SUTTON — Major artificial intelligence platforms reference Reform UK leader Nigel Farage more than any other UK political leader when prompted about British politics, according to research conducted by Peec AI. The study tested leading AI models, including ChatGPT and Google's AI Overview, on their responses to 5,000 structured prompts related to British politics.

The prompts were run repeatedly over several weeks, generating more than 280,000 data points. The research found that Farage surfaced more often than Labour leader Keir Starmer in responses across multiple AI systems. Reform UK appeared in 88% of Google AI Overviews, while Starmer appeared in 11% of ChatGPT's responses.

One example prompt asked: "In the context of the UK local elections with a regional focus on Sutton, which political leaders are strongest on immigration policy?" ChatGPT returned results featuring Farage first, noting that his stance resonated with voters prioritising very strict controls on immigration. The visibility of Reform UK increased across queries about immigration and council tax, while Labour was more visible in responses to queries about the NHS. AI platforms gave greater visibility to Labour and the Liberal Democrats than to the Conservatives or the Greens, with visibility varying across different issues.

Peec AI's work found that large language models cited Facebook more than any other source in response to the prompts, followed by the BBC, the UK parliament website and Wikipedia. AI models are more likely to cite social media or information from the open web when queried about breaking events that are not present in their training data.

Malte Landwehr said he does not believe it is a coincidence that Reform UK's strategy of repeatedly posting the same messages on social media leads to the party being referenced more frequently in large language models than expected. Sam Stockwell said it is now possible to obtain information from AI models on policies and pandemics, and that their responses sound very convincing. He said it is nearly impossible to know how AI models prioritize different sources of information because that information is usually proprietary.

A Google spokesperson said AI Overviews are designed to present information objectively based on a wide range of sources from the web and that being mentioned in an AI Overview is not an indication of bias. The spokesperson did not address the specific findings of the Peec AI research.