- Published
- Reading time
- 3 minutes
- Filed under
- product · brand-defense · claims
KnitKnot field note
Introducing Evidence Packets for disputed AI claims
A bad AI answer is easy to screenshot. Evidence Packets preserve the answer, the outside and company pages used to verify a false claim, and when each was collected.

Kevin Kho
Co-founder, KnitKnot
Evidence plate / productObserved · sourced · reviewedA screenshot of a bad ChatGPT answer gets a team’s attention. It rarely settles the next question: what can we prove?
The answer may cite several pages. One could be a competitor’s site. That citation alone does not tell you which page supplied the false statement, or what was on the page when ChatGPT answered. Your own page may contradict the claim today, but it may have changed since the answer was recorded.
We built Evidence Packets for that handoff. From a verified misrepresentation Issue in KnitKnot, you can prepare a private case file with the saved AI answer, the outside page that supports the disputed statement, the company page that contradicts it, and a record of when each was collected.
From an Issue to a case file
Open a verified misrepresentation Issue and select a saved ChatGPT answer. KnitKnot shows the disputed quote and the two page quotes before you prepare the packet. If the evidence passes the check, Prepare evidence packet captures the pages and assembles a downloadable ZIP.
The ZIP has a readable PDF index, the full saved question and answer, the retained page text used to verify the quotes, new captures of both pages, and a manifest with file hashes. The files stay private to the workspace. A new packet creates a new version; it does not rewrite an earlier one.
This is a record of a KnitKnot benchmark answer. It is not a capture of a buyer’s personal ChatGPT session.
Why a citation isn’t enough
Consider an answer that says your product lacks a feature and cites a competitor comparison page. The citation is a lead. It does not prove that the competitor wrote the false sentence or caused ChatGPT to repeat it.
For an Evidence Packet, we require a tighter chain. The exact disputed words must be present in the saved answer. An outside page must itself support that statement, and a company page must explicitly contradict it. The claim has to refer to the same product, capability, plan, region, and relevant time. If those pieces do not line up, KnitKnot holds the packet and shows why.
We also keep the dates separate. The packet records when ChatGPT answered, when we first retained each page’s text, when the proof was checked, and when the pages were captured again for the packet. A fresh capture cannot show what ChatGPT saw earlier. If a required quote has disappeared from a page, we preserve the historical receipt and hold the packet rather than pass off the current page as historical proof.
For your team and counsel
Evidence Packets organize the material for your team and counsel. They do not decide whether anyone acted unlawfully, draft a cease-and-desist letter, or contact a publisher. A reviewer can open the original records, inspect the quotes and timestamps, and decide what the evidence supports.
If you have a verified misrepresentation Issue, open it in KnitKnot and select Prepare evidence packet. If the button is unavailable, the Issue shows which part of the proof is missing. For the wider workflow, see how KnitKnot traces claims back to their sources.