AI search tools get 8.2% of election answers wrong, Perplexity 22.6%

Searchable evaluated 7,449 answers from ChatGPT, Gemini, Google AI Mode and Perplexity on local US political issues: 610 (8.2%) contradicted official government data. Perplexity had the highest error rate at 22.6%, versus 4.4% for ChatGPT, 6.2% for Gemini and 2.9% for Google AI Mode. About 62% of the 43,079 webpages the tools consulted were government sites.

• 7,449 answers tested, 610 false — an 8.2% error rate
• Perplexity erred on 22.6% of answers, Google AI Mode on 2.9%
• 62% of 43,079 sources consulted were government sites
• Questions with false premises were 4x more likely to get false answers