Research
Six public cases where a fabricated claim reached print, why a model states it with full confidence, and what a verification stage has to do.
8 of 31 sources cited45 claims checked, 5 corrected
All research
A detector scores how a text was written. A citation check lists what a reader cannot verify, and only one of them hands an editor work to do.
A generated draft arrives looking finished. Four failures hide inside it, and a better prompt cannot reach any of them.
Detection and watermarking claims usually cite a lab or a university. Here are the people behind them, one result each, with the limits they published.
One survey of 1,015 B2B marketers reports both. Put side by side, they show which part of the work teams dropped when the tools arrived.
Most advice about beating a detector targets vocabulary, which is why it does not work. What moves the score is structural, and so is the fix.
One article free on sign-up. No credit card. Agencies start with a pilot.