A pre-print from MIT Media Lab suggests extensive reliance on OpenAI’s ChatGPT can sharply reduce neural engagement during writing tasks. Researchers asked 54 undergraduate students to compose three 20-minute SAT-style essays over a four-month period, alternately using no digital aid, Google Search and ChatGPT. Throughout the experiment, participants wore electro-encephalogram headsets that monitored activity in 32 brain regions. The cohort that used ChatGPT showed the lowest overall brain activity, with EEG data indicating about a 50 % drop in connectivity compared with those writing unaided. More than 80 % of ChatGPT users were unable to recall or recognise excerpts of the essays they had just produced, and two independent English teachers judged their submissions to be less original and largely formulaic. By the final session many ChatGPT users copied material directly from the chatbot, while the unaided group displayed the highest levels of creative and semantic processing. When roles were reversed, students who had previously relied on ChatGPT struggled to write without it, whereas those gaining first-time access to the tool maintained strong neural engagement. Lead author Nataliya Kosmyna said the findings raise concerns about “cognitive debt”, particularly for developing brains, and called for guidelines on how generative AI is introduced in classrooms. The study has not yet undergone peer review and its small sample means broader conclusions remain tentative, but it adds empirical weight to calls for measured use of large language models in education and the workplace.