Are AI chatbots effective at changing political opinions?
A recent large scale study of over 77,000 adults based in the UK, who were paid to participate, asked for people’s views on political topics, such as taxes and immigration - then an AI chatbot tried to change their mind, to an opposing view.
Often the chatbot succeeded.
The study, presented in the journal, Science (Dec 2025), is part of wider research designed to look at how AI could affect politics and democracy, through swaying opinions.
The most persuasive tactic? The AI chatbots ability to 'rapidly access and strategically deploy information' during the back and forth of the chat, reinforcing their position.
Which was more effective than alternate debating tactics such as appeals to morality or arguments personalized to the individual.
The conversational style of data delivery and discovery was 41-52% more effective than giving the data in a static format all at once.
Not only that, the effect was shown to be long-lasting, with the 36-42% of the persuasive effect still in place one month later.
Ironically, the researchers identified that “The most persuasive models and prompting strategies tended to produce the least accurate information”, with 19% of the claims made by the chatbot being false.
They concluded that the chatbot “exceed the persuasiveness of even elite human persuaders, given their unique ability to generate large quantities of information almost instantaneously during conversation”.
17 different language models were included in the test, including some that were fine-tuned.
The information may be flawed, but the quantity and confidence that it is provided gives a sense of authority and believability.
The authority bias (https://en.wikipedia.org/wiki/Authority_bias) is a tendency to attribute greater accuracy to the information from a figure in authority, which leads to being influenced by their opinions. In this case the chatbot is establishing itself as a source of authority.
Previous studies into persuasiveness have not demonstrated such a strong effect.
IMPLICATIONS FOR OUR FIELD
1. The increased persuasiveness of AI Chatbots to change humans 'core' beliefs.
2. That the strategic deployment of credible looking information can sway opinions, establishing a sense of authority in the person, or bot, providing it.
3. That the accuracy of the information is less of a factor - which can lead to manipulation, by the model, the prompt or the data being used.
4. That those who control the model, the prompt or the data can manipulate populations at scale.
Source
Link to the original Science research: https://www.science.org/doi/10.1126/science.aea3884
Link to the NBC article: https://www.nbcnews.com/tech/tech-news/ai-chatbots-used-inaccurate-information-change-political-opinions-stud-rcna247085
BESCI AI OPINION
This one really made me think - the implications are wide-ranging.
In the NBC article, David Brockman, a professor from Berkeley, argues that in a real world situation, the arguments from each side would cancel each other out. This assumes that both sides have the money and access to those being influenced.
I can't help compare what we are seeing, especially in persuasion at scale, to the James Bond movie 'Tomorrow Never Dies' where the baddie media baron uses his control of global media to do evil and profit from it.
Those that control the model, the prompts/agents, or the data gain persuasive power.
