Student Use of and Views on GenAI for Writing

Synthesis: Kuznetsov, Sheely and Baker (2026) surveyed 504 sociology students and interviewed 12 more to find out how students actually use generative AI for coursework and writing. Two thirds use it, but selectively — asking follow-up questions about course content, summarizing readings and generating practice questions dominate, while generating assignment text is rare. The sharpest finding is a guidance gap: 81 percent of students had been given some instructions about AI use, yet only 46 percent found them very clear, and about one in four reported no guidance or unclear guidance. The authors argue the fix is not detection but explicit, course-level policy paired with Scaffolding that removes the pressures pushing students toward AI.

Overview

Writing has always been central to sociology teaching, and the scholarship of teaching and learning has a long record of work on how students develop it. Generative AI complicates that work in a specific way: students can now produce competent-looking prose without doing the thinking the assignment was designed to elicit, and instructors cannot reliably tell the difference. But the authors note that much of the resulting debate has run ahead of evidence about what students are actually doing.

The study answers three questions: what students think about using GenAI in coursework, both for students generally and for themselves; how they are using it; and how well professors address AI use in course design and in the evaluation of written work. The setup is deliberate: views and behavior are treated as linked, since a student who thinks AI use is broadly acceptable may still avoid it personally, and a student who uses it heavily may be uneasy about doing so. The findings are then read through a writing education lens, asking what instructors should change rather than what students should stop doing.

Study Design & Method

An exploratory mixed-methods design combining a large survey with a small number of interviews, at a public research university in Canada.

• Survey. Administered as part of an annual student survey run by a research center, in two terms of the 2024–2025 year: 465 students enrolled in the winter term and 379 in the summer (844 invitees), yielding 504 respondents who completed the GenAI questions (273 winter, 231 summer). Participation earned a course bonus point regardless of consent. The 2024–2025 topical module covered student views on and use of GenAI and the extent to which professors provided guidance on acceptable uses.
• Question design. After being asked whether they had used GenAI, respondents were randomly assigned to open-ended or closed-ended items. The closed-ended version listed ten possible uses to check off. Open-ended prompts asked what they used GenAI for, what schoolwork it helped with, what concerns they had, and what they considered its proper role in coursework — 920 open-ended responses in total, from a few words to a paragraph, which the authors also coded into the closed-ended categories and used to validate the interview findings.
• Focus groups and interviews. Students in two required theory courses for majors and specialists in Sociology and in Criminology, Law, and Society were invited in April 2025; of 157 contacted, 12 participated across four focus groups of two to three students each and two one-on-one interviews, held April–May 2025 and lasting 50 minutes to two hours. Most were second-year students (75 percent), matching typical program entry. Interview topics covered high school writing ability, writing supports used, writing process and challenges, AI thoughts and use, and students' recommendations to instructors. Transcripts were produced with Otter.ai, cleaned by two undergraduate research assistants and coded by the first author in ATLAS.ti.
• Typology. Qualitative and quantitative data were combined into a four-part continuum of student orientations: the hater, the guilt-ridden user, the savvy adopter and the wholesale adopter. The authors stress it is not exhaustive and is a continuum rather than a set of distinct categories, and that most students fell into the two middle categories rather than the extremes.

Key Findings

• Most students use AI, but selectively. 65 percent reported using GenAI for coursework in the past year. Almost 90 percent used it for four or fewer of the ten listed tasks, with the modal respondent using it for two.
• Study and comprehension uses dominate. The most common were asking follow-up questions about course content or lectures (66 percent), summarizing course readings (56 percent) and generating sample test questions (42 percent). Creating outlines or getting ideas for written assignments (42 percent) and getting feedback on written work (38 percent) were also substantial.
• Generating prose is rare. Only 3 percent reported using GenAI to generate text for assignments, 2 percent to generate a full draft, and 6 percent to generate discussion questions or responses — the behaviors most often feared when institutions talk about academic integrity.
• The typology runs from rejection to wholesale adoption. The hater is concerned with hallucination, environmental impact, integrity and a dulling effect on thinking and Creativity; the guilt-ridden user employs AI while uneasy about it; the savvy adopter integrates it deliberately and critically; the wholesale adopter hands writing over entirely. The authors report most students sit in the middle two.
• Reliability is the leading concern. Of the concerns students raised, the top was the reliability of AI output (33 percent), followed by fear of committing an academic offense (28 percent) and worries about their own skill development (20 percent), with 16 percent reporting no concerns and 9 percent worried about subpar quality.
• Guidance is widespread but often unclear. Among survey respondents (N = 502), 81 percent said their professors and teaching assistants had provided guidelines on AI use, and 46 percent called those instructions very clear. The remainder reported no guidance (19 percent), very unclear (1 percent), somewhat unclear (5 percent) or only somewhat clear (29 percent) guidance.
• Students want instructors to engage with the technology. Interview participants asked for clearer, more open communication and commonly suggested that instructors should test the tools themselves to understand how students use them and how AI performs on course tasks. One suggested building class exercises in which students critically evaluate AI answers to course questions.
• Use is often driven by unmet needs. A participant described generating practice test questions because her professors did not provide any, leaving her to extract the material herself through AI. The authors read such accounts as evidence that some AI use responds to gaps in feedback, scaffolding and support rather than to a desire to avoid learning.

What this means for practice

• Instructors. State permitted and prohibited uses at the course level and require declaration of AI use — the model the authors point to is co-developing the policy with students, as done in three sociology courses, and asking them to be able to show their work: 81 percent of students had been given some guidance, but only 46 percent found it very clear and about one in four reported none or unclear guidance.
• Instructors. Ask what unmet need drives a use before assuming misuse — students generated practice test questions because their professors supplied none and turned to AI for feedback they could not otherwise get, so examining the scope and tenor of instructor and TA feedback is one way to reduce AI's appeal.
• Instructors. Choose the policy from the assignment's purpose: prohibit AI where professional writing skill is the goal but pair the prohibition with low-stakes scaffolded writing, and permit AI for feedback on already-written drafts where critical thinking is the goal.
• Instructors. Split large assignments into components and scaffold feedback through the term, since time pressure and overwhelm were what interviewees described as the trigger for reaching for AI.
• Instructors. Test the tools on your own course tasks, as interview participants suggested, and consider a class exercise in which students critically evaluate AI answers to course questions.

Limitations

• Survey estimates come from 504 respondents among 844 invited students in sociology courses at one Canadian public research university (273 in the winter term, 231 in the summer), with a course bonus point offered regardless of consent.
• Interview evidence rests on 12 volunteers out of 157 students contacted, spread across four focus groups of two to three students and two one-on-one interviews; 75 percent were second-year students.
• Use and views are self-reported, and response rates varied by question, so reported sample sizes differ across results — the guidance figures, for instance, use N = 502.
• The four-part typology is offered as a continuum rather than a validated classification, and the authors state it is not exhaustive.

Citation

Kuznetsov, A., Sheely, A., & Baker, J. (2026). Student Use of and Views on GenAI for Writing. Teaching Sociology.