AI chatbots exhibit significant editorial bias and surface Russian disinformation when retrieving news, according to a study published Wednesday by Agora Digitale Transformation, a Berlin-based think tank.
The analysis found that ChatGPT leans heavily on journalistic sources but cites Die Welt nearly five times more often than Deutschlandfunk, a pattern researchers linked to a 2023 global partnership between OpenAI and Axel Springer, Die Welt's parent company. Claude and Gemini drew roughly half their responses from journalistic outlets. However, Claude accessed domains tied to the Pravda network, a Russian influence operation that has distributed over 8.5 million articles, on seven occasions — effectively presenting propaganda alongside legitimate reporting.
The researchers warned that such distortions are difficult for users to detect, raising concerns about the reliability of AI-powered news aggregation as a growing information channel.