🧠 Your Brain on ChatGPT: What MIT's EEG Study Reveals About AI Writing

We've all experienced the relief of letting ChatGPT help us write — whether it's polishing emails, drafting essays, or brainstorming ideas. But here’s a serious question:
What’s happening to your brain while ChatGPT writes for you?
MIT researchers decided to find out. They attached EEG headsets to 54 university students and had them write essays across multiple sessions — some using ChatGPT, some using Google, and some with nothing but their own minds.

What they discovered isn’t just about writing. It’s about what AI use is doing to your cognitive engagement, memory, and creativity.Let’s walk through what the study found, what the brain scans revealed, and how you can use AI more wisely without losing your mental edge.

The Experiment: LLM vs Search vs Brain

Participants were split into three groups:
1. LLM Group – used ChatGPT (GPT-4o) only
2. Search Group – used Google (AI content blocked)
3. Brain-only Group – used no tools at all
Each person wrote three SAT-style essays across three sessions, then a fourth "flip" session: LLM-to-Brain users went tool-free, Brain-to-LLM users got to try ChatGPT

All participants wore EEG headsets to track neural engagement. Their essays were scored by human teachers and an AI judge. Afterward, they were interviewed for memory, satisfaction, and ownership.

Cognitive Debt:

The researchers coined a powerful concept: cognitive debt. Much like financial debt, cognitive debt builds up when we offload thinking to AI systems. You gain short-term convenience, but over time, your ability to deeply engage, recall, and create is weakened. This study offers some of the first direct neurological evidence of this tradeoff.

What the Brainwaves Showed :

The EEG scans revealed stark differences in how much participants' brains were "working" across the three groups.

Alpha/Beta Band Connectivity:
1.Brain-only participants showed the strongest neural network activity (alpha & beta frequencies linked to attention and memory).
2.Search users showed moderate connectivity.
3.ChatGPT users had the weakest connectivity by far.

Memory Breakdown: What Did They Remember?

After writing, participants were asked to quote something from their essays.
Results: 83% of ChatGPT users could not recall or quote anything they had written. 89% of Brain-only and 89% of Search users could recall correctly.Even more startling: 0% of ChatGPT users quoted accurately.

This demonstrates a significant breakdown in memory encoding when using LLMs — even just minutes after writing.

Who Felt Like They Wrote It?

Participants were asked whether the essay they wrote felt like theirs.

• Brain-only group: 89% said they felt full ownership

• Search group: Mixed — many felt partial ownership

• ChatGPT users: Reported the lowest sense of authorship

  The more people relied on AI, the more disconnected they felt from the outcome.

But Were They Happy With It?

Surprisingly, most ChatGPT users still felt satisfied with their work:

• 94% of ChatGPT users were happy with their essays

• 100% of Search users were satisfied

• 83% of Brain-only users were satisfied

This suggests a potential illusion of competence — a dangerous long-term effect in learning environments.

What Happened When They Switched Tools?

In Session 4, users flipped tools to test lasting effects.
LLM → Brain: These participants still showed low brain activity, as if their brains stayed “off” even without the tool.
Brain → LLM: Maintained higher neural activity, close to Search users.This shows that repeated reliance on AI may cause lasting disengagement, even when AI is removed.

Writing Quality and Language Patterns:

Natural Language Processing (NLP) analysis showed:
—> ChatGPT users’ essays were more homogeneous over time
—> Frequent reliance on default LLM phrases and structures
—> Less novelty and topic diversity

In contrast:
—> Brain-only essays were more original
—> Search users wrote more fact-rich, structured responses
Even though ChatGPT-generated essays scored high on grammar and cohesion, they lacked personal voice and conceptual diversity.

Final Thought: Use It, Don’t Lose It

AI makes writing easier. But easier isn't always better — especially when it comes to your brain.

Every time you offload thinking to a machine, you're building cognitive debt. That debt compounds over time, leading to shallower learning, weaker recall, and less agency over your own work.

Write first with your own mind. Let AI sharpen what you’ve already created — not replace it. Full paper: arxiv.org/abs/2506.08872

Thanks for reading,
— Mervin Praison