AI chatbots may have the potential to sway voter opinions
Enrique Shore / Alamy
Could the persuasive abilities of AI chatbots signal the decline of democracy? A substantial study investigating the impact of these tools on voter sentiments revealed that AI chatbots surpass traditional political campaign methods, such as advertisements and pamphlets, in persuasiveness, rivaling seasoned campaigners as well. However, researchers see reasons for cautious optimism regarding how AI influences public opinion.
Evidence shows that AI chatbots, like ChatGPT, can migrate the beliefs of conspiracy theorists, winning converts to more reasonable positions and attracting support during human debates. This capability raises valid worries about AI possibly skewing the digital scales that determine election results or being misused by malicious entities to manipulate users towards certain political figures.
The concerning part is that these fears have merit. A survey involving thousands of voters who participated in recent elections in the US, Canada, and Poland found that David Rand and researchers at MIT discovered that AI chatbots effectively swayed individuals to back specific candidates or alter their stance on certain issues.
“Conversations with these models can influence attitudes towards presidential candidates—contributions often deemed deeply entrenched—more than previous studies would suggest,” Rand remarks.
In their American election analysis, Rand’s team surveyed 2,400 voters, asking them about the most significant policy issues or characteristics of a potential president. Subsequently, voters rated their preferences for the leading candidates, Donald Trump and Kamala Harris, on a 100-point scale and answered additional questions to clarify their choices.
The answers were inputted into a chatbot, such as ChatGPT, with the objective of persuading the voters to support an already favored candidate or switch their support to a less favored one. The interaction took about six minutes, consisting of three question-and-answer exchanges.
Following the AI interaction and a one-month follow-up, Rand’s team discovered that voters adjusted their candidate preferences by an average of 2.9 points.
Furthermore, the researchers examined AI’s capacity to influence views on specific policies and noticed a substantial change in opinions regarding the legalization of psychedelics, shifting voter support by approximately 10 points. In comparison, video ads impacted views by only about 4.5 points, and text ads swayed opinions by merely 2.25 points.
The magnitude of these findings is remarkable. Sasha Altai of the University of Zurich stated, “These effects are considerably larger than those typically observed with traditional political campaigning and are comparable to the influence stemming from expert discussions.”
Nevertheless, the study reveals a more hopeful insight: these persuasive interactions predominantly stemmed from fact-based arguments rather than personalized content, which tends to exploit users’ personal information available to political operatives.
Another study of approximately 77,000 individuals in the UK assessed 19 extensive language models across 707 distinct political issues, concluding that AI performed best when employing fact-based arguments, as opposed to tailoring its discussions to the individual.
“Essentially, it’s about creating a compelling argument that prompts a mindset shift,” Rand explains.
“This bodes well for democracy,” notes Altai. “It indicates that individuals are often more influenced by factual evidence than by personalized or manipulative strategies.”
There is a need for further research to confirm these findings, asserts Claes de Vries at the University of Amsterdam. He adds that if replicated, the controlled environments of these studies—where participants engaged with chatbots extensively—might differ significantly from individuals’ typical political interactions with friends or colleagues.
“The structured setting of interaction about politics with a chatbot is quite different from how people usually engage with political matters,” he mentions.
Despite this, De Vries notes growing evidence that individuals are indeed turning to AI chatbots for political advice. A recent survey of over 1,000 voters in the Netherlands ahead of the 2025 national elections found that about 10% sought AI guidance regarding candidates, political parties, and election matters. “This trend is particularly noteworthy as the elections approach,” De Vries points out.
Even if people’s engagements with chatbots are brief, De Vries asserts that the integration of AI into political processes seems unavoidable, as politicians seek tools for policy recommendations or as AI generates political advertisements. “As researchers and as a society, we must recognize that generative AI is now a vital aspect of the electoral process,” he states.
Topics:
- artificial intelligence/
- US election
Source: www.newscientist.com
