How does GenAI impact student writing and learning?
We routinely outsource tasks to technology, from dishwashers to Google Maps, without much concern. But educators know that encouraging deep understanding and long-term learning sometimes requires doing things “the hard way” before utilizing shortcuts: students use manipulatives to practice math before picking up calculators to ensure they fully understand fundamentals. Along similar lines, it’s important to consider how GenAI use impacts students’ writing processes and cognition, and to tailor GenAI policies and assignment design to encourage only those uses of GenAI that support learning.
AI, Learning, & Cognitive Offloading
Significant concerns have emerged about GenAI tools’ impact on students’ learning. Consider one large-scale, widely-reported study of over 20,000 Chinese schoolchildren: students who used AI on homework completed it more quickly and scored higher than those who didn’t, but ultimately earned significantly lower exam scores (Strömberg et al., 2026), suggesting the tools ultimately undermined learning. Concerns about AI tools’ impact on learning are compounded by the risks of hallucinations, which novices are ill-positioned to identify.
The story isn’t all bad. Research findings on the impact of GenAI tools on student learning are mixed, with some emerging research showing that AI learning tools can enhance learning (An et al., 2026; e.g., Kestin et al., 2025). Importantly, though, research showing GenAI benefits in learning often focuses on specialized software designed for educational contexts, where reference content and GenAI roles have been carefully tailored to maximize learning (Bauer et al., 2025). (As of fall 2026, UMN technology services have not approved any of these specialized AI tools.) Learners must remain “active agents in the learning process” to see significant learning benefits of GenAI tool use (An et al., 2026).
In contrast, general-purpose chatbots like those approved for use at UMN are more likely to provide confident answers than to support inquiry, which can encourage learners to offload their thinking to GenAI tools rather than use them to augment their efforts. Cognitive offloading—outsourcing a mental task to an outside technology—can negatively impact students’ learning when they forgo tasks that would help them learn. Generally, technology companies, responsive to user preferences, are pursuing GenAI technologies designed to create efficiency by eliminating human labor, rather than augmenting it, as shown by the emergence of agentic GenAI tools.
Multiple studies have illustrated the risks that overreliance on GenAI tools poses for learning and cognitive engagement. For instance, Gerlich (2025) found that participants who relied more on GenAI tools showed reduced critical thinking ability, likely due to cognitive offloading. Cognitive offloading is also associated with more superficial information processing and lower analytical effort (Gerlich, 2025). Users relying on a GenAI tool often default to uncritically adopting its recommendations (Bauer et al., 2025; Beck et al., 2026; Rosbach et al., 2026) or display an “anchoring effect,” where the range of responses they consider is subconsciously limited by the GenAI’s suggestions (Rosbach et al., 2026). Outsourcing and anchoring are especially likely when users are under time pressure or facing an otherwise high cognitive load (Rosbach et al., 2026)—common states for college students!
Conversely, many traits and habits buffer against excessive cognitive offloading, including expertise on the topic (Rosbach et al., 2026); AI literacy (Tian et al., 2025), prior knowledge of AI (Tully et al., 2025), skepticism about AI (Beck et al., 2026); critical thinking skills (Hou et al., 2025); and intrinsic motivation to learn about the topic (Cui et al., 2026).
AI in the Writing Process: Possibilities & Consequences
Research in writing studies shows that writing is a process of thinking—a crucial site for the cognitive effort essential to learning (Bazerman & Tinberg, 2015). Unlike most published writing, most student writing tasks focus on the writing process and the learning it supports, not merely the product such effort produces. It’s important, then, to consider the potential risks of offloading any given stage of the writing process to GenAI tools.
Although “AI use” is sometimes discussed monolithically, the degree and risk of cognitive offloading varies significantly depending on when and how a user employs a GenAI tool. Consider these questions in determining the extent and risk of cognitive offloading:
- What are the relevant learning outcomes? Students should spend the most time practicing skills that are central to the course. For example, if your course focuses primarily on content in your discipline, students will miss important opportunities to engage with content if they rely on GenAI to brainstorm or generate ideas, but using GenAI for translation or editing might be unproblematic. Conversely, if your course focuses on sentence-level editing, students might benefit from generating initial error-ridden drafts with GenAI to correct, but you probably wouldn’t want them to use GenAI to polish.
- What stage of the writing process are students in? Generally, using GenAI early in the process invites greater cognitive offloading than using GenAI later in the writing process. When humans think independently before using GenAI, they are likely to engage more critically with its outputs. (Notably, this is a big downside of the common approach of allowing GenAI for brainstorming but not final drafting: even if a student wrote every word of a text, their thinking about the content may be superficial.)
