Study Finds AI Search Tools Are Inaccurate 60% of the Time

Research highlights widespread errors and overconfidence in AI-generated search results, with premium tools often performing worse than free versions.

• The Tow Center for Digital Journalism tested eight AI search tools, including ChatGPT, Perplexity, and Microsoft's Copilot, for accuracy in retrieving and citing news content.
• On average, AI search tools were correct less than 40% of the time, with Perplexity performing best at 63% accuracy and Grok-3 the worst at just 6%.
• ChatGPT responded to all queries but was completely accurate only 28% of the time, while Microsoft's Copilot declined over half of the queries and was 70% inaccurate for those it answered.
• Premium versions of AI tools often provided more confidently incorrect answers compared to their free counterparts, raising concerns about transparency and value for users paying up to $200 per month.
• The study also revealed that many AI tools fabricated links, ignored website restrictions, and misattributed content, underscoring their unreliability for factual searches.

3 Articles

TechSpot Digital Trends 9to5Mac