Campaigns spend $137 million on AI ads as voters struggle to spot fakes

The rise of synthetic ads

American political campaigns and PACs are rapidly adopting artificial intelligence to produce advertisements. According to a report by the Wesleyan Media Project, these groups have already spent more than $137 million on AI-generated content during the current midterm cycle.

While some of this spending was low in previous years, the current volume shows a significant shift in strategy. These ads range from simple cosmetic changes, like animating a still photo of a candidate to smile, to complex deepfakes where a candidate appears to say things they never actually uttered.

One specific example involves a video of Texas Senate candidate James Talarico. In the clip, he appears to read old tweets and offer commentary that he never actually spoke, a production created by the Senate Republicans' campaign arm. This represents a move toward high-fidelity fabrication in political messaging.

How voters perceive the fakes

Research suggests that the average viewer is poorly equipped to spot these synthetic messages. A study by Utah Valley University found that voters correctly identified deepfakes only 16% of the time. Interestingly, people who claimed to be extremely familiar with deepfakes performed no better than those who had no familiarity at all.

The issue is compounded by how these ads are labeled. In some cases, the 'AI GENERATED' tag is small and faint, making it easy to miss. In other instances, ads carry no labels whatsoever, such as a PAC ad in Oklahoma that featured a candidate next to Hillary Clinton with no disclaimer.

When labels do appear, they are often confusing. NYU’s Center on Technology Policy found that only 21% of people identified AI as the technology involved when an ad used the phrase 'manipulated by technical means.' Furthermore, studies show that these labels can actually decrease trust in the candidate, which may discourage campaigns from using them.

The cost of political slop

For political operatives, AI offers a massive reduction in production costs. A veteran campaign ad maker noted that a scene which previously would have cost $2 million to film can now be generated for approximately $200,000. This makes it much easier for smaller groups to flood the airwaves with content.

Experts categorize these new ads into different types. Some are 'cosmetic' animations, while others are described as 'slopaganda'—content designed to stir feelings rather than make a logical argument. These can include absurd scenarios, such as a candidate pitching a 'political delusion pill' or a celebrity appearing as a superhero.

The most concerning category is content built to be believed as fact. These deepfakes can be used to manufacture scandals or fake endorsements. Because the technology is becoming more accessible, the risk of these deceptive ads appearing on television and in social media feeds is expected to increase significantly.

Proposed rules and next steps

Current laws are struggling to keep up with the pace of technological change. While at least 31 states have deepfake election statutes, the actual rate of labeling is nearly identical in states with laws versus those without. This suggests that current regulations are not effectively forcing transparency.

One proposed solution is to move away from defining 'deceptive' technology and instead focus on the person shown. Under this litmus test, if a real person did not agree to be shown in a specific way, the ad would require a plain-language label in both the audio and on-screen text. This would be easier for station managers to verify quickly before an ad airs.

With the 2028 presidential election approaching, there is a push for federal standards to replace the current patchwork of state laws. A recent poll shows that a vast majority of both Democrats and Republicans want rules on AI in politics, as many voters report they have already been fooled by AI-made content during the current cycle.