CORE GUIDE
Fact-check and revision
AI-assisted long-form work becomes trustworthy when claims are extracted, checked against authoritative sources and revised with provenance instead of asking the same model to 'fact-check itself'.
Mental model
Fact-checking is a claim–evidence workflow: identify verifiable claims, attach appropriate sources, classify support or conflict, then revise the manuscript while preserving a trace of why the statement changed.
Why it matters
Long documents mix factual claims, synthesis, opinion and illustrative language. A generic review prompt tends to miss subtle errors or confidently approve statements that resemble training data. Separating claims from prose creates a reviewable unit and lets the author focus expensive verification on consequential or uncertain assertions.
01
Turn prose into a claim queue
Extract claims with location references, classify which require external evidence, retrieve authoritative sources, and record support, contradiction or unresolved status. Revise only after the evidence review, keeping citation and source notes attached. Re-run consistency checks so a correction in one chapter does not leave the same outdated claim elsewhere.
02
Example: one outdated statistic appears in three chapters
A manuscript cites an adoption percentage copied from an old report and paraphrases it differently across chapters. Claim extraction groups the related statements, the reviewer finds a newer authoritative dataset, and revision updates all occurrences plus their interpretation. The evidence record shows why the number changed.
Common failure modes
- Asking the drafting model to certify its own claims without independent sources.
- Checking citations for existence while ignoring whether they support the exact claim.
- Correcting one occurrence and leaving inconsistent versions elsewhere in a long document.
Engineering heuristics
- Prioritize verification by claim consequence and uncertainty.
- Keep claim location, source and review status in a structured matrix.
- After factual edits, run cross-document consistency checks.
Takeaways
- 01Fact-checking operates on claims, not prose vibes.
- 02Independent sources matter more than model confidence.
- 03Revision should preserve provenance and propagate corrections consistently.
Reading evidence
This records actions you actually took; it does not claim mastery, proficiency, or certification.
Used in
This Concept is reused across these canonical learning paths.