ChatGPT, Claude, Gemini, and Grok are all bad at crediting news outlets, but ChatGPT is the worst (at least in this study)

Canadian researchers have conducted a study examining how four AI models—ChatGPT, Claude, Gemini, and Grok—handle crediting news outlets when discussing Canadian news events. The study, which compared both paid and free versions of the models, aimed to determine whether these AI systems would provide proper attribution to individual news sources. The findings reveal that AI models are generally poor at citing news sources unless explicitly asked to do so, and among the tested models, ChatGPT emerged as the worst offender in this regard.

The research involved testing the four AI models on 2,267 real Canadian news stories, both in English and French, without enabling web search capabilities. The results showed that all models demonstrated extensive knowledge of Canadian current events, indicating that they had systematically ingested Canadian journalism. However, despite this deep understanding of domestic politics, provincial affairs, and local reporting, the models rarely provided any source attribution in their responses. Specifically, 92% of knowledgeable responses from the models failed to credit any news outlet.

When web search was enabled, and the models were tested on 140 specific articles through each company's API, the results remained consistent. Every model produced responses that covered enough of the original reporting that many consumers would have little incentive to visit the source material. Notably, the models often linked to Canadian news sites, further highlighting their reliance on these outlets for information.

Taylor Owen, Beaverbrook chair in media, ethics, and communication at McGill University and a coauthor of the study, explained that the AI models' lack of attribution is a significant concern. Owen noted that while the models have absorbed Canadian journalism, they rarely disclose the origin of the information they provide. This issue has been further underscored by a lawsuit filed in November 2024 by Canada's CBC, Globe and Mail, Toronto Star, Postmedia, Metroland Media, and The Canadian Press against OpenAI for copyright infringement. The case, which is the first of its kind in Canada, is currently ongoing.

The study's authors, Owen and Aengus Bridgman, an assistant professor at McGill, emphasized that their findings highlight a critical need for AI models to improve their ability to credit news sources accurately. As AI systems continue to play a growing role in shaping public discourse and information dissemination, the lack of proper attribution poses a threat to journalistic integrity and the credibility of the information shared by these models.

In conclusion, the Canadian research has revealed that AI models, particularly ChatGPT, are not adept at crediting news outlets when discussing Canadian news events. This shortcoming not only undermines the importance of acknowledging journalistic contributions but also raises concerns about the potential for misinformation and copyright infringement. As AI technology advances, it is crucial for developers and stakeholders to prioritize the integration of accurate attribution mechanisms to ensure transparency and respect for the work of journalists.