Evaluate a source before using it in an AI brief

A useful source is suitable for the claim you need to make. The same page can be strong evidence for a provider’s published price and weak evidence for whether its product is best. Evaluate the claim-source relationship rather than giving a whole website a permanent trustworthy label.

Ask what the source can know

A software vendor can describe its own plan. A regulator can publish requirements within its authority. A participant can describe their experience. Each has a different scope. Consider who produced the information, how they obtained it, and whether they have an incentive to present a particular conclusion.

A fictional evidence comparison

Do not turn “official” into “independently proven.” Primary sources are valuable because they are close to the facts, but their claims still have scope and incentives.

Use this evaluation worksheet

For this source, record author or organization, publication/update date,
claim being used, method or basis, geographic and population scope,
commercial interests, and information missing from the page.
Explain what the source supports and what it does not support.
Do not rank a source by design polish or writing confidence.
Source excerpt and URL: [paste]
Claim I need to establish: [sentence]

Follow the evidence chain

If an article cites a study, open the study rather than attributing the conclusion to the article alone. Check whether the cited work measured the outcome your sentence names. A study about intention is not automatically evidence of actual behavior.

When sources disagree, compare dates, definitions, methods, and populations. They may answer different questions. Record the disagreement instead of selecting the source with the more attractive result. The final brief should tell the reader which conclusions are secure, which are conditional, and what additional evidence would be useful.