AI chatbots influence voters and provide incorrect information in elections
Research shows voters worldwide use chatbots like ChatGPT and Gemini for election guidance, with studies finding these systems provide incorrect political information, concentrate recommendations among a narrow set of candidates, and influence voter preferences more strongly than traditional political advertising.
What happened
Studies show that voters in multiple countries ask chatbots including ChatGPT and Gemini for voting recommendations. In the Netherlands, a University of Amsterdam study found that 10% of voters were inclined to follow chatbot recommendations on who to vote for, with another 13% uncertain.
A 2024 MIT study of 2,400 voters found that a chatbot favourable to Kamala Harris shifted Trump supporters 3.9 points toward her, approximately four times the measured impact of political advertisements in 2016 and 2020. A chatbot favourable to Trump shifted Harris supporters 2.3 points toward him.
In Chile, 27% of voters approached the election via chatbots, rising to 44% among higher-income groups. Younger voters aged 18 to 34 were more inclined to follow chatbot recommendations than those over 55.
Reported impact
Studies show that some chatbot responses about political party positions were incorrect. In Dutch research, 80% of chatbot recommendations on one platform focused on only two parties regardless of the question asked.
Electoral regulators face challenges as voters seek election guidance from AI chatbots rather than authoritative sources.
Index analysis
The incorrect responses and concentrated recommendations suggest chatbots may not understand how they generate suggestions or may be shaped by factors including media coverage of particular political figures.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.