--- site: agentic-rfp-software.ai slug: /reponse-tracable langue: en title: Why an AI-produced tender answer must cite its source, and how to verify it before signing. description: Traceable means that, for every claim submitted, you can trace back to the document it rests on, its version, who drafted it, who verified it, with what score, and who signed it. What this changes for the signatory, the buyer, the auditor, and for you in six months. rubrique: Understand intention: Info requete_type: factual derniere_revue: 2026-09-05 statut: draft, pending review --- # Why an AI answer must cite its source, and how to verify it An AI-produced tender answer must link every claim to the document it rests on: the document, the page, the version, who drafted it, who verified it, who signed it. ## What is a traceable answer? A request (a tender, a security questionnaire, a due diligence questionnaire, the DDQ a customer sends before awarding you a contract) receives dozens or hundreds of answers. Each one asserts something about your company: a certification, a control, a timeline, an organization, a commitment. An answer is traceable when every one of these claims is linked, with no intermediary, to five things. 1. **The document** where the claim finds its proof: a policy, a procedure, a certificate, an audit report, an already-validated answer. 2. **The page** of that document, because a long policy proves nothing until you know which part to read. 3. **The version** of the document, because a policy changes, and a claim true under the previous version can be false under the current one. 4. **The confidence score** assigned to the answer by an independent check, along with the discrepancies found between what is claimed and what is proven. 5. **The validator**: the named person who approved the answer, and the date. To these five links is added an overall property: replayability. Six months later, you can reconstruct the entire process, from the file received to the answer submitted, with the same documents, the same versions, the same scores, the same signatures. An answer that lacks these links can still be accurate. It is not defensible, because no one can show why it is. ## What is traceability for, and for whom? Four people need to trace a claim back to its evidence, at different times and for different reasons. | Who | When | What they are looking for | What happens without traceability | |---|---|---|---| | **The signatory** | Before submission | Knowing what they are committing the company to, and being able to review what matters instead of everything | They sign blind, or they review everything, which amounts to the same thing once volume exceeds capacity | | **The buyer** | During evaluation | Verifying a claim in one click, comparing proven answers rather than asserted ones | They file the response among those they cannot verify, and compare on price | | **The auditor** | Months later | Reconstructing what was claimed, from what, and who validated it | The submitted answer has become an orphan claim: no one remembers where it came from | | **You yourself** | At the next request | Reusing a validated answer knowing whether its source is still in force | You reuse an answer whose source has changed, and answer falsely with a sentence that used to be true | The third case is the one that gets forgotten, and it is the costliest. An answer to a security questionnaire commits the company for the life of the relationship. When the customer's auditor, or a regulator, asks two years later what a given claim rested on, a traceable answer has the response ready; a non-traceable one forces you to reconstruct, from memory, what you believed at the time. ## What a precise source pins down in a traceable answer: the document, the page, the version Many tools display a source. Few display a page and a version. The difference is not a detail. A source with no page is a folder: "see the security policy." To verify it, you have to read the whole policy. No one does, not the signatory, not the buyer. The citation exists, verification does not happen. A source with no version is a source that is potentially stale. An access-management policy moves from version three to version four; the account-revocation deadline changes; the answer in the questionnaire, meanwhile, does not change, because it cites "the access-management policy" and that label is still true. The answer has become false with no one having changed anything. A traceable answer cites the page and the version because those are what make verification possible and staleness visible. When the cited version is no longer the one in force, the answer is flagged before it is reused. ## What happens to a claim with no source? This is the question that separates systems. Generic AI, faced with a question whose answer is not in the documents, produces a plausible-sounding sentence: that is what it is built to do. Professional AI tools, marketed as