A university study links the use of ChatGPT to a higher likelihood of young people sharing false information

Researchers from the School of Communication at the University of Navarra have analyzed the relationship between the use of ChatGPT, trust in artificial intelligence, and behaviors related to misinformation in the educational community. The study research sample reveals a gap between familiarity with these tools and critical evaluation of them: students are the ones who use ChatGPT the most and who trust its answers the most, but they are not the ones who verify the information they receive the most.

The study by Xavier Bringué and Charo Sádaba, professors at the School of Communication, has been published in the scientific journal adComunica. It is based on an online survey conducted as part of the 6th Study on Screen Time and GAD3 in Spain in January 2025, surveying 922 participants, including students, families, and teachers from Education Secundaria, high school diploma and training Professional.

The analyses also reveal a more nuanced dynamic than might appear at first glance: while frequent use of ChatGPT is associated with a higher likelihood of having shared false information, it is the trust placed in the tool —not its use—that reduces the tendency to verify information. Meanwhile, trust in social media emerges as the strongest predictor of having shared false information.

Differences Among Profiles in Terms of Usage, Trust, and Verification

The study compares the use of ChatGPT, the level of trust placed in it, verification practices, and the perceived ease of detecting errors among the three profiles within the educational community.

In terms of usage of tool, student body ranks well above families and faculty, with statistically significant differences compared to both groups. In turn, faculty reports significantly higher usage than families.

The level of trust placed in ChatGPT mirrors part of this patron saint: student body sample significantly higher levels than families and faculty, while there are no significant differences between these last two groups.

The picture changes when we look at verification practices: it is the “ faculty ” who report a higher frequency of cross-checking information, with significant differences compared to both the “ student body ” and families, who do not differ from one another on this point.

Finally, in terms of perceived ease of detecting errors, faculty and student body received the highest scores, while the group that reported the greatest difficulty in identifying incorrect or misleading information generated by AI was group .

The Relationship Between Usage, Trust, Verification, and Error Detection

The study also analyzes how these variables relate to one another. It finds a association between ChatGPT use and trust in tool, as well as a relationship between use and the frequency of verification, and a moderate, positive association between use and the perceived ease of detecting errors.

Trust in ChatGPT is also positively associated with the perceived ease of detecting errors, but shows no sample significant relationship with verification when analyzed in isolation. However, by constructing a model that controls for all other variables, the authors find that greater trust in the tool does significantly reduce the tendency to verify information, while greater use of it slightly increases that tendency. The researchers interpret this result as a sign of overconfidence: the more familiar the Username is with ChatGPT, the more they believe they know when the tool is wrong, which may reduce their need to cross-check the received response with external sources.

Spreading False Information

One of the study’s key findings is the link between the use of ChatGPT and the spread of misinformation. According to data, student body is the group where the highest proportion of users reported having shared misinformation, followed by faculty and, to a lesser extent, families.

Logistic regression models confirm that the use of ChatGPT as an information source is positively and significantly associated with the probability of having shared false information, even after controlling for the effect of trust in social media and basic sociodemographic variables. In contrast, neither fact-checking practices nor the perceived ease of detecting errors proved to be significant predictors in this model.

The most striking finding, however, is that the strongest predictor of having shared false information was not the use of ChatGPT, but rather trust in social media. The authors conclude that the risk of misinformation does not lie solely in the use of generative artificial intelligence, but rather in the ecosystem of informational trust—and especially in the trust placed in social media—within which that use takes place.