Co-Writing with AI: An Empirical Study of Diverse Academic Writing Workflows
This empirical study investigates how university students integrate AI into diverse academic writing workflows by identifying three distinct, value-oriented usage configurations—early-stage learning, late-stage quality refinement, and peripheral productivity enhancement—that emerge from the interplay of task demands and individual factors like AI literacy and authorship concerns.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine academic writing as a long, winding road trip. For decades, students have been driving this road alone, relying on their own maps, their own fuel, and their own navigation skills. But recently, a new kind of GPS has arrived: Generative AI.
This paper, titled "Co-Writing with AI," is like a travel log from 107 students and 12 experienced drivers. The researchers wanted to find out: How are students actually using this new GPS? Are they letting it drive the car? Are they using it just to pick the music? Or are they ignoring it entirely?
Here is what the study found, broken down into simple concepts.
1. The Big Discovery: It's Not "All or Nothing"
You might think students either use AI for everything (letting it write the whole essay) or nothing (banning it completely). The study shows that's not true.
Instead, students treat AI like a specialized toolkit. They don't use every tool for every job. They pick and choose specific moments in the writing process to bring out the AI, and they put it away for the rest of the time.
The researchers mapped the writing process into five "stops" on the road trip:
- Ideation: Figuring out what to write about.
- Sourcing: Finding facts and references.
- Planning: Making the outline.
- Drafting: Writing the actual sentences.
- Reviewing: Checking for mistakes and polishing.
Most students only use AI at two or three of these stops, not all five.
2. The Three "Driving Styles" (Workflows)
The study found that students naturally fall into three distinct patterns, or "driving styles," based on what they value most: Learning, Quality, or Speed.
🧠 Style A: The "Learning Explorer" (Early-Stage Focus)
- The Goal: To understand the topic and learn new things.
- The Strategy: They use AI heavily at the beginning of the trip (Ideation and Sourcing) but stop using it once they start driving the car themselves.
- The Analogy: Imagine a student using AI as a tour guide at the start of a hike. The guide points out interesting rocks, explains the history of the trail, and suggests a path. But once the student starts walking the path, they put the guide away. They want to do the walking themselves so they don't lose their fitness (skills) or forget the view (their own voice).
- Why? They fear that if they let AI write the actual essay, they won't learn the material, and their own "voice" will disappear.
🎯 Style B: The "Quality Inspector" (Late-Stage Focus)
- The Goal: To get the best possible grade and ensure the final product is perfect.
- The Strategy: They do the hard thinking, planning, and outlining themselves (no AI allowed here!). But once the draft is written, they bring AI in for the end (Drafting and Reviewing) to polish the language.
- The Analogy: Think of this student as a chef who cooks the entire meal from scratch, chopping the vegetables and seasoning the sauce by hand. But before serving the dish, they hire a food stylist (AI) to make the presentation look beautiful and ensure the plating is perfect.
- Why? They trust AI to fix grammar and flow, but they don't trust it to come up with the original ideas or arguments. They worry AI might lie about facts or give generic ideas.
🚀 Style C: The "Momentum Keeper" (Throughout the Trip)
- The Goal: To get the work done, keep moving, and avoid getting stuck.
- The Strategy: They use AI throughout the whole process, from the first idea to the final check.
- The Analogy: This student treats AI like a co-pilot or a junior assistant. If the road gets bumpy or they feel tired, the co-pilot takes the wheel for a bit to keep the car moving. They use AI to break the "blank page" fear, to summarize long texts quickly, and to double-check their work at the end.
- Why? They value efficiency and momentum over strict control. They are willing to trade a little bit of "perfect originality" for the benefit of getting the essay finished and feeling less overwhelmed.
3. The Hidden Tension: "Who is the Driver?"
The study highlights a constant tug-of-war students feel between getting the job done and owning the work.
- Trust vs. Fear: Students who use AI a lot generally trust it more. They see it as a tool to boost productivity. Students who use it less often worry about "authorship"—the feeling that if AI helps too much, the essay isn't really theirs anymore.
- The "Shallow" vs. "Deep" Use:
- Shallow Use: Asking AI to "fix my grammar" or "give me a list of ideas" and just accepting the answer. (Like asking a friend for a quick tip).
- Deep Use: Asking AI for an idea, arguing with it, changing the idea, and then checking if the sources are real. (Like having a debate with a friend).
- The study found that "Deep Use" happens mostly when students are planning or drafting, while "Shallow Use" happens when they are just checking spelling or looking for quick facts.
4. What This Means for Schools and Tools
The paper concludes that we need to stop asking, "Did you use AI?" (Yes/No). Instead, we should ask, "How did you use AI, and at what stage?"
- For Teachers: Instead of banning AI, they should teach students when it's okay to use it and when to put it away. For example, "Use AI to brainstorm, but write the outline yourself."
- For Tool Makers: AI tools shouldn't just be "write my essay" buttons. They should be smart enough to know which stage of writing the student is in. If a student is brainstorming, the tool should offer ideas. If the student is writing the final draft, the tool should offer grammar checks but not rewrite the whole paragraph.
Summary
The paper argues that students aren't lazy cheaters or blind robots. They are strategic managers. They are carefully deciding which parts of their writing journey to handle alone and which parts to share with AI, based on whether they want to learn, get a good grade, or just finish the work.
The future of writing isn't about humans vs. AI; it's about humans learning how to share the steering wheel at the right times.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.