The Real Cost of Deepfake Ads Is Not Deception. It Is Doubt.
The 2026 midterms were the first U.S. election in which AI-generated ads became routine. Most of them fooled no one. That is precisely the problem. When synthetic images become ordinary campaign material, the damage is not only that false things look real, but that real things become easy to deny.
Executive summary
- The moment: the 2026 U.S. midterms normalized AI-generated political advertising. By early September, the Wesleyan Media Project had identified at least 164 such ads, worth just under $80 million, and most carried no AI disclaimer.
- The common defense: these ads are obvious, labeled or satirical, so nobody is deceived.
- The argument: deception is the wrong measure. The deeper cost is epistemic. Every normalized fake weakens the public's ability to trust recordings, and that weakness benefits whoever needs a real piece of evidence to be dismissed.
- The lesson: leaders should judge synthetic media not only by whether it misleads, but by what it does to the credibility of evidence in general.
A season of synthetic campaigning
The numbers are still incomplete, but the direction is clear. In its September update, the Wesleyan Media Project (2026) reported "a minimum of 164 political ads generated or enhanced through the use of AI," up from 33 in May. According to the same report, 69 percent of those ads, and 78 percent of the spending behind them, carried no disclosure of AI use. The researchers stress that their count is an undercount, since there is no federal requirement to label AI in political advertising.
The examples are familiar by now: candidates placed in scenes they never attended, opponents dressed in costumes, rivals shown together although they never met. Al Jazeera described the trend in October as "blurring truth," and quoted Scott Babwah Brennen of New York University on "slopaganda": low-quality synthetic content that seeks to convey a general impression of a candidate rather than a factual claim (Roy, 2026).
The usual defense from campaigns is reasonable on its face. No one believes a senator actually wore a costume. The ad is labeled, or it is satire, and satire has a long and legitimate history in politics. If no one is deceived, where is the harm?
The wrong question
The question "Was anyone fooled?" assumes that the only ethical problem with synthetic media is false belief about a particular event. That assumption made sense for older forms of manipulation. It misses what is new here.
The philosopher Regina Rini (2020) has offered a more useful frame. Audio and video recordings, she argues, "function as an epistemic backstop." They do not only prove things after the fact. Their mere possibility disciplines what people say: politicians, witnesses and officials know that a recording might exist, so they have reasons to be careful with the truth. Recordings support the honesty of testimony even when no one checks them.
A deepfake ad that fools no one still erodes that backstop. Each time a fabricated scene circulates as normal campaign material, the public learns, a little more, that images and video are things campaigns make up. The lesson is not tied to a single ad. It accumulates.
The liar's dividend
Legal scholars Robert Chesney and Danielle Citron (2019) anticipated where that accumulation leads. They called it the "liar's dividend": as the public becomes aware that convincing fakes are possible, people caught on genuine recordings gain a new defense. They can simply claim the evidence is fake.
This is no longer only a hypothesis. In five survey experiments with more than 15,000 American adults, Kaylyn Jackson Schiff, Daniel Schiff and Natalia Bueno (2025) found that politicians who falsely claimed a real scandal was misinformation gained support across partisan groups, compared with alternatives such as staying silent or apologizing. The effect appeared for scandals reported in text. Against video evidence, the strategy was weak.
That last finding deserves attention. Today, video still works as a backstop. The authors' results suggest that the liar's dividend is limited precisely by the credibility of recordings. A campaign season that teaches voters to see video as routine fabrication is spending down that credibility. The dividend grows as the backstop shrinks.
Who pays, and who collects
There is an asymmetry here that campaigns rarely consider. The candidate who uses a satirical deepfake today may benefit from a few days of attention. The costs are distributed across everyone, and they arrive later: in the next scandal, the next leaked recording, the next crisis in which the public must decide whether to believe its own eyes.
The people who collect the liar's dividend are not necessarily the ones who produced the fakes. They are whoever, in the future, needs a genuine piece of evidence to be dismissed. In that sense, normalized synthetic advertising is a commons problem: a shared resource, public trust in recordings, is degraded by many small and individually defensible uses.
What responsible practice looks like
None of this requires banning satire or AI tools. It requires a higher standard than "nobody was fooled."
- Label clearly, and at the start. A disclosure in small print at the end of an ad protects the sponsor, not the public. Labels should be as visible as the image itself.
- Do not fabricate the real person in a realistic scene. Caricature and animation signal their artificiality. Photorealistic fabrication of an identifiable opponent does not.
- Never answer evidence with "it's fake" unless you can show it. For leaders facing a crisis, the temptation to claim misinformation is now backed by research suggesting it can work. It is also the clearest way to spend credibility that cannot be recovered.
- Protect your own archive. Organizations should keep verifiable originals of their statements and recordings, so that authenticity can be demonstrated when it matters.
Closing note
Campaigns tend to judge new tools by their immediate effect on voters. The ethics of synthetic media require a longer view. A deepfake that persuades no one can still teach millions of people that nothing they see can be trusted, and that lesson does not stay in the campaign that taught it. In a democracy, evidence is a public good. The 2026 midterms showed how cheaply it can be spent.
References
Chesney, R., & Citron, D. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. California Law Review, 107, 1753. https://www.californialawreview.org/print/deep-fakes-a-looming-challenge-for-privacy-democracy-and-national-security
Rini, R. (2020). Deepfakes and the epistemic backstop. Philosophers' Imprint, 20(24).
Roy, S. (2026, October 9). AI deepfake ads grow more popular in US midterm campaigns, blurring truth. Al Jazeera. https://www.aljazeera.com/economy/2026/10/9/ai-deepfake-ads-grow-more-popular-in-us-midterm-campaigns-blurring-truth
Schiff, K. J., Schiff, D. S., & Bueno, N. S. (2025). The liar's dividend: Can politicians claim misinformation to evade accountability? American Political Science Review, 119(1), 71–90. https://doi.org/10.1017/S0003055423001454
Wesleyan Media Project. (2026, September 4). At least $80 million spent on AI ads this cycle. https://mediaproject.wesleyan.edu/?p=56458
