Trace the Claim Home
Use this whenever a claim reaches you through a summary, screenshot, search result, social post, video, textbook, museum page, or generative AI answer. The job is not to distrust everything. The job is to reconstruct the evidence chain far enough that you know what supports the claim.
What exactly is being asserted? Rewrite it narrowly enough that evidence could confirm, complicate or contradict it.
Where did you encounter it? AI answer, website, video, textbook, article, post, conversation?
Does the carrier identify a person, document, dataset, archive, study, quotation or institution?
Can you open the cited or implied source itself? Does it really support the same claim?
What observations, records, testimony, measurements or documents does the upstream source rely on?
What genuinely independent evidence agrees, disagrees or changes the interpretation?
Claim-tracing worksheet
The claim I am testing:
Where I encountered it:
What evidence the carrier points to:
The furthest upstream source I actually opened:
What that source can establish:
What it cannot establish:
Independent evidence I checked:
My current confidence and why:
When the carrier is AI
AI can help you identify useful vocabulary, possible scholars, dates, institutions, interpretations and search directions. But an AI-generated sentence can be several abstraction layers away from the evidence. If the claim matters to your argument, do not stop at the answer. Move upstream.
A good final state is not “the AI was right”. It is “I can now show what evidence supports this claim, what its limits are, and where the interpretation comes from”.