The biggest AI chatbots are supposed to help people find information, not manufacture misinformation. But a new investigation suggests that separating those two jobs isn’t as straightforward as many AI companies would like users to believe. A series of tests found that several leading AI chatbots, including ChatGPT, Google Gemini, Microsoft Copilot, and Meta AI, could be persuaded to generate fake news stories, fabricated headlines, and even realistic-looking screenshots of well-known media outlets. While every chatbot showed some weaknesses, ChatGPT reportedly performed the worst, producing the most convincing fake content with surprisingly little effort.
The findings come from an investigation by German newsroom CORRECTIV, which tested how the AI assistants responded when asked to create false reports involving sensitive topics such as election fraud, government coups, war, vaccines, climate change, and conspiracy theories.
The fake news problem isn’t going away
According to the investigation, ChatGPT generated the majority of the requested fake content, sometimes after only minor prompt adjustments. In several cases, it reportedly produced highly realistic mock-ups of articles that closely resembled established news organizations. One example included a fabricated screenshot designed to look like a legitimate news article displayed inside a web browser on a desktop computer. In another instance, ChatGPT reportedly refused to create misleading news content in text while still generating the accompanying fake image anyway — a contradiction that raises fresh questions about how consistently its safeguards are enforced.
The report also found an interesting inconsistency. ChatGPT allowed fake versions of several German news outlets to be recreated, while similar requests involving publications such as The New York Times and the BBC were blocked. OpenAI did not explain why those differences exist. Instead, the company said it continuously improves its safety systems and noted that using ChatGPT to deceive people violates its policies.
No chatbot came away with a perfect score
ChatGPT wasn’t the only AI assistant that struggled. Google’s Gemini also created fabricated articles and fake screenshots involving multiple news organizations, although the final results weren’t considered as convincing. Google pointed to its policies against harmful misinformation, while acknowledging that generative AI systems still operate within the limits of their training. Microsoft Copilot also generated fake news-style content during testing, though it generally included an “AI-Generated” label and produced designs that were easier to distinguish from genuine articles.
Meta AI proved the most resistant overall. The chatbot created only one fake media post during testing and refused to generate several prompts that included specific false claims, citing company policies around copyright and misuse. The findings arrive as concerns about AI-generated misinformation continue to grow. A recent fake news post circulating on social media, designed to resemble reporting from a major German broadcaster, was created using ChatGPT and briefly fooled users online. Legal experts cited in the investigation warned that producing highly convincing fake news with AI could have serious consequences. Depending on how the content is used, creators could potentially face claims ranging from defamation and reputational damage to violations involving forged digital documents. As AI tools become faster, cheaper, and more accessible, the challenge is making sure they don’t become some of the internet’s most effective misinformation machines.
