AI chatbots can make inaccurate political claims more consequential by presenting them in fluent, interactive conversations that may also persuade people. Experiments show that chatbot dialogues and AI-written political messages can shift attitudes; one 2025 study also found inaccurate claims in candidate-advocacy chatbots. This establishes a plausible risk—not proof that chatbot misinformation changed an election result.
How can a chatbot amplify political misinformation?
Here, “amplify” does not mean that a chatbot necessarily spreads a claim to more people than a social network does. It means an inaccurate claim may carry more weight when it appears inside a responsive exchange: the system can address a person’s question, supply reasons, and continue the conversation. If the answer is persuasive but wrong, its presentation may make it harder for a user to recognize the error.
The concern comes from the combination of two findings: some chatbot conversations can shift candidate preferences, and candidate-advocacy models have made inaccurate claims in experiments. The cited work does not establish that the inaccurate claims caused the attitude changes. Persuasion could also result from accurate facts, argument quality, or other features of the exchange.
What do the studies show?
The evidence covers different kinds of political influence. Candidate preference, policy attitude, factual accuracy, and belief in misinformation are separate outcomes; a change in one does not prove a change in the others.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute#1 Best Overall
| Study and setting | What was measured | What the result supports—and does not establish |
|---|---|---|
| Lin and colleagues, preregistered experiments tied to the 2024 U.S. presidential election and the 2025 Canadian and Polish elections | Participants were randomly assigned to converse with a model advocating for one of the leading candidates; researchers assessed candidate preference and analyzed model claims. | The researchers report significant candidate-preference effects and inaccurate claims. In all three country contexts, models advocating for right-leaning candidates made more inaccurate claims than models advocating for left-leaning candidates. This is a finding about the tested models and conditions, not a universal rule about parties or AI systems. The study does not isolate inaccurate claims as the cause of persuasion. |
| Bai and colleagues, three preregistered policy-message experiments, 2025 | Across 4,829 participants, researchers compared attitude change after LLM-generated policy messages with a neutral-message control and lay-human messages. | LLM-generated messages shifted policy attitudes relative to the neutral control and were similarly effective to lay-human messages in these experiments. These were messages, not interactive chatbot sessions, and the finding is about attitudes—not candidate choice or voting behavior. |
| Potter and colleagues, EMNLP 2024 | The paper examined political preferences in 18 open-weight and closed-source LLMs, then had 935 U.S. registered voters interact for five exchanges with Claude-3, Llama-3, or GPT-4. | The abstract reports that about 20% of Trump supporters reduced their support for Trump after interacting. Participants were not instructed to persuade users toward Biden. The result does not show that every chatbot has the same political leaning, that misinformation caused the change, or that an election outcome changed. |
| UK conversational-AI survey preprint, 2025 | A representative public survey asked about conversational-AI use for information relevant to electoral choice in the week before the 2024 UK election. | The authors report that 32% of chatbot users and 13% of eligible voters used conversational AI for such information. The preprint concerns use and political knowledge, not an experiment proving persuasion or increased belief in false claims; its authors caution that greater use need not mean greater belief in political misinformation. |
Do chatbots influence voters?
In controlled experiments, chatbot conversations have shifted candidate preferences. Lin and colleagues tested conversations advocating for leading candidates in three election contexts: the 2024 U.S. presidential election and the 2025 elections in Canada and Poland. They report significant effects on preference and say the effects were larger than those typically observed for traditional video advertisements.
That result is evidence of influence under the study conditions, not a measurement of how voters behave in an actual campaign. It does not establish that a chatbot conversation changed a vote, affected turnout, or decided an election. Nor does it show that misinformation was the mechanism behind the observed preference changes.
The earlier EMNLP 2024 voter study offers a related but distinct signal: after five exchanges with one of three named models, the abstract reports that roughly one in five Trump-supporting participants reduced their support. That finding should remain attached to its U.S. voter sample, models, and interaction design; it is not a general estimate of chatbot effects on voters.
Can AI-written political content persuade without a chatbot?
Yes, in the policy-message experiments reported by Bai and colleagues. Their three preregistered experiments included 4,829 participants and found more attitude change after LLM-generated messages than after a neutral control message. The authors found the generated messages similarly effective to messages written by laypeople in those experiments.
The authors associate persuasiveness with the messages’ use of facts, evidence, logical reasoning, and a dispassionate voice. This gives a reason to examine AI-generated political content beyond chat interfaces, but it should not be collapsed into the chatbot findings: a one-way policy message is not an adaptive dialogue, and policy attitude is not the same outcome as candidate preference.
Does using chatbots for political information increase belief in misinformation?
The available evidence here does not establish that it does. The 2025 UK preprint reports that conversational AI was used for electoral-choice information by 32% of chatbot users and 13% of eligible voters during the week before the 2024 election. Those figures describe use, not whether users believed false information or changed their political views because of it. The authors explicitly caution that greater use may not increase public belief in political misinformation.
A separate background source helps explain why the distinction matters. Pfänder and Altay’s 2025 systematic review and preregistered meta-analysis combined 303 effect sizes from 67 experimental articles, involving 194,438 participants in 40 countries across six continents, on how people judge true and false news. It is evidence about news judgment broadly, not a direct measure of chatbot misinformation or its effect on political decisions.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What remains uncertain—and how should readers evaluate a political answer?
The cited studies support a credible risk, but they do not demonstrate that chatbot misinformation changed a real election outcome. They also do not establish which safeguards reliably prevent false claims from influencing political decisions. Disclosures, watermarks, or model guardrails should not be treated as proven remedies on the basis of these findings.
For an individual political decision, treat a chatbot answer as a starting point rather than the final authority. Separate checkable factual assertions from interpretation or advocacy, and verify consequential claims against reliable, independent sources. This is a prudent way to handle uncertainty; the studies above do not test it as a specific intervention.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
