��֭����

As artificial intelligence technologies make their way into political ads and campaigning, Americans are expressing growing concern. But they’re not just worried about deepfakes and deceptive content’s impact on elections —  they also fear how the government might use the fight against misinformation to restrict free speech.

In a recently released FIREpoll of registered American voters, conducted by Morning Consult, one concern stood out: government regulation itself. Nearly half of respondents (45%) said they are “extremely” or “very” concerned that government regulation of election-related AI content could be abused to suppress criticism of elected officials. That’s a powerful signal that while Americans see the risks posed by AI, they don’t trust government regulators to police political expression fairly.

When asked to choose between protecting free speech in politics or stopping deceptive content, a plurality (47%) said protecting free speech in politics is more important, even if that means allowing some deceptive content. Just 37% prioritized stopping deceptive content, even at the expense of limiting speech that would otherwise be protected by the First Amendment. These sentiments are held across the political spectrum, but are stronger among Independents and Republicans, than among Democrats.

This isn’t just a preference — it’s a principled stand in favor of the core freedoms the First Amendment exists to protect. Political speech lies at the heart of those freedoms, and Americans clearly recognize that any government attempts to police what can or can’t be said pose a far greater threat to democracy than free speech itself.

Regulation threatens participation

The chilling effects are already measurable. About 28% of voters said they’d be less likely to share content on social media if the government began regulating AI-generated or AI-altered content. (That’s right: All content, not just AI-generated or AI-altered content.) That may not sound dramatic at first glance, but that’s  than the average voter turnout during the last midterm primaries. As our political culture is increasingly shaped online, discouraging speech — even unintentionally — can have real consequences for public discourse.

These findings suggest a troubling trajectory: Government regulations justified in the name of protecting the public from AI could end up silencing the public instead. 

While some polls show that a similar percentage of voters (41%) say it’s important to protect people from misinformation, that concern cannot be used to justify censorship. About 39% said that preserving freedom of speech should be the government’s top priority when crafting AI laws. Only 12% said that view doesn’t describe them at all. In other words, most Americans believe that protecting speech isn’t just one goal among many — it’s the central concern.

And they’re right to think so. The First Amendment doesn’t permit the government to restrict speech simply because it believes the public might be misled. The solution to bad speech is more speech, not less.

These results should serve as a warning to policymakers: The public views efforts to regulate AI in political campaigns as a risk to free expression. FIREhas been actively engaged in legislative advocacy to safeguard First Amendment rights, including vague and overbroad bans or disclosure requirements imposed on AI content. 

If voters already believe regulation will be abused — and are already pulling back from political expression using AI — that’s not just a theoretical harm. It’s a chilling effect in action.

Instead of rushing to regulate, elected officials should reaffirm their commitment to protecting political speech, no matter the medium. The technology may be new, but the principle is not: In a free society, the government doesn’t get to decide which ideas are too dangerous to be heard.

The poll was conducted May 13-15, 2025, among a sample of registered voters in the US. A total of 2,005 interviews were conducted online across the US for a margin of error of plus or minus 2 percentage points. Frequency counts may not sum to 2,005 due to weighting and rounding. Topline results are available here.