A study by the European Broadcasting Union and the BBC found that 48 percent of answers from four widely used AI assistants contained accuracy issues, according to a Statista chart published November 28, 2025. The assistants were the free versions of ChatGPT, Gemini, Copilot and Perplexity. Journalists analyzed the responses, which were collected in May and June 2025.

Within that 48 percent, 17 percent of answers carried significant errors, which the study traced mainly to sourcing problems and missing context. The chart's remaining share covers minor inaccuracies.

The chart compares those results with an earlier round of testing from December 2024. That round used a smaller sample of answers. In it, 72 percent of responses from the same four assistants showed inaccuracies, and 31 percent carried major issues. The share of flawed answers fell by 24 percentage points between the two rounds, and the share with major problems fell by 14 points.

Statista data journalist Tristan Gaudiaut wrote that ChatGPT's public release on November 30, 2022 moved chatbots into the mainstream, and that hallucinations, misinterpretations of context and inaccuracies still appear in the most capable models available. The analysis says the remaining error rates raise questions about reliability in high-stakes settings such as healthcare, legal advice and education.

The two rounds differ in design. The December 2024 figures came from a smaller answer sample than the May and June 2025 figures, so the change over time compares two datasets of different sizes. The assistants named in the chart come from OpenAI, Google, Microsoft and Perplexity, all US companies.

Source: Statista - https://www.statista.com/chart/35534/ai-chatbots-accuracy-rate-of-inaccurate-responses-2025/