AI tools such as ChatGPT and Gemini are becoming part of our personal, academic, and professional lives. Using AI for writing is one of the most common uses of generative AI in the workplace (Bick et al., 2024; Chatterji et al., 2025). AI-assisted writing involves asking AI to help develop ideas, organize what we want to say, produce a draft, or revise a piece of writing. For example, an employee might give AI meeting notes and ask it to summarize decisions and responsibilities. A student might ask it to suggest essay arguments, draft paragraphs, or identify weaknesses in their reasoning. Within seconds, AI can produce clear, coherent text. But what happens to our thinking when AI does this work for us?
When we use tools to reduce the mental effort in doing a task, we are engaging in a process called “cognitive offloading” (Risko & Gilbert, 2016). We do this whenever we write a shopping list instead of remembering every item or set a calendar reminder instead of keeping an appointment in mind. These tools can help us manage tasks that would otherwise place heavy demands on our memory and attention. However, people have long worried that relying on technology could weaken their own thinking skills. In the past, similar concerns have been raised about the use of calculators in schools (National Research Council, 2001) and the prevalent use of search engines like Google in everyday life (Carr, 2008). This question is particularly relevant to writing, where some of the mental effort we save may be part of how we develop our ideas and understanding.
When AI produces the text, which parts of our thinking are we handing over to it? To answer this question, we must first understand the mental processes involved when we engage in the act of writing without AI. Writing involves more than putting fully-formed thoughts onto a page. We must plan what we want to say, retrieve the relevant knowledge from our long-term memory, connect ideas, and organize them for the reader (Flower & Hayes, 1981; Hayes, 2012). Expressing those ideas also requires choosing words that convey the intended meaning and arranging them into grammatical sentences (Kellogg, 1996). These activities overlap and repeat, rather than following a fixed sequence. Writers generate stretches of text, evaluate them, and move back and forth between planning, composing, and reviewing. For example, searching for the right word may help us clarify what we mean, and constructing a sentence may reveal that a relationship between ideas is unclear. Writing therefore helps us develop and revise our own thoughts beyond communicating them to others.
Because writing involves these different iterative mental processes, using AI to help us write can mean handing over different kinds of thinking, depending on which parts of writing we ask AI to handle and how actively we engage with its written output. Consider a student writing an essay for a class assignment. At one extreme, student A asks AI to write the entire essay and submits it after a quick read. Much of the work of generating ideas, weighing arguments, organizing them, and choosing words has been handed over to AI, without reviewing the written output. At the other extreme, student B writes the essay themselves and asks AI to identify unclear explanations or weaknesses in the argument. They decide which suggestions to accept or reject and revise the essay accordingly. In between these two extremes, student C might develop their own arguments and supporting reasons, ask AI to produce an outline, and then ask AI to produce the first draft. How much further thinking student C engages in depends on how carefully they evaluate and revise the draft.
Research is beginning to examine what people remember and can do independently after using AI to help them write, with evidence of both drawbacks and learning benefits. In one study on essay writing, those who wrote an essay with ChatGPT’s help had more difficulty accurately quoting their own text and reported less ownership than those who used a search engine or no tools (Kosmyna et al., 2025). In another study of Chinese university students learning to write essays in English (Chen, 2026), students in one group were allowed to use ChatGPT for suggestions and feedback on writing an essay, but could not ask it to draft paragraphs or provide complete arguments. They also had to explain why they accepted or rejected ChatGPT’s suggestions. Another group of students could ask AI to draft paragraphs and develop arguments for them. On a later writing task without AI, the first group scored higher on writing quality, argument depth, and higher-order thinking than the second group, although the evidence was tentative. This suggests that greater cognitive offloading of ideas and reasoning during the initial writing task was associated with poorer subsequent higher-order thinking performance.
While these studies point to possible drawbacks of AI-assisted writing, others have found that using AI can improve people’s ability to write independently. Lira et al. (2026) investigated whether using AI during writing practice would help or hinder people’s ability to write independently afterward. Participants first received guidance on effective writing, then practiced improving a cover letter either with or without an AI writing tool. Those who practiced with AI exerted less mental effort on the task but subsequently performed better when revising another cover letter without the help of AI. A further experiment found that simply studying an AI-revised cover letter produced similar benefits, suggesting that seeing how AI improved the text helped participants learn professional writing skills. These findings suggest that reduced mental effort does not necessarily always mean reduced learning.
These findings give us a reason to look beyond whether AI makes writing easier and consider what we want to gain from the task. For anyone using AI to write, a useful starting point is to ask: Are we simply trying to complete a writing task, or are we trying to develop our own understanding and skills? Saving time on writing a simple email may be valuable. But when the goal is to learn how to write and think critically, the time spent on planning and deciding what to say and how to say it is part of the value of writing. Using AI does not have to make us “lazier, less smart, and learn less,” as many people fear (Lira et al., 2026). As AI takes on more of our writing tasks, we need critical thinking skills to recognize its errors, evaluate its arguments, and know when and how to revise what it has written. By making these judgments ourselves, we can remain in charge of our writing as AI supervisors, rather than having AI replace our thinking.
References
Bick, A., Blandin, A., & Deming, D. J. (2024). The rapid adoption of generative AI (Working Paper No. 32966). National Bureau of Economic Research. https://doi.org/10.3386/w32966
Carr, N. (2008, July/August). Is Google making us stupid? The Atlantic. https://www.theatlantic.com/magazine/archive/2008/07/is-google-making-us-stupid/306868/
Chatterji, A., Cunningham, T., Deming, D. J., Hitzig, Z., Ong, C., Shan, C. Y., & Wadman, K. (2025). How people use ChatGPT (NBER Working Paper No. 34255). National Bureau of Economic Research. https://doi.org/10.3386/w34255
Chen, X. (2026). Layer-sensitive cognitive offloading in generative AI-assisted writing: Supported performance and independent no-AI outcomes. Frontiers in Psychology, 17, Article 1906199. https://doi.org/10.3389/fpsyg.2026.1906199
Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32(4), 365–387. https://doi.org/10.58680/ccc198115885
Hayes, J. R. (2012). Modeling and remodeling writing. Written Communication, 29(3), 369–388. https://doi.org/10.1177/0741088312451260
Kellogg, R. T. (1996). A model of working memory in writing. In C. M. Levy & S. Ransdell (Eds.), The science of writing: Theories, methods, individual differences, and applications (pp. 57–71). Lawrence Erlbaum Associates.
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 [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2506.08872
Lira, B., Rogers, T., Goldstein, D. G., Ungar, L., & Duckworth, A. L. (2026). Coach not crutch: Evidence that AI can improve writing skill despite reducing effort (Version 4) [Preprint]. arXiv. https://arxiv.org/abs/2502.02880v4
National Research Council. (2001). Adding it up: Helping children learn mathematics (J. Kilpatrick, J. Swafford, & B. Findell, Eds.). National Academy Press. https://doi.org/10.17226/9822
Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
Author
Asst. Prof. Suphasiree Chantavarin, Ph.D.
Head of Cognitive Psychology Area
