Most arguments about AI in writing collapse into one question: should students be using these tools at all? That framing misses the research. Reza, Thomas-Mitchell, and colleagues (2025) at the University of Toronto reviewed 109 HCI papers on AI writing tools and interviewed 15 writers. They argue the right question isn’t whether AI belongs in the writing process. It’s where in the process it belongs, for whom, and why.
The Content-Form Split
The interview study produced a split the field hasn’t named clearly. Reza et al. (2025) found that writers fall into two camps based on what they want to own. Content-focused writers (academics, technical writers, journalists) protect the planning stage because their ideas are their primary contribution. Form-focused writers (novelists, poets, screenwriters) protect translating and revising because their sentence-level decisions are what defines them as writers.
The authors put it this way: “writers value ownership most strongly over components of the composition process they see as their primary contribution. Writers tend to be more open to delegating composition tasks to an AI for areas that are tangential to their perceived primary contribution” (p. 18).
That’s the pattern. Content-focused writers welcome AI help with translation, polish, and grammar but resist AI in ideation. Form-focused writers happily delegate brainstorming and structure but resist AI at the sentence level. This is the opposite of what most AI writing tools assume, which is that AI help is uniformly good across the writing process.
The Four Design Strategies and What They Miss
The systematic review identified four design strategies in current AI writing tools. Structured Guidance treats AI as a coach for skill development. With Guided Exploration, AI becomes a facilitator for navigating an idea space. Active Co-Writing assigns AI an active partner role with substantial content generation. Critical Feedback restricts AI to editor and organizer functions.
Active Co-Writing is the most common strategy (37.1% of systems), and the most concerning. Reza et al. (2025) found that half of creative writing tools deploy this strategy, even though creative writers are exactly the population most likely to feel violated when AI generates substantial content for them. The field has been building tools for creative contexts using strategies that work against what creative writers actually want.
Sperber et al.’s (2025) PAIRR research, which I’ve covered before, made a similar argument from a different angle: AI feedback works best as a supplement to peer review, not a replacement. The pattern is consistent. AI tools that try to do too much end up reducing the writer’s investment in the work.
The Forgotten Process
The paper’s most interesting structural argument is about monitoring, the meta-cognitive process of overseeing one’s own writing alongside the AI’s contributions. Only a handful of papers across all writing contexts engaged with it. Most current systems give writers no global toggle to turn AI off, no local toggle to override AI suggestions for intentional rule-breaking, and no clear interface for tracking what the AI has touched.
Reza et al. (2025) argue that “as AI systems advance, monitoring and management of the Human-AI collaborative relationship becomes increasingly important” (p. 23). That argument transfers cleanly to teaching. Students using AI tools without monitoring affordances can’t track which parts of the work are theirs and which parts belong to the model.
Hawkins, Taylor-Griffiths, and Lodge’s (2025) work on feedback literacy and AI-enhanced essay writing, which I’ve covered before, makes the same point from the pedagogical side. The judgment muscles only develop with practice, and practice requires the ability to see what’s happening.
What This Means for Teaching Writing
The pedagogical implication is clear. A research paper isn’t the same as a short story or a personal essay. Each demands a different kind of AI integration. Students writing argumentative academic work need to protect their planning and ideation. Students writing creative fiction need to protect their voice and sentence rhythm. Blanket policies that treat all writing the same will work against half of the students they’re meant to help.
This also means workshops on “AI in writing” need to get more specific. Generic “use AI ethically” instructions don’t do the work. Teachers have to walk students through the chessboard pattern Reza et al. describe: where AI help respects ownership, and where it erodes it.
Where I’d Push the Argument Further
The paper has limits the authors flag clearly. The interview sample is small (15 writers), age-limited (max 34), and educationally homogeneous (mostly university-educated). The systematic review covers only the ACM Digital Library. Most participants used ChatGPT primarily, which means alternative tool ecosystems are underrepresented.
I’d add one more concern. The content-form split is useful, but most academic writers I work with do both. A literature review demands content ownership. The discussion section often demands form ownership too, because voice and argumentative texture matter. Real writing rarely fits cleanly in one camp. The framework is a starting point but definitely should not be viewed as a final taxonomy.
The authors put their broader argument plainly: “ownership in AI-assisted writing is not fixed but actively negotiated—shaped by writers’ goals, contexts, and the interaction design of the tools they use” (p. 25). That negotiation is what teachers should be helping students do, deliberately and out loud.
Find where your contribution as a writer lives. That’s the part to protect. The rest can come from AI. Students need to learn to make this call themselves.
References
- Hawkins, B., Taylor-Griffiths, D., & Lodge, J. M. (2025). Summarise, elaborate, try again: Exploring the effect of feedback literacy on AI-enhanced essay writing. Assessment & Evaluation in Higher Education. https://doi.org/10.1080/02602938.2025.2492070Â
- Reza, M., Thomas-Mitchell, J., Dushniku, P., Laundry, N., Williams, J. J., & Kuzminykh, A. (2025). Co-writing with AI, on human terms: Aligning research with user demands across the writing process (arXiv:2504.12488). arXiv. https://arxiv.org/abs/2504.12488
- Sperber, L., MacArthur, M., Minnillo, S., Stillman, N., & Whithaus, C. (2025). Peer and AI Review + Reflection (PAIRR): A human-centered approach to formative assessment. Computers and Composition, 76, 102921. https://doi.org/10.1016/j.compcom.2025.102921