AI chatbots sometimes gave wrong answers about local political issues despite consulting official government sources, according to a new study. In the tested sample, Perplexity had the highest error rate, a timely warning as voters turn to AI for election information.
Watch Desk analysis
What happened
Searchable, an AI search analytics company, evaluated 7,449 answers from ChatGPT, Gemini, Google AI Mode and Perplexity. Tom’s Guide reports that 610 answers, or 8.2 per cent, contradicted official government data. Perplexity’s error rate in the study was 22.6 per cent, compared with 6.2 per cent for Gemini, 4.4 per cent for ChatGPT and 2.9 per cent for Google AI Mode.
The test covered 106 underlying questions across five metropolitan areas and their states, on issues including crime, housing, employment, education and healthcare. Searchable found that questions containing false assumptions were more than four times as likely to produce a false answer. About 62 per cent of the 43,079 web pages consulted by the systems came from federal, state or local government sites. Two graders had to agree before an answer was classed as false.
Why it matters
The results point to a gap between finding a credible source and interpreting it correctly. A chatbot can cite an official page and still get the answer wrong, so a list of links is not a substitute for checking what those sources actually say. That matters particularly for local issues, where an answer can shape how someone judges a candidate’s record.
The study is not a measure of each system’s accuracy across every political question. It tested a defined set of prompts, including some built around misleading or debunked claims. Its findings are best read as evidence about those tests, not a universal ranking of chatbot reliability.
Our read
Treat an AI answer about a candidate or local policy as a lead, not a verdict. Ask a neutral question, check the date and open the cited government source to see whether it supports the claim. For polling places, registration and voting hours, go directly to the relevant election office. Slightly less futuristic, considerably more dependable.
What to watch
- Whether further testing finds similar error patterns across different regions and election topics.
- Whether AI tools make it easier to check that a cited source actually supports an answer.
- How systems handle questions that start with a false or unverified assumption.
Discussion spark: Should election chatbots be expected to challenge false assumptions in a question, or is it the user’s job to check the premise before asking?
Sources and evidence
- ChatGPT, Gemini and Perplexity are getting election facts wrong – and a new study shows why – Tom's Guide (29 October 2026, 14:49 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.