reliable, get it wrong in a measured share of cases, on a preregistered protocol (Magesh et al., *Journal of Empirical Legal Studies*, 2025), and the most dangerous error is not a visible absurdity, it is the plausible claim no one checks. A traceable answer cannot contain such a claim, by construction: a sentence with no document, no page, and no version is not an answer, it is a gap. It is declared as such, the expert who can supply the evidence is named, and the action to close it is recommended before submission. The submitted file then contains two things: proven answers, and an explicit list of what was closed or remains open. It is less flattering than a document with no holes. It is what a serious buyer prefers. On the Optivalue.ai platform, which publishes this site, every answer carries its document, page, and version, a confidence score from 0 to 100 produced by a second AI team independent of the first, and the named, traced signature of a validator; the process is exported with its context, ready to be audited. ## The concession that clarifies everything A draft does not need to be traceable. An internal note, a first version you will fully rework, a summary for yourself: generic AI does this fast and well, and demanding a page and a version for every sentence would be absurd. Traceability becomes necessary the moment text commits someone in front of someone else. A draft commits no one. Your response does. The question is never "should it be traced," but "from what point": the answer is "as soon as it goes out." ## Three questions that draw the line To find out whether an answer, or a tool, is genuinely traceable: 1. **Pick an answer at random: can you open the evidence that proves it, at the right page, in one move?** If the source is a file name, or a folder, traceability stops at the document's door. 2. **Is the cited version the one in force on the submission date, and would you know if it were not?** If nothing flags that a cited document has changed, you will one day reuse a stale answer. 3. **Can you replay the process in six months, and find who drafted, who verified, with what score, and who signed?** If the answer is "we have the final document," you have a text, not a record. ## Frequently asked questions **Does traceability slow down the response?** It slows down drafting a single sentence and speeds up everything else: review, which concentrates on low-scoring answers; the buyer's evaluation, which checks in one click; reuse, which knows whether the source still holds. Time shifts from downstream to upstream, and it shrinks. **Should even the obvious answers be traced?** Yes, because those are the ones no one checks. The head-office address, headcount, a certification: obvious, until the day one of them changes and the answer does not. **What happens when a source document is updated?** Every answer that cites it is flagged: the cited version is no longer the one in force. Each answer is then re-verified against the new version, or removed from the reference base. That is the mechanism that stops a true answer from silently becoming false. **Does traceability protect you in a dispute?** It lets you establish what was claimed, on what basis, on what date, and who validated it. That is what a judge or an auditor asks for. This site does not give legal advice, but it is not hard to see the difference between "we can show it" and "we remember it." **Is a traceable answer longer?** No. The links to the document, page, version, and score are not text added to the answer; they are properties attached to it. The reader sees a sentence, and can open its evidence. --- ## See traceable answers on your own documents Bring a real tender, a security questionnaire, or a DDQ. You will see the sources cited to the page and version, the score of each answer, and the gap analysis on your document, not a prepared demo. [Discover the Optivalue.ai platform](https://www.optivalue.ai/en/?utm_source=agentic-rfp-software.ai&utm_medium=satellite&utm_campaign=reponse-tracable) ยท [Book a demo](https://www.optivalue.ai/en/demo/?utm_source=agentic-rfp-software.ai&utm_medium=satellite&utm_campaign=reponse-tracable) Written by the compliance and presales team at Optivalue.ai. Last reviewed: 5 September 2026. [Markdown version](/en/reponse-tracable.md) ### Sources cited - V. Magesh, F. Surani, M. Dahl, M. Suzgun, C. D. Manning, D. E. Ho, "Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools," *Journal of Empirical Legal Studies*, vol. 22, 2025, p. 216-242. https://onlinelibrary.wiley.com/doi/full/10.1111/jels.12413 --- ## Further reading - [How artificial intelligence helps you respond to a tender or RFP](/en/) - [Why two companies using the same AI produce the same response](/en/une-ia-qui-n-apprend-que-de-vous) - [Are your tender responses protected when you use AI](/en/vos-reponses-sont-un-secret-des-affaires) - [How to choose tender-response software](/en/comment-evaluer-un-agent)