Synthetic video tools put truth, and the next election cycle, under more pressure

By Jared Schroeder

An American service member in her home. A law enforcement official in his car. A bespectacled, gray-haired man sitting in a park. Each of these people, along with several others from all walks of life, took to TikTok earlier this spring to ask Americans to show their support for the Republican Party.

Their requests were compelling. Instead of simply reading text on a social media platform, the viewer could see and connect with the individual sharing the message. The people who shot and posted these videos could be a relative, neighbor, or coworker. They looked like someone we know or could know. Video, unlike print, has always had a powerful seeing-is-believing quality. 

The only catch regarding the TikTok posts was that none of these people set their phones to record and then posted their ideas. All of the videos were synthetic, meaning the people, settings, and message were generated by AI. The everyday Americans were characters imagined by an AI tool, rather than fellow citizens seeking to share ideas. The videos, which TikTok has since taken down, represent the next challenge for democracy. We are fast approaching the first AI-slop-infused election cycle.

AI-generated testimonials and other types of videos are going to influence voters’ opinions in the midterm elections this year. Our free expression system is ill-equipped to navigate these types of believable, nonhuman political videos. 

This might seem like de ja vu. After all, bots and algorithms have already influenced U.S. elections. While U.S. elections remain vulnerable to these decade-old types of social-media manipulation, the rise in quality and accessibility of synthetic media creation tools has put truth, and each citizen’s ability to identify it, under even more pressure.

For more than a decade, people have recommended improving news and technology literacy to help citizens parse through the growing amounts of misleading online content. This approach seems unlikely to succeed as an inoculation against misleading or false AI-generated political videos. The content is too believable, and the human brain is wired to accept video depictions as reality. The unique nature of video, and how people receive it, make it different from other types of AI-generated content. Improved literacy can help, but it’s not the solution.

That leaves two other avenues: government regulation and more aggressive corporate limitations on the use and spread of AI-generated political videos on their spaces. 

Social media platforms generally have guidelines of AI-generated content. TikTok requires realistic synthetic video content be labeled. Meta, which owns Instagram and Facebook, has increased its efforts to label AI content. These efforts by those who own the spaces that generally host AI-generated video content are helpful but are often unevenly enforced. The TikTok testimonials seeking support for the Republican Party garnered tens of thousands of followers before they were taken down. Social media firms have given us little evidence that they will halt their products from harming the democratic process.

Tik Tok and Meta have guidelines of AI-generated content.

Regulation is also a fraught solution. There is no federal law limiting political AI video content. Nearly two-thirds of states have political deepfake and synthetic-media laws, but there are concerns about their constitutionality. The Supreme Court has long established that political speech receives the highest level of First Amendment protections. So far, that’s meant these laws have struggled in the courts.

A federal court struck down a California law in 2025 that limited “deceptive audio or visual media of a candidate within 60 days of an election” after the parody site The Babylon Bee argued it limited its First Amendment rights. The Babylon Bee succeeded again in Hawaii this year, when another federal court struck down the state’s limits on deceptive election-based content.

The decisions, along with the limited effectiveness of literacy and corporate control efforts, leave the U.S. democratic process at a crossroads. As it stands, American voters will be misled by AI-generated video content this fall and in future elections. These videos can be created by foreign actors seeking to undermine U.S. elections, special interest groups, or a candidate’s opponent in the election. 

At the same time, government efforts to limit the threat conflict with how the First Amendment has been interpreted. Political candidates, their supporters, and special interest groups have long received expansive latitude to make iffy claims.

What is crucial is that AI-generated political videos are fundamentally false representations of reality and, because of the nature of video, inherently more believable than other types of media. Any regulatory solution must protect human-created political expression, like the types of claims human communicators have long made and the parody work at the center of the California and Hawaii cases, while still limiting intentionally false and misleading AI-generated video. 

One solution would be to classify AI-generated video differently than other forms of expression. Government regulations have long placed certain requirements, including required information about the origins of the message, on political advertainments. A similar approach would be appropriate for AI-generated videos.

Thus, legislation must focus on the nature of the media. With careful definition, lawmakers should construct legislation that requires labels on AI-generated video during a 60-day period before an election. These requirements must specifically exempt human-authored news, commentary, and parody content and instead focus on AI-generated political videos that make false and misleading claims that are meant to be believed as truthful by the audience. A labeling requirement, rather than regulating the content of the videos, ensures ideas are still shared and the government does not act as the arbiter of what is true.

There is no easy solution to the wave of AI-generated political videos that will appear online during this election cycle. Experimenting with narrowly tailored labeling regulation might present the best option.

———————————————————————-

Jared Schroeder is an associate professor at the University of Missouri School of Journalism and a member of the Overby Center panel of experts. He is the author of “The Structure of Ideas: Mapping a New Theory of Freedom of Expression in the AI Era.”