What happened
Lots of people now ask a chatbot, “Is this real?” when they see a shocking picture or video. Full Fact, a UK fact-checking charity, tested how well that works. From February 2026, it asked Grok, ChatGPT and several versions of Google’s Gemini about pictures and videos it had already checked.
On 28 August 2026, it published the results: 39 major errors, and 67 wrong answers in total. For example:
- Several chatbots said a computer-made picture of a Ryanair plane window was a real photo.
- Grok and one version of Gemini said a computer-made picture of a children’s funeral in Iran was real.
- Grok said a video of a fire at Glasgow Central Station showed Tel Aviv after an Iranian missile attack.
Why it went wrong
Chatbots don’t really look things up the way a fact-checker does. They give an answer that sounds likely. During breaking news, there is little reliable information to go on, so they guess, and they say it with the same confidence as when they’re right.
Impact
- People use these answers to decide what to share. A wrong “yes, it’s real” spreads fake pictures faster.
- OpenAI said accuracy “remains an important focus.” Google said retested claims “were consistently debunked.”
The lesson
- Don’t ask a chatbot to fact-check breaking news. Check trusted news sources or fact-checkers instead.
- Look for the original. Where did the picture first appear, and who took it?
- If it makes you angry, slow down. Fake pictures are made to be shared before anyone checks.