Two studies suggest that AI chatbots can shift political views more effectively than TV campaign ads, by presenting many claims, regardless of their accuracy
New research suggests AI chatbots can shift people's political views more effectively than campaign ads on TV.
Washington PostWill Oremus
Context & Ripple Effects
Earlier research had already found that AI assistants could influence users without their awareness through collaboration and social norms, a concern now sharpened by evidence of subtle assistant influence. Separately, the rise of AI-generated political advertising had prompted calls for campaign guardrails, creating a policy backdrop for this more conversational form of persuasion.
The new studies matter because they shift attention from AI as a tool for producing campaign material to chatbots as a direct channel through which political claims can be delivered at volume.
First-order effects
Chatbot providers and political-risk researchers face stronger evidence that conversational AI can affect political attitudes even when the claims it presents are not accurate.
Campaigns and advocacy groups gain a reason to evaluate chatbot interactions alongside television advertising as a potential persuasion channel, while users face a higher burden to assess claims encountered in dialogue.
Second-order effects
Pressure on providers to constrain misleading political responses is likely to rise, extending concerns already associated with AI-generated campaign advertising to general-purpose assistants.
Political-advertising guardrails may prove incomplete if they focus on labeled ads while persuasive claims can instead be delivered through one-to-one chatbot exchanges.
Third-order effects
If replicated across settings, the findings point to political persuasion becoming more interactive and less easily separated from everyday AI use, complicating disclosure, accountability, and enforcement models built around discrete ads.
The key policy question shifts from whether campaigns use AI to how platforms govern persuasive, inaccurate content in systems that users may treat as conversational advisors.
The trend: AI is moving from a campaign-production tool toward an always-available, conversational layer of political influence, increasing the importance of safeguards around trust and accuracy.
Yes, this is a crucial point. It's definitely an important finding that chatbots can have these effects, but the real-world impacts are trickier to predict than it might seem. [image]
🚨 New today in @ScienceMagazine !!🚨 We're publishing the results of the largest AI persuasion experiments to date: 76k participants, 19 LLMs, 707 political issues We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more... 🧵: [image]
🚨 New in Nature+Science!🚨 AI chatbots can shift voter attitudes on candidates & policies, often by 10+pp 🔹Exps in US Canada Poland & UK 🔹More “facts”→more persuasion (not psych tricks) 🔹Increasing persuasiveness reduces “fact” accuracy 🔹Right-leaning bots=more inaccurate [image]
Today (w/ @UniofOxford @Stanford @MIT @LSEnews) we're sharing the results of the largest AI persuasion experiments to date: 76k participants, 19 LLMs, 707 political issues. We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more 🧵 [ima…
Today, our new study with colleagues at @OIIOxford @Stanford @MIT and @LSEnews was published in @ScienceMagazine. It is the largest empirical investigation to date of how conversational AI can influence people's political attitudes. [image]
Now to the Science paper Here we scaled things up: 🔹76,977 UK participants 🔹707 political issues 🔹19 different LLMs (open & closed source) We asked: what factors boost AI persuasion the most - model size, personalization, prompting, or post-training? https://www.science.org/...
Can chatting with AI sway your political beliefs? A new joint study from the OII and @AISecurityInst offers unprecedented insights into how conversational AI can exert influence over people's political beliefs. @KobiHackenburg @HelenMargetts Read more: https://www.oii.ox.ac.uk/..…
In @Nature , we ran experiments where voters had short (~6 min) text chats with an AI told to promote one candidate/side We tested this in: 🇺🇸 2024 US presidential race 🇨🇦 2025 Canada national election 🇵🇱 2025 Poland prez election +MA psychedelics ballot https://www.nature.com/..…
New: AI chatbots can change voters' minds, according to a pair of in-depth studies published just now in Science and Nature. — How they do it is interesting — and concerning. Gift link: wapo.st/49RSstP