AI political ads reached $80mn—and most did not disclose the tool
The Wesleyan Media Project has identified at least 164 political advertisements generated or enhanced with AI in the 2026 US election cycle, carrying just under $80mn in estimated spending. Eighty per cent of the ads—and 83% of the spending—came from Republican candidates or pro-Republican groups. Sixty-nine per cent lacked a disclaimer disclosing AI use.
Those figures are a floor, not a census. Researchers tracked reported uses, visible disclosures and material confirmed after coders flagged it; subtle editing may remain undetected. They also combine a broad range of practices, from obvious satire and inexpensive virtual sets to deepfakes that depict candidates saying or doing things they never did. That distinction matters because the democratic harm does not scale neatly with the production technique.
The sharpest case in the Financial Times’ reporting comes from Texas. Ads supporting Republican Senate candidate Ken Paxton show Democrat James Talarico appearing to recite old comments, operate a van offering children gender-transition procedures and act as a masked surgeon. Some underlying quotations are genuine; the scenes and delivery are synthetic. The device gives authentic fragments the authority of invented visual evidence.
More than 30 states regulate AI in political advertising or require disclosure, but constitutional challenges and uncertainty over federal races leave a patchwork. The market provides a counter-pressure: roughly three-quarters of respondents in a September industry survey considered AI-generated political imagery unacceptable, and disclosures may function as warnings that make voters distrust an ad.
The result is a revealing transitional equilibrium. AI has already reduced the cost of scenes that once required actors, locations and crews, but campaigns remain unsure whether its use persuades or repels. A useful disclosure rule should therefore identify synthetic speech or conduct in a way viewers can notice before absorbing the claim—not simply add small print after the production economics have already changed.