We are kind of screwed.

We, being everyone who has used AI in the past few years, asking for drafts and things from the very beginning and then editing the draft afterwards. Or asking AI to create your pitch perfect LinkedIn copy, or to write a business proposal or email. That we.

I spent the weekend reading “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task”, which is a research paper from the MIT Media Lab. There is a level of cognitive debt that’s created when you use AI, and throughout the study they discovered said debt between three different control groups.

The first group was an the LLM group, which used AI to write from the very beginning. The second a Search Engine group, which searched for sources to help them write. The third was a no-tool brain-only group, that didn't rely on any external tools to write.

They did 3 sessions with each group, with an optional 4th session, where people from the LLM group was asked to not use any tools when writing. They recorded each participants’ brain activity with an EEG to see how their brains were firing while doing each of the tasks.

"Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels." (Kosmyna et al., 2025)

The brain-only group had a high memory recall, worked fine and normally, which you’d kind of expect, with shorter essays but a high quoting ability. The Search Engine group had initially also a high quoting ability, meaning they were able to quote from sources they wrote. Later on there was less of a perceived ownership of their writing. They did the research for their essays basically.

Percentage of participants within each group who struggled to quote anything from their essays in Session 1  (Kosmyna et al., 2025)

The LLM group though repeatedly “focused on a narrower set of ideas”, with 80% not remembering sources that they used. There was less ownership of their work, and even if they didn’t use an LLM to write, they still continued writing similarly as if they used an LLM with AI-type vocabulary and patterns even a few weeks after using AI to write. As quoted by the study “Participants who relied solely on ChatGPT wrote less creatively, remembered less, and reported less personal connection to their work.” (Kosmyna et al., 2025)

In comparison, what was seen in the Brain only group is that when they write first, then use AI, then write again, their brains were firing normally in the EEGs. AI becomes a tool and not a crutch.

"...the directed connectivity analysis reveals a clear pattern: writing without assistance increased brain network interactions across multiple frequency bands, engaging higher cognitive load, stronger executive control, and deeper creative processing. Writing with AI assistance, in contrast, reduces overall neural connectivity, and shifts the dynamics of information flow. In practical terms, a LLM might free up mental resources and make the task feel easier, yet the brain of the user of the LLM might not go as deeply into the rich associative processes that unassisted creative writing entails." (Kosmyna et al., 2025)

Where does that leave us? In 2 years we will have ChatGPT babies going into kindergarten and entering the educational system. If we look at token usage through OpenRouter, we can see it drops sharply between June to July, which aligns with students heading to summer.

Tokens processed on OpenRouter from May 15th to August 12th (OpenRouter, n.d.)

Narrative wise, we need to change how we approach AI when it comes to content creation. For example, instead of relying on AI to create a first draft, use it to edit the first draft you create. Whether you’re looking at doing a social media copy, or writing a newsletter or email, you should always start off writing it by hand.

The other method is to not edit the content that AI generates, and instead train the model to write and sound like you as much as possible so there needs to be minimal editing. There are solutions for this, for example our StyleGPT tool that lets AI sound 95% closer to how you write. Or any of the generic Humanize AI tools that try to make AI sound more human instead. Or hire a third-party to constantly edit all of your AI generated content, so you don't get affected. (Dystopian but free market?)

As for students, there isn’t a good solution. What do you think?


Sources:

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv. https://doi.org/10.48550/arXiv.2506.08872

OpenRouter. (n.d.). OpenAI. OpenRouter. Retrieved August 12, 2025, from https://openrouter.ai/provider/openai