A new study from the Massachusetts Institute of Technology has found that relying heavily on AI chatbots to verify information may erode critical-thinking skills over time, even as users feel they are getting sharper at spotting falsehoods.
“When we’re interacting with AI, we feel we’re becoming better at certain tasks and there’s enough research that shows we are not,” said Anku Rani, a PhD student at MIT and co-lead author of the study.
During the four-week study, released in April, researchers tracked 67 participants and quizzed them on whether pairs of news-related headlines and images were real. Participants used an AI assistant that runs on GPT-4o and is integrated with Google search. The chatbot could offer clues — in one example, it advised a user to examine a police badge that revealed an image was fake.
The study found a trade-off. AI helped participants make the right call 21% more often when they used it. But when reviewing new images without the chatbot’s help, their accuracy dropped 15.3% by the experiment’s fourth week. “These results indicate that while AI may help immediately, it may ultimately degrade long-term misinformation detection abilities,” the study noted.
About one-quarter of participants said they believed their detection skills were improving, even as their test scores declined, according to the study. Researchers also found that users who relied on AI systems that simply told them what to do often “go along with the system because it sounds knowledgeable.”
The study’s authors said an AI system’s approach matters. Chatbots that are more prescriptive — giving direct answers — appeared to worsen users’ independent judgment, while a more probing, guided-questioning style helped maintain critical thinking. That finding echoes prior research documented by Main Street Independent. An April study reported that AI chatbots from multiple companies were found to flatter users and push harmful advice, and a March study in the journal Science similarly reported that chatbots flatter users and can recommend dangerous information.
The MIT study is not the first to find that technological shortcuts can weaken underlying skills. A 2025 Lancet study found that doctors who used AI classification tools to detect cancer eventually became worse at doing so on their own. A neuroscientist at the Possibility Institute, a metascience research group, has also warned that outsourcing thinking to AI could weaken the brain’s defenses against dementia.
The study’s authors acknowledged limitations. Participants were predominantly from the US and UK, and researchers said a more diverse sample could indicate whether skill degradation occurs across cultural contexts and educational systems. Longer studies could also show whether the effects of AI over-reliance continue at the same rate.
The researchers said their findings are particularly relevant for educators who increasingly depend on AI tools for learning. “As AI becomes increasingly sophisticated, ensuring these tools build critical thinking skills rather than cognitive dependency becomes essential for maintaining public resilience to misinformation,” the study noted.