgement.

Can the risk register label a proposal non-responsive?

No. The register can identify missing, ambiguous or conflicting evidence and link it to a mandatory requirement. A responsible human authority must make and document any responsiveness determination under the governing rules.

Can ChatGPT perform reference checks?

It can draft an authorised plan, consistent question set and evidence form from supplied rules. Human officials should approve the scope, contact references and evaluate the responses. The model should not search for unofficial references, infer contact details or convert an inability to verify a claim into an adverse finding.

What if the award decision has not yet been made?

Do not use Prompt 24 to ask ChatGPT to identify a successful supplier. Request a neutral evaluation summary that reports approved results, open issues and required approvals without recommending a supplier. Draft the award memorandum only after the authorised decision and rationale exist in the record.

Can a ChatGPT Project serve as the procurement archive?

No. OpenAI’s Projects documentation describes Projects as workspaces for chats, files and instructions, and does not establish them as a procurement record or archive. Projects can organise chats, files and instructions, but availability and sharing depend on plan and workspace settings. The organisation must use its approved repository, access controls, retention schedule and record-ownership process.

How should teams handle calculations created through data analysis?

Require a reproducible calculation table showing the supplied formula, each input, treatment of blanks, rounding and exceptions. Then independently reconcile the output with the commercial worksheet and approved rules. OpenAI documents data analysis and calculations with Python, but that capability does not validate procurement-policy compliance or the correctness of extracted source values.

What should happen when a citation points to the wrong page?

Correct the citation through the controlled correction process, record the changed version and check whether the error affects the rationale or score. If it does, return the matter to the authorised evaluator or committee rather than silently editing the conclusion.

May the team upload confidential proposals?

Only after confirming organisational authorisation, the actual account and workspace, permissions, retention controls, connected applications and sensitivity restrictions. Proposals may contain personal, export-controlled, security-sensitive or competitively sensitive information. OpenAI’s business privacy statement applies to the listed business offerings and should not be generalised beyond the relevant offering and configuration.

Useful Links

• OpenAI Help Center: Projects in ChatGPT
• OpenAI Help Center: File Uploads FAQ
• OpenAI Help Center: Data analysis with ChatGPT
• OpenAI: Enterprise privacy
• OpenAI Academy: Using projects in ChatGPT