Turn an AI research answer into a claim-and-evidence record
List consequential claims, attach original sources and decide which statements are verified, qualified or removed before publication.
In brief
A research answer can combine correct background with unsupported specific claims. A claim record makes those statements individually checkable before you use or publish the text.
It is especially useful for prices, release dates, supported versions and advice that changes a user’s action.
Make each statement small enough to check
Split “new, cheaper and available everywhere” into separate claims. They may have different evidence and dates. Prioritize claims affecting compatibility, cost, safety or access rather than recording every harmless connective phrase.
Keep a usable evidence table
The following structure is a blank working record, not completed research. Preserve the source passage or reference within suitable privacy and quotation limits.
| Claim | Original evidence | Checked date/scope | Decision |
|---|---|---|---|
| Exact statement | Document and supporting section | Date, product and region if relevant | Keep / qualify / correct / remove |
Check the final text again
Use only supported statements when rewriting. Confirm that the rewrite did not add a new number, causal claim or broader region. Publish a correction when a material prior claim was wrong, preserving the original publication date and a truthful update record.
Before you act
A source link is not enough unless it supports the specific statement. Record the date and scope of verification, and leave an honest unknown when evidence is missing.
What to do next
- Extract the factual claims that matter to the decision.
- Find the original evidence and note its date, passage and applicability.
- Keep, qualify, correct or remove each claim, then review the rewritten answer against that record.
