Fact-checking
Checking every name, date, number, claim, quote, and citation against evidence before the piece goes out.
See it
What it is
Fact-checking verifies the parts of a draft that can be tested against evidence: names, titles, dates, numbers, claims, quotations, links, and citations. The checker works claim by claim, preferring primary sources, recording what supports each statement, and preserving the context around a quote or statistic. The New Yorker's checking department is the usual reference point for how far this can go: the checkers work through a piece word by word, calling sources back to confirm the details a writer already believes are settled.
Reach for it before publication, and earlier when a high-stakes claim shapes the whole piece. Mark every checkable assertion, trace it to the underlying report, record, dataset, transcript, or named expert, and confirm the source says what the draft says it says. Dates and versions matter because a once-correct number can expire.
Gotcha: a citation is not proof by decoration. A real source may not support the exact sentence beside it, and several articles can all repeat the same original error. Follow the chain to the earliest available evidence, distinguish verified fact from reasonable inference, and leave an unresolved claim unresolved rather than rounding uncertainty into confidence.
Ask AI for it
Fact-check this draft claim by claim. Build an evidence table with claim, verdict, primary source, exact supporting passage or data field, publication date, version, and correction. Verify scholarly citations against the Crossref REST API, including DOI, title, author, and publication year. Check quotations against the supplied transcript in context, and trace repeated statistics to their original dataset or report. Use only sources you can identify by URL or DOI. Label unsupported, outdated, disputed, and inferred claims separately; never invent a source or silently soften a failed claim.