Millions use it, but a third give false information

According to the report, chatbots are no longer backing down by saying "I don't know enough about this" but are instead opting to provide an answer, even if it's inaccurate. This trend could have serious consequences, especially at a time when disinformation campaigns are growing in influence.

WHO IS AT THE TOP?

The chatbot with the highest error rate was Inflection AI's Pi, which included 57 percent of its responses with false information. Perplexity followed with 47 percent. ChatGPT and Meta's Llama included 40 percent of the responses with false information, while Microsoft's Copilot and Mistral's Le Chat were at 35 percent.

The bots with the lowest error rates were Anthropic's Claude (10 percent) and Google's Gemini (17 percent). Perplexity, in particular, experienced a notable decline: While the system, which operated with zero errors in 2024, returned incorrect information in 46 percent of its responses in August 2025.

Mistral's performance remained steady. The bot generated 37% of misinformation in both years, and repeated false claims about French President Emmanuel Macron and his wife Brigitte Macron in 31% of its French replies and 58% of its English replies.