Deepfake Democracy: UVU Study Finds AI Videos Sway Voters Like Real Ones

OREM, UT – June 24, 2026 – A groundbreaking study from Utah Valley University has delivered a stark warning for the future of democratic elections: AI-generated deepfake videos are just as effective as authentic content at influencing voter opinion. The research, conducted by the university's Center for National Security Studies (CNSS) and the Gary R. Herbert Institute for Public Policy, also found that Americans from all demographics are profoundly unable to distinguish synthetic media from reality, raising urgent questions about election security and public trust.

Amid growing concerns over the role of artificial intelligence in political discourse, this new report moves the threat from theoretical to tangible. The findings suggest that the tools to manipulate public opinion on a mass scale are not only effective but also widely accessible, and that current public defenses, like media literacy education, are falling short.

The Unsettling Power of Persuasion

The study's core experiment presented a politically representative sample of 632 U.S. voters with a fictional ballot initiative. After an initial vote, participants were shown one of four videos—two real and two AI-generated, arguing for or against the initiative—before being asked to vote again. The results were alarming: there was no statistically significant difference between the persuasive power of the real videos and the AI deepfakes.

In fact, the synthetic content was sometimes even more effective. While about 50% of viewers who watched a real speaker changed their opinion, that number rose to 55% for those who viewed a synthetic video. Furthermore, biometric data gathered by UVU's Neuromarketing SMART Lab, which measures unconscious responses, showed that deepfake videos received equal or higher credibility scores than their authentic counterparts on measures of trustworthiness and persuasiveness. This suggests that on a subconscious level, the AI-generated content was not only believable but highly compelling.

“The findings of this study present a clear and urgent challenge for policymakers, election security professionals, and the public: AI-generated deepfakes are already capable of influencing political opinion as effectively as authentic media,” said Brandon Amacher, Director of UVU's Emerging Tech Policy Lab and an instructor at the CNSS.

A Crisis of Perception: The Failure to Detect

Perhaps more striking than the persuasive power of the deepfakes was the public’s inability to identify them. When asked to determine if the video they watched was real or synthetic, participants failed overwhelmingly. A staggering 60% of individuals who viewed a deepfake video incorrectly identified it as authentic. Only 16% correctly identified the AI-generated content, a sharp decline from a similar study conducted just two years prior, indicating that the technology’s sophistication is rapidly outpacing the public's detection abilities.

This inability to detect fake content cut across all lines. Democrats, Republicans, and Independents all performed at similarly low rates, with correct identification ranging from just 15–19 percent. No age group or demographic showed any meaningful advantage in spotting the fakes. The study also dismantled a common assumption in the fight against disinformation: that awareness of deepfakes builds immunity.

“Equally concerning is the finding that familiarity with deepfakes provides no meaningful protection,” Amacher noted. “This undermines a common assumption embedded in current media literacy policy – that educating the public about the existence and nature of deepfakes will reduce their impact.” Participants who reported being "extremely familiar" with deepfake technology performed no better than those who had never heard of it. Troublingly, individuals who were most confident in their ability to spot a fake were three times more likely to be wrong.

Underscoring the accessibility of this threat, researchers confirmed that all synthetic media used in the study was created using publicly available, low-cost software on a student-owned laptop over a single weekend. This proves that creating convincing, politically influential deepfakes is no longer the exclusive domain of state actors but is now available to almost anyone.

Beyond the Ballot Box: The Erosion of Public Trust

While the implications for election integrity are immediate, the researchers at Utah Valley University warn that the problem extends far beyond the ballot box. The widespread proliferation of convincing synthetic media threatens to erode public trust in institutions, the media, and reality itself. If any video or audio clip can be flawlessly faked, it becomes easier for citizens to dismiss inconvenient truths as disinformation and for malicious actors to sow chaos and division.

This creates a paradox where the public becomes simultaneously more gullible to fakes and more cynical about authentic information. The study highlights a future where national security is threatened not just by external attacks, but by an internal collapse of shared truth. When leaders, experts, and journalists can be convincingly impersonated, the foundations of informed public discourse and policymaking begin to crumble.

This challenge requires a response that goes beyond simply telling people to be more vigilant. The study’s authors argue that relying on the human eye and ear is a failed strategy. Instead, the focus must shift toward systemic, technology-driven solutions that can verify and authenticate content at its source.

Forging a Path Forward with New Solutions

In response to their own sobering findings, the experts at Utah Valley University are not just sounding the alarm but are actively pioneering solutions. They advocate for a multi-layered defense system that includes provenance labeling to track a file's origin, cryptographic watermarking to invisibly tag content as real or synthetic, and robust digital identity frameworks.

Putting these ideas into practice, the Gary R. Herbert Institute and the CNSS have launched a pilot program in partnership with a digital identity startup. This initiative offers Utah's political candidates a free service to authenticate their digital identity, creating a verified chain of custody for their campaign communications. By allowing candidates to certify their own content, the system aims to give voters a reliable way to distinguish official messages from fraudulent deepfakes. UVU is serving as a third-party evaluator for the project, gathering critical data to refine these solutions for broader application.

This proactive, solutions-oriented approach demonstrates a commitment to tackling the challenges of the digital age head-on. The university's research makes it clear that the era of AI-driven disinformation is here, but its work also lights a path forward, emphasizing technological safeguards and new policy frameworks to protect the integrity of our information ecosystem.