An AI won't tell you it doesn't know. It hands you a confident answer with a footnote, and the footnote is fake: plausible authors, a real journal, a DOI that goes nowhere. Lawyers have been sanctioned in open court for citing cases that never existed. A federal health report cited studies that were never written. Most people can't spot one, and we'll test that live, with an interactive audience game: AI-generated bibliographies go up on screen, you vote real or fake from your phone, and the room's tallies land on screen in real time before we check the answers.

Then the bigger problem. Citations are trust technology, a 350-year-old system for "don't take my word for it," and courts, medicine, policy, and science all run on it. Fabricating a source now costs nothing. Checking one still costs minutes. And the record was leaky before AI ever showed up: papers retracted years earlier were cited at least 44,046 times in 2024 alone.

You leave knowing the tells that give a fake citation away, a 30-second check that works on any source, and what it would take to make "cited" mean something again.

All levels. 30 min preferred, flexes longer. Audience participation needs only a phone camera and a browser (QR link, no app, no signup, own cell data); anyone without a phone still gets the full show, and the talk degrades gracefully if connectivity is poor. Companion submission: "Brain, or AI model?" (separate material).