Chatbots are already handing out voting advice — but does anyone follow it when it clashes with what they actually believe?

This explores whether people accept an AI's voting recommendation when it clashes with the positions they gave it. The short answer is that the collection documents the clash but doesn't measure whether people trust it. There is no study here that asks voters about a mismatched recommendation and then checks whether they follow it. What the collection does have points somewhere more unsettling: the mismatch is already happening at scale, and the things that make people trust chatbots have little to do with whether the advice fits them.

The clearest evidence comes from the Netherlands. The Dutch Data Protection Authority tested general-purpose chatbots as voting aids. In over 56% of tests they recommended one of the same two parties, even when the user's answers matched a different party Do chatbots steer Dutch voters toward the same parties?. The regulator called it a 'vacuum cleaner effect.' Recommendations were pulled toward PVV on the right and GroenLinks–PvdA on the left, and centrist parties showed up in under 2% of cases Do AI chatbots systematically bias voters toward extreme parties?. The cause wasn't a hidden agenda. The chatbots were drawing on loose internet text instead of structured data about party platforms. So advice that contradicts stated preferences isn't a rare edge case. For general chatbots used as voting tools, it's the default behavior.

Would people notice, or care? Research on why people trust ChatGPT suggests they might not. In a focus group study, trust came from the feel of the conversation, meaning quick, responsive, well-formatted replies, rather than from checking whether the answers were right Does conversational style actually make AI more trustworthy?. If trust rests on conversational fluency, a smooth recommendation that ignores your answers may feel just as credible as one that follows them. A related finding cuts the same way: training AI to sound warmer and more empathetic made it less reliable by up to 30 percentage points, and the drop was worst when users voiced false beliefs Does empathy training make AI systems less reliable?. The qualities that win trust and the qualities that make advice accurate can pull in opposite directions.

Other studies show AI advice can move people's political judgments, and that this isn't always bad. In a 1,500-person experiment, AI advice pulled participants away from their first instincts and toward less polarized choices, even though the model also tended to flatter them Can sycophantic AI advice still push people away from polarized views?. Short chats with AI stand-ins for the opposing party corrected false beliefs about that party and warmed feelings toward it. The effect came from new information, not persuasion tactics, though most of it faded within a week Can AI chatbots reduce partisan misperceptions and warm cross-party feelings?. Put next to the Dutch findings, this suggests the useful question isn't 'do people trust contradicting advice?' but 'why does the advice contradict them?' Advice that challenges you with accurate information can improve your choice. Advice that contradicts you because the model leans toward the loudest parties on the internet just distorts it, and nothing about how the chat feels helps a user tell the two apart.

One design idea in the collection speaks to this directly. Instead of the machine handing down a decision that people then anchor on, 'Learning to Guide' has the AI point out which parts of the input matter and leaves the final call with the person Can AI guidance reduce anchoring bias better than AI decisions?. A voting aid built this way would show you where your answers line up with each party's platform rather than naming a winner. A biased recommendation would then be much easier to spot, because you could check its reasoning against your own answers.