Communication in Media and Society
DOI 10.55206/WHCO8050
Tihomira Doncheva
Sofia University St Kliment Ohridski
E-mail: tihomiradd@uni-sofia.bg
Abstract: Factchecking is among the key mechanisms for addressing information disorder. In the context of the growing integration of artificial intelligence, particularly large language models, into the information environment, the need for reliable information is increasing while access to it is more challenging. Thus, generative artificial intelligence can be both – an asset and a risk to factcheckers. The present study aims to examine the extent to which Claude.ai and ChatGPT can factcheck claims in Bulgarian. The experiment was conducted using a free skill developed by Nauka.bg. The claims were selected from the “Verified” section of Factcheck.bg, the largest fact-checking platform in Bulgaria. Through comparative content analysis, the study found that in most cases both models correctly identify false claims in accordance with Factcheck.bg’s publications. Claude.ai demonstrates a more authoritative tone and provides concise explanations supporting its verdicts. ChatGPT is more hesitant but offers greater detail, precision, and clarity in its explanations. Notably, ChatGPT rarely references its sources, whereas Claude.ai occasionally links to unreliable sources. A positive trend is Claude.ai’s note to consult official factchecking organizations and verified information sources.The results confirm the hypothesis that artificial intelligence can assist factcheckers but cannot replace human expertise in the verification process.
Keywords: disinformation, factchecking, artificial intelligence, large language models, ChatGPT, Claude.ai.
Rhetoric and Communications Journal, issue 69, October 2026
