Plain, mechanistic writing on how AI handles evidence — and how it should. Specific, not salesy; written for people who read methods sections for a living.
A running record of publicly documented cases where AI-fabricated citations reached published research, survived peer review, or entered a court filing. Every case found by someone else — a journal, a research team, a judge — and linked to its primary source.
Open the Case File →Five checks, fastest first — resolve the DOI, search PubMed and Crossref, match author, journal and year, and open the source. Plus the three kinds of fake citation to watch for.
Read the guide →A walk through exactly how a language model produces a flawless author list, journal and DOI for a paper that was never written — and why formatting is the easy part to fake.
Read the article →A citation can be correct in draft one and wrong by draft three. We define drift precisely and show how it creeps in as you sharpen your prose.
Read the article →A case report and a randomized trial are not the same evidence. Why grading belongs on every citation an AI hands you — not buried in the methods.
Read the article →Join the waitlist and we'll send new pieces along with your early-access invite. No spam.
Join the waitlist