- What is the role of GenAI? Although we tend to picture students simply copying and pasting from GenAI, sophisticated users adopt more varied strategies. Mollick & Mollick (2023) outline seven roles GenAI tools might play in educational settings, most of which invite much less cognitive offloading than typical GenAI usage.
- Can GenAI successfully complete this task? Unless you’re deliberately generating text with errors to practice catching errors, GenAI assistance won’t be helpful if it can’t consistently complete the task successfully. Novices typically don’t have the relevant content knowledge required to critically analyze and ultimately reject bad outputs.
| Writing Process Stage | Low Cognitive Offloading Use of GenAI | High Cognitive Offloading Use of GenAI |
|---|---|---|
| Brainstorming | The student freewrites or concept maps, then prompts GenAI to coach: Read my initial freewrite and ask me three open-ended questions to prompt further thinking. | GenAI prompt: Suggest ten possible topics for this assignment. |
| Research | The student creates a list of research questions, then prompts the GenAI with a specific question: Suggest keywords I should use in academic database searches to find research in genomics related to these research questions. | GenAI prompt: Create a research question for a first-year writing paper on [topic]. GenAI prompt: Suggest six sources related to my topic. |
| Reading | The student reads and summarizes an article, then generates a GenAI summary to check their work. | GenAI prompt: Summarize this article and suggest two discussion questions about it. |
| Data Collection | The student writes a survey draft, then prompts the GenAI: Review this survey draft and suggest plain language edits that would make it more accessible to an audience of first-year, traditional-age college students. | GenAI prompt: Create a survey for first-year students about their experiences living in dorms. |
| Data Analysis | The student organizes their data and identifies an appropriate statistical test, but asks GenAI to generate R code to run that test: Suggest the steps and R code to run a MANCOVA on the attached data. Independent variable: test_condition, dependent variable: test_score, covariates: student_gpa, hours_studied The student selects an appropriate data visualization type and prompts GenAI for directions to use the software: Tell me the steps to create a stacked bar chart in Excel, change the colors, and add data labels. | GenAI prompt: Identify trends in these data. GenAI prompt: Create a visualization that best represents the attached data. |
| Drafting | The student prompts the GenAI to respond as a tutor: You are a tutor who helps students with college writing. Engage me in a conversation with lots of non-leading questions that will (1) help me assess whether this outline is clear, (2) help me revise my outline as needed, and (3) get me motivated to start writing. | GenAI prompt: Outline this essay. GenAI prompt: Write a draft of this essay. |
| Revision | The student prompts the GenAI to respond as a reader: You are a patient newly diagnosed with diabetes. You have only a high school education. Read this draft pamphlet, and tell me about your experience reading it. What is clear and what is unclear? How would you respond to it? The student assesses the response and makes revisions based on their intended impact on their audience. | GenAI prompt: Tell me three ways I should change my draft to improve my grade. GenAI prompt: You are a first-year college student. Revise this draft of a SOAP note for a dental hygiene course to meet expectations in the discipline. |
| Editing | The student prompts GenAI to help them improve their editing skills: The student uses GenAI to generate practice opportunities: Create a sample text with common grammatical errors. I’ll correct it and ask you to check my work. | GenAI prompt: Correct the grammar errors in this text. GenAI prompt: I’m studying for a test in my journalism course. What grammar errors might I see on my quiz? |
| Citation | The student writes citations and prompts GenAI to check for consistency with a style guide. (Better option: use a citation generator, which also saves work without introducing the risks of GenAI tools in a very specific process.) | GenAI prompt: Create citations for the attached articles. |
| Translating | The student generates a translation of a section of their text—Provide a translation of this paragraph while maintaining meaning and tone as closely as possible—and manually checks it for accuracy. | GenAI prompt: Translate this paper from English to Spanish. |
| Coding | Students in an advanced course outsource simple steps they can already perform confidently, but complete high-level tasks themselves. | Students in an introductory course have GenAI generate the kinds of code they are learning to complete problem sets. |
Ultimately, decisions about where students can use GenAI tools in your courses should be grounded in an understanding of the capacities of GenAI tools, the potential impacts of GenAI use on students’ cognition and learning, and your course learning outcomes. Once you’ve made these decisions, you’re ready to develop your GenAI policy, design assignments that thoughtfully incorporate GenAI tools, and set up class structures that discourage unauthorized GenAI use.
Graphic credit: The Economist, based on data from Strömberg et al.