Guides

AI-Made Political Ads: The Counts Keep Growing, and Most Carry No AI Label. What Creators and Agencies Should Take From It

5 November 2026 · RenderBob team

Wesleyan Media Project counts of AI-made US political ads rose from 33 in May to 220 by 11 October. A 3 November report cites nearly 300. Most of the counted ads carried no AI disclaimer.

Researchers counting AI-made US political ads have reported larger totals each time: 33 ads in May, 164 by early September and 220 by 11 October. CNBC on 3 November cites nearly 300. Most of the ads in the counts that give a disclaimer rate carried no AI label.

The counts so far

• May. Wesleyan's 4 September release says its first report found 33 ads and about $20 million in spending.
• Early September. That release puts the total at a minimum of 164 ads and spending just under $80 million. CT Mirror quotes co-director Erika Franklin Fowler saying almost 170 ads and at least $80 million, and gives the report's figure as 164. It does not reconcile her words with the report. The release says 69% of the ads and 78% of the spending had no disclaimer. CT Mirror rounds the ad share to almost 70%.
• 11 October. Fortune, on 21 October, reports more than $137 million on 220 videos made at least partly with AI. More than 70% had no disclosure. Co-director Travis Ridout is quoted. The project found less than $50,000 on such ads in the 2024 campaigns.
• 3 November. CNBC says a Wesleyan analysis released the previous week found more than $187 million spent on nearly 300 ads generated or enhanced by AI. It says the total is almost certainly an undercount, and that most of those ads had no disclaimer. CNBC does not print the exact ad count.

How to read the numbers

• Definitions differ. Wesleyan's September release says it counts ads with an AI disclaimer, ads reported as AI-made, and ads its coders flagged and then confirmed. Fortune does not restate how "at least partly with AI" was defined for the October count.
• Periods differ. Each report uses a different cut-off date.
• Counts are likely low. Fowler calls the September figures an undercount because little AI use is disclosed. Wesleyan's release and CNBC say the same about the later tallies.
• Who is spending. Wesleyan's September release says Republican candidates or pro-Republican groups accounted for 80% of the ads and 83% of the spending. Fortune's October count is 81% of the ads and 88% of the spending. CNBC says more than 90% of the spending in the later tally came from Republican candidates, a Republican party committee, or a Republican-leaning group. Ridout, in Fortune, attributes part of the gap to differing attitudes toward AI.
• Content varies. CT Mirror says the ads ranged from satire to hyperrealistic deepfakes, used in attacks and to flatter candidates.

The rules, as reported

CT Mirror says no federal law requires disclosure of AI use in political ads. Some states require disclaimers on AI-generated political content and some ban election deepfakes outright. CNBC says there is no comprehensive federal law, and that about 30 state laws set different requirements. We did not check the statutes or the platform rules, and they differ.

What creators and agencies can take from it

The same questions apply to any AI-assisted video you deliver, political or not.

• Decide your labelling rule before the job, and write it into the brief.
• Check the rules where it will run: the state or country, and the platform.
• Keep a record of which shots used generation, which model, which references and which prompt.
• Do not depict real people in generated scenes without written permission.
• Mark generated shots in the edit so a label can be added or a shot replaced quickly.
• If most AI use goes unlabelled, a clearly labelled piece with a record behind it is easier to defend.

What to watch

The Wesleyan Media Project's final totals, any state or platform changes after the election, whether disclosure rates improve, and how researchers define "AI-made" as tools blend into ordinary editing.

RenderBob 3 and this story

RenderBob 3 puts generated clips on their own timeline lane and keeps references and prompt logs with the studio, so a team can show which shots were generated, with which model, and label or replace them.