The author examines why AI chatbots are increasingly spreading misinformation. According to a study by NewsGuard, a company specialising in assessing the credibility of news sources, the ten best-known AI chatbots now spread almost twice as much false information as they did a year ago. In 2024, the chatbots tested produced misinformation in 18% of cases; in 2025, this figure had risen to 35%.
Perplexity recorded an error rate of more than 46%, compared with no false information at all in the previous year. NewsGuard attributed an error rate of 40% to both ChatGPT and Meta AI. Microsoft Copilot and the French AI chatbot Mistral each produced incorrect answers in 36.67% of cases. One possible explanation identified by the authors is that the models now tend to generate an answer to virtually every question. A year earlier, 31% of queries went unanswered – and refusing to answer can also help prevent false information.
Ultimately, users should be aware that chatbots are not a particularly reliable way to find information on breaking or very recent news, especially when the same information is already available from a wide range of current sources.
By Maria Kleiner
Source: t3n digital pioneers, 15 September 2025, by Kim Rixecker
