Writing With AI Leads to Less Thinking, Researchers Show

June 30, 2025, 2:09 pm | Read time: 6 minutes

ChatGPT has become an indispensable part of daily life for many. Whether it’s summarizing texts, sorting out relationship issues, or generating the latest recipes, this AI is up for the task. It provides quick, convenient, and surprisingly convincing answers to almost anything. But what actually happens in our minds when we increasingly rely on artificial intelligence for thinking, writing, and problem-solving? A new study from the U.S. reveals that those who regularly write with AI might be doing more harm than good to their brains.

The use of AI tools like ChatGPT in schools, universities, and daily life is rapidly increasing–especially in writing. But how does the use of such systems change our thinking, memory, and relationship with our own text? A study from the Massachusetts Institute of Technology currently provides fascinating insights into this. Using brainwave measurements, language analyses, and interviews, researchers have comprehensively examined for the first time how AI support in writing affects cognitive processes and learning experiences.

Follow the FITBOOK channel on WhatsApp now!

What and why was studied?

The researchers wanted to know if writing with the help of AI changes thinking. They examined three groups:

A group that worked with an AI writing assistant like ChatGPT (Large Language Model)
A group that gathered content through traditional internet research
And a group that wrote entirely without aids–using only their own knowledge

The goal was to measure the cognitive consequences of using such systems–specifically the effects on attention, memory, language, and the feeling of being the author of one’s own text. The researchers were particularly interested in whether regular AI use creates a “cognitive debt”–a gradual decline in mental self-performance.

Study design and methods

The study lasted four months and included 54 participants. Each person initially wrote three essays–always under the same condition (with or without aids). In a fourth session, the conditions were then switched:

Those who previously wrote with an AI writing assistant like ChatGPT had to work without it
Those who previously worked without aids were given access to an AI writing assistant like ChatGPT

During writing, brain activity was measured using electroencephalography–a method where brainwaves are recorded via electrodes. The focus was on specific frequency ranges of the brain, such as the alpha and beta bands, which are associated with concentration, memory retrieval, and mental activity.

The essays were also analyzed linguistically–including sentence patterns, technical terms (so-called ‘Named Entities’), and typical expressions. The texts were evaluated by human teachers and a specially developed AI evaluation system. Interviews with participants provided additional insights into self-perception and memory ability.

What is the result of the study?

The brainwave measurements showed a clear connection: The stronger the technical support, the lower the measured brain activity.

Participants who wrote without aids showed the strongest and broadest brain activity.
Those who worked with a search engine were in the middle range.
Users of an AI writing assistant showed the least activation–especially in the alpha and beta areas of the brain, which are important for attention, memory performance, and information integration.

In other words: The more technical support–the less mental self-performance.

Linguistic Differences: Less Variety Through Large Language Models (LLM)

The content of the essays also showed clear differences:

The texts from the LLM group (e.g., ChatGPT) were often linguistically smooth but not very individual–they frequently contained similar phrases and followed a typical AI pattern.
The texts from the “Brain only” group appeared more creative and independent, even if they were sometimes less perfect linguistically.
The search engine group was again in between: better structured than completely without aids, but with more self-performance than the LLM group.

Memory and Quoting Ability: Those Who Write with AI Remember Less

A central result of the interviews:

LLM users often could not remember their own phrases, even though they had written them just a few minutes earlier.
Correctly quoting their own texts was also more difficult for them.
The “Brain only” group, on the other hand, had significantly better memory scores and could usually accurately reproduce their statements.

This suggests that writing with AI reduces the active processing of content–and thus also the anchoring in memory.

Impact on the Feeling of Being Responsible for the Text

Many participants commented on the perceived responsibility for their text:

In the LLM group, the feeling of having “truly written” the text themselves was the lowest.
The search engine group felt somewhat more involved.
The group without help had the highest level of text responsibility–for them, it was clear: The text belongs to me.

Fourth Session: What Happens During the Switch?

The condition switch in the fourth session was particularly revealing:

Those who previously worked with AI and now had to do without showed significantly lower brain activity–a sign of underchallenge or habituation.
Conversely, participants showed more activation when they initially had to do without help and then could use an AI language model–compared to using it from the start.
The sudden switch to AI apparently led to more cognitive engagement with the output–the brain worked more actively again.

Why are these results important?

The study shows: Those who regularly rely on AI for writing could gradually reduce their mental self-performance–without realizing it. While tools like ChatGPT help achieve results quickly, one’s own concentration, memory, and sense of language measurably suffer.

This presents a clear challenge for the education sector:

How can AI be meaningfully integrated without weakening learning?
And how can students learn to use such tools responsibly?

The topic is also highly relevant for teachers, school authorities, and universities. Because: Those who no longer work on texts themselves may forget how to understand them–or even retain them at all.

Limitations of the study

As well-founded as the study is, there are limitations: It is a preprint, meaning it has not yet been peer-reviewed. The participant group was relatively small, with 54 people. The exact use of AI (e.g., what inputs were made) was not standardized–differences in approach could have influenced the results. And: Four months is a considerable time, but not long enough to clearly demonstrate long-term brain changes.

Brain Activity Measured Writing with ChatGPT Leads to Less Thinking

Physical limitations often don't occur alone; those affected frequently also experience mental health issues, sometimes even depression.

Pectus Excavatum What Is a Funnel Chest–and How Dangerous Is It?

FITBOOK Editor Shares Her Story How I Rediscovered Running as a Mom

Conclusion

The MIT study provides an important impetus for thought: AI can make many things easier–but also mentally disconnect. Especially with prolonged use, memory can suffer, mental self-activity can decline, and the sense of one’s own text can disappear. Those who think they are saving time by writing may risk a gradual mental disengagement. The central question remains: How can AI help with learning–without us forgetting how to think for ourselves?

This article is a machine translation of the original German version of FITBOOK and has been reviewed for accuracy and quality by a native speaker. For feedback, please contact us at info@fitbook.de.