Imagine this: you open a chatbot to ask a harmless question — maybe “What are the pros and cons of Candidate A vs. Candidate B?” A few minutes later, you discover that your opinion toward one candidate has subtly shifted. That’s no longer a hypothetical: recent research shows that conversational artificial-intelligence tools can influence political opinions. This effect — sometimes stronger than that of campaign ads — raises important questions about how we consume information, especially in election seasons.
What the studies actually found
Two new studies, published in Nature and Science, explored whether large-language-model (LLM) chatbots can sway voters’ opinions and intentions.
- In the Nature study, researchers recruited over 2,300 U.S. adults ahead of the 2024 presidential election. Participants chatted with an AI that had been instructed to support one of the main candidates. On a 0–100 preference scale, people who initially favored one candidate but engaged with a chatbot pushing the other saw their scores shift — on average — by 2 to 4 points.
- In similar experiments conducted in Poland and Canada prior to their 2025 elections, the effect was even larger: participants’ candidate preferences moved by roughly 10 percentage points after a brief chat with a persuasive AI.
- The related Science study, with more than 77,000 UK participants, explored hundreds of political issues. The authors found that chatbots frequently swayed opinions not by emotional appeals or dramatic storytelling, but by offering many fact-based arguments — though these “facts” weren’t always accurate.
In short: when trained to argue for a particular candidate or policy, chatbots can significantly influence how people think — sometimes more than traditional campaign advertising.
Why chatbots are so persuasive
Why do we shift our political views after talking to a chatbot — even one we know is a machine? The studies provide some clues:
- Conversational, real-time replies. Unlike a static ad or a piece of mail, a chatbot reacts to your questions and follows up accordingly. That back-and-forth feels more personal and tailored.
- Information overload wrapped in civility. The most effective AIs were those that loaded their arguments with many facts and presented them politely. That “information-dense” style tends to feel authoritative — even if some pieces are misleading.
- Perceived expertise. Because chatbots speak like experts (or like neutral advisers), they may be more trusted than a flashy campaign video or a billboard — even when their content is shaky.
Interestingly, the studies found a trade-off: the more convincing the chatbot, the more likely it was to deliver inaccurate information.
Risks, misuse, and real-world consequences
The findings don’t exist in a vacuum. Generative AI is already reshaping how political content is produced and shared — for better and worse. Experts at Carnegie Mellon University (CMU) warn that the same technology that enables helpful, accessible communication can also be misused to spread disinformation.
Among the major risks:
- Fake local news and “pink slime” sites: CMU researchers have documented a surge in hyper-localized fake-news websites that recycle the same stories with different photos or names, giving the illusion of multiple independent sources. Such sites have succeeded in tricking people into donating money or believing a candidate enjoys broader support than reality warrants.
- Deepfakes — beyond text: Generative AI isn’t just about chat. It can produce videos, voice recordings, images, and memes — all potentially convincing, but entirely fabricated.
- Erosion of trust: Even if a single interaction doesn’t sway many voters, the cumulative effect of many misleading or manipulative AI-driven messages could corrode public confidence. CMU’s ethical AI researchers warn that over time, such diffuse harm might shift democratic norms, moving people away from reasoned debate toward gut reactions.
- Difficulty in measurement or regulation: Right now, there’s no reliable way to track how much AI-driven persuasion is happening at scale — or whether it ultimately changes election outcomes. Because many interactions are private (1-on-1 with a chatbot), even researchers struggle to assess the full impact.
It’s not that AI is inherently malevolent. But like any powerful tool, it can be repurposed — sometimes by bad actors who benefit from spreading confusion or reinforcing divisions.
What you — a voter — can do to stay grounded
Given how persuasive AI chatbots can be, especially when their answers sound authoritative or fact-filled, it’s wise to be a deliberate and critical consumer of political information. Here are some practical, research-backed strategies:
- Pause and reflect: If a chatbot or article evokes a strong emotional reaction — anger, fear, excitement — that may be a sign you’re being manipulated. Taking a break gives you space to think. CMU experts suggest treating those emotional “spikes” as a warning sign.
- Check facts from multiple, credible sources: Don’t rely solely on a single conversation or website. Look for confirmation (or rebuttal) from recognized news outlets, official campaign sites, or reputable fact-checks.
- Ask simple follow-up questions: If a chatbot makes a claim (“Candidate A planned to cut taxes by X”), ask for sources, dates, or counterarguments. If it refuses or gives vague answers — treat that as a red flag.
- Know the limitations: Even when chatbots try to be factual, the studies show they sometimes deliver misleading or incomplete information. The more “convincing” they are, ironically, the less accurate.
- Balance AI-based info with human discussion: Talking to friends, family, or trusted community members about political issues can help ground your views and reduce the influence of one-sided information.
In short — treat AI chatbots as tools, not oracles.
What researchers and policymakers are calling for
The urgency of these findings is already influencing conversations about regulation and responsible AI design.
- Researchers at CMU emphasize the need for “guardrails”: technical safeguards (such as “red-teaming” or stress-testing models for harmful outputs), policy-based protections, and public education.
- Some proposals under discussion in the U.S. include requiring clear disclosure when political content is generated by AI, or banning AI-driven ads that intentionally misrepresent opponents. But as of now, no major federal law has been passed — meaning voters are largely on their own.
- On the flip side, experts at CMU note there are positive uses for generative AI: translating campaign materials for non-English speakers, simplifying complex policy positions, or helping candidates engage disenfranchised voters. The catch: it only works for good when used responsibly.
In a world where AI-driven persuasion is becoming increasingly common, transparency, education, and accountability will be essential.
If you’ve ever wondered whether AI chatbots could really change minds — the answer is “yes, to an extent.” These systems are powerful enough to nudge opinions, especially when they resemble trusted, fact-filled advisers. But their effectiveness depends heavily on how they’re used — and whether users stay curious, skeptical, and informed.
With a little awareness and some critical thinking, you can stay one step ahead — and help preserve the integrity of your own beliefs.
