AI giving misleading advice on voting

Ahead of the elections for the Welsh Senedd and Scottish Parliament on Thursday, some of the most popular artificial intelligence chatbots have been found to give false and misleading information to prospective voters.

People are increasingly using AI chatbots, such as ChatGPT, Claude, and Grok, to gather information, including information on how to vote in upcoming elections. Research by the AI Security Institute found that around a third of chatbot users sought advice from AI on how to vote, including 13% of eligible voters in the 2024 general election.

Investigations by the BBC showed that several of the leading AI chatbots gave inaccurate and confusing responses to questions relating to the Senedd elections this week. The team from BBC Wales posed as fictional voters and asked ChatGPT, Copilot, Claude, Gemini, and Grok who they should vote for, as well as factual information about the candidates in their area and how the voting system worked.

Although some platforms initially refused to answer the question about who to vote for, all eventually gave a recommendation. In many cases the chatbots also gave helpful insights into the different political parties and all platforms described the new electoral system for the Senedd accurately.

However, the artificial intelligence systems did make significant mistakes. One chatbot described Rhun ap Iorwerth as being the leader of Plaid Cymru “until recently” despite the fact that he still leads the party. Other platforms left out important information or misrepresented a party’s position on certain issues.

Some of the AI platforms were not able to correctly identify the constituency of a certain town, and more than one listed candidates who were not actually standing in that constituency (including, in one case, a candidate who had died in 2025). Even more accurate lists of candidates given by chatbots were incomplete.

OpenAI, who make ChatGPT, said that the AI gave accurate information but could make mistakes, and that they were focused on improving factual accuracy. Google said that their Gemini chatbot was designed to “provide a balanced presentation of multiple points of view” but cautioned people to double-check information given.