News - Jagiellonian University

Conversations with chatbots influence voters’ political decisions

Conversations with chatbots influence voters’ political decisions

AI-based chatbots are capable of genuinely changing voters’ attitudes—and they do so more effectively than traditional political advertising. Even a short conversation with a language model can shift voting preferences by several percentage points, according to an international study published in Nature. The study was co-authored by Dr Gabriela Czarnek from the JU Faculty of Philosophy.

The impact of artificial intelligence on democratic societies and democratic procedures is the cause of growing concerns. A major part of these worries relates to the potential for widely accessible chatbots based on large language models (LLMs) to influence voters’ attitudes. Trained on massive datasets, these tools process and generate human-like speech, enabling conversations with users.

However, it has remained unclear how effective generative AI is - and will be - in shifting voter attitudes, especially in presidential campaigns. Some scientific studies show that AI can outperform humans in persuasion, and that engaging in dialogue with it is more convincing than static messages such as campaign ads.

In the study Persuading voters using human–artificial intelligence dialogues published in Nature, an international team of researchers from American, Canadian and Polish institutions examined whether conversations with state-of-the-art AI models could meaningfully change the political attitudes of Americans, Canadians, and Poles. One of the co-authors is Dr Gabriela Czarnek from the Institute of Psychology at the JU Faculty of Philosophy.

'In the context of the US 2024 presidential election, the 2025 Canadian parliamentary election, and the 2025 Polish presidential election, we randomly assigned participants to conversations with an AI model that advocated for one of the two main candidates', explained Dr Czarnek.

The study involved more than 2,300 Americans, who at the end of 2024 held a conversation with a chatbot. Participants indicated their preference between Democratic candidate Kamala Harris and Republican candidate Donald Trump on a 0–100 scale, as well as their likelihood of voting. They then spoke with a suitably 'programmed' chatbot designed to shift their attitudes toward a given candidate.

The AI model received instructions to keep its messaging positive, respectful, and grounded in facts. The chatbot was also required to use persuasive arguments and analogies to build rapport with the participant. It was provided with information about each participant’s intended vote so it could personalize its message. After the conversation, participants completed the survey again. More than a month later, they were contacted once more to assess the durability of the effects.

'We observed significant persuasive effects on candidate preferences when the discussion focused on specific political issues - effects larger than those typically seen with, for example, traditional video advertisements. The AI model supporting Donald Trump shifted potential Kamala Harris voters by 2.3 percentage points toward the Republican candidate. Meanwhile, the model supporting Kamala Harris moved likely Trump voters by 3.9 percentage points toward the Democratic candidate', says Dr Czarnek.

This effect is about four times stronger than the impact of traditional advertising tested in the 2016 and 2020 elections, where shifts in favor of candidates did not even reach 1 percentage point.

In similar experiments with Canadians and Poles, the effects were much larger. Conversations with chatbots shifted the voting intentions of people who had previously supported opposition candidates by roughly 10 percentage points.

In the experiment conducted during the week preceding the Canadian federal election in April 2025, more than 1,500 Canadians were randomly assigned to speak with AI models advocating either for Liberal Party leader Mark Carney or Conservative Party leader Pierre Poilievre. The persuasive effect in Canada was almost three times stronger than in the US experiment. However, when the AI model was deprived of the ability to reference facts and data, the effect dropped by more than half.

In May 2025, during the two weeks preceding Poland’s presidential election, more than 2,100 Poles were randomly assigned to conversations with AI models advocating either for Civic Coalition candidate Rafał Trzaskowski or Law and Justice-backed candidate Karol Nawrocki. As in Canada, the persuasive effect was nearly three times larger than in the US, and removing the AI’s ability to reference facts reduced the effect by as much as 78%.

Analysis of the persuasion strategies used by the AI models indicates that they relied on referencing facts and data rather than psychological techniques. They were almost always polite and used language accessible to voters. They tended to emphasise positive outcomes and tailor their message to participants’ values. Interestingly, AI models rarely employed strategies commonly discussed in psychological or political science literature, such as direct calls to vote, eliciting anger, social influence tactics, or testimonial-based appeals.

'However, not all of the information presented as factual was actually correct. In all three countries, AI models advocating for right-leaning candidates more often provided inaccurate claims than models promoting centrist candidates. This aligns with earlier research showing that in the US, right-leaning voters are more likely to share misleading content. Generative models appear to replicate information inequalities observed in society', noted the Jagiellonian University researcher.

As the authors conclude, dialogue with language models can meaningfully shift voters’ attitudes and voting intentions. This has significant implications for the future of political persuasion, political advertising, and democracy.

Source: JU Institute of Psychology

Recommended
JU ranked among most sustainable Polish universities
JU ranked among most sustainable Polish universities
Poland and Europe's future discussed at Open Eyes Economy Summit
Poland and Europe's future discussed at Open Eyes Economy Summit
JU hosts an international conference of university management leaders
JU hosts an international conference of university management leaders
Prof. Wojciech Szczeklik named Researcher of the Year
Prof. Wojciech Szczeklik named Researcher of the Year