News
220 AI-Made Political Ads and $137 Million: What the Wesleyan Count Says About Disclosure
ยท RenderBob team
Gizmodo, citing the Wesleyan Media Project, reported 220 AI-made or AI-enhanced midterm ads and more than $137 million in spending. It says 27% included a disclaimer.
On 16 October, Gizmodo reported Wesleyan Media Project figures of 220 AI-generated or AI-enhanced ads in the 2026 US midterms, with spending above $137 million, across 34 states. Gizmodo says 27% of those ads included a disclaimer. An AI Weekly summary of a New York Times story the same day rounded the share without a disclaimer to about 70%.
Two counts, two documents
Wesleyan's own release of 4 September, which we read, counted a minimum of 164 ads and spending just under $80 million. It said 69% of those ads had no AI disclaimer. Wesleyan called the total an undercount. The method was news mentions, a search for ads that carried a disclaimer, and the project's own tracking of broadcast, CTV, and digital ads. There is no federal rule that requires a disclaimer. The 220-ad figure is Gizmodo's account of later Wesleyan data. We have not read a newer Wesleyan release, or the Times story, so the later totals stay reported rather than checked against the paper.
Who paid, as reported
Gizmodo says Republican candidates and groups accounted for 81% of the identified ads and 88% of the spending. The September release put 80% of that earlier set with Republican sponsors or aligned groups, and 11% with pro-Democratic sponsors. The labelling rate is the part that transfers to film work.
What carries over
- Labels are inconsistent. On Gizmodo's figure, most of the counted ads had none. Our 16 October note on audience surveys cited Hub's finding that 72% of respondents wanted AI use disclosed every time.
- Catching undeclared AI and asking creators to disclose it are different jobs. YouTube's move towards automatic labels, which we covered, tries to catch what creators do not declare. Provenance that travels with the file, as in our Article 50 note, lasts longer when platforms keep it.
- Real people placed in scenes they did not perform sit at the top of the risk ladder in our documentary piece. Gizmodo says nearly half of the identified ads depicted real politicians in synthetic scenes.
For your own work
Disclose at the point of publication, name what was generated, and keep the prompt log and the reference list. If a project touches an election, check the platform rules and the election law that applies before you release it.
What to watch
A Wesleyan write-up of how the 220 ads were coded, and whether the share that carries a disclaimer moves before election day.
More from the blog
- Kling 4.0 Launched on 15 October, with HDR, Extension, and the API Still Pending
Kuaishou says Kling 4.0 launched on 15 October, with clips up to 30 seconds and 15 references. The user guide still marks 10-bit HDR and two-minute extension as coming soon.
- Artificial Lands at Neon After Amazon MGM Drops It: What the Move Means for a Single Buyer
Amazon MGM dropped Luca Guadagnino's Artificial, and Neon acquired it in late June. Screen Daily has reported a 25 December US release. Amazon's statement does not explain the drop.