Is ChatGPT Rewiring Your Brain? New Studies Raise Concerns About Cognitive Offloading and Language Change

Is ChatGPT Rewiring Your Brain?

Artificial intelligence has spread rapidly in recent years. Tools such as Microsoft Copilot promise help with planning and brainstorming, while ChatGPT is widely used for homework and professional tasks. Some people avoid these systems on principle, while many others welcome the efficiency they offer. A growing body of research, however, asks whether reliance on AI assistants may change how we think, learn and use language.

What the 2025 EEG study found

A 2025 preprint on arXiv, Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, examined brain activity in participants who wrote essays across multiple sessions under three conditions: unaided (brain-only), using internet search engines, and using a large language model (LLM). Researchers measured neural activity with electroencephalography (EEG).

The study reported a consistent pattern across three sessions: the unaided group showed the strongest EEG signals associated with cognitive engagement, the search-engine group showed intermediate signals, and the LLM group showed the weakest neural activity. In a fourth session, participants switched between the brain-only and LLM-only conditions; those who moved from unaided writing to LLM use showed some short-term gains in memory and activity, while participants who switched from LLM use back to unaided writing underperformed relative to peers.

The authors also reported behavioral findings: LLM users were less likely to feel ownership of their text and had more difficulty accurately recalling or quoting passages they included. Compared with the other groups, LLM users scored lower on several language and behavioral measures and showed weaker indicators that the authors interpreted as reduced neural connectivity or engagement. The paper also noted concerns about the reliability of some AI-enhanced search features.

Language change linked to AI

Separate research published as Model Misalignment and Language Change: Traces of AI-Associated Language in Unscripted Spoken English analyzed more than 22 million unscripted words from online sources and podcasts. The team found a post-ChatGPT increase in vocabulary and phrasings that are commonly favored by LLMs — examples highlighted include words such as "meticulous," "strategically," "garner," and "surpass." The authors raise the possibility that AI-recommended phrasing could influence everyday speech and reduce some individual stylistic variation.

Implications for students and educators

Psychologists and educators warn that overreliance on AI could short-circuit key learning processes, especially for students. When tools do thinking for learners, opportunities to make errors, reflect, and strengthen memory and problem-solving skills are reduced. To guard against dependence, experts recommend teaching foundational cognitive skills such as:

• Meta-learning — learning how to learn effectively.
• Cognitive flexibility — adapting strategies and forming new mental pathways.
• Self-efficacy — building confidence that effort improves ability.

Balanced take and next steps

These findings are thought-provoking but preliminary. The EEG study is a preprint (not yet peer-reviewed) and, like all single studies, has limitations in scope and generalizability. The observed patterns are consistent with the concept of cognitive offloading — using external tools to reduce mental effort — but do not by themselves prove long-term brain damage. More independent, longitudinal research is needed to confirm lasting effects and to identify which learning contexts are most at risk.

Meanwhile, educators, students and professionals can benefit from using AI as an aid rather than a replacement for active thinking: use assistants to brainstorm and draft, but practise retrieval, revision, and explanation without AI so critical skills remain strong.

Key sources: arXiv preprints titled "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task" and "Model Misalignment and Language Change: Traces of AI-Associated Language in Unscripted Spoken English."