Are students really letting AI write for them? Study observes what happens during writing

Is AI transforming academic writing? By directly observing the student writing process, a pilot study suggests that generative tools are mainly used to unlock ideas and rework texts, without entirely replacing the author.

Debates on generative AI in higher education have relied on studies of already completed student work, or on declarative survey data. This research shows that artificial intelligence tools can support learning, but it has also raised concerns, including over-reliance on students, cheating and the possible degradation of critical thinking and engagement.

While these types of studies provide interesting snapshots of reported practices, their methodologies can miss a key aspect: how writing actually happens when students write with the help of teachers. AI.

The pilot study I conducted among undergraduate students at Kennesaw State University takes a different approach. Using think-aloud protocols – a method in which participants verbalize their thoughts as they complete a task – our research observes how students interact with generative AI tools during the process itself writing. This method makes it possible to understand decision-making processes as they occur.

Our initial results suggest a more complex reality than the often-advanced narrative that students simply let AI write their homework. Rather, many seem to be negotiating when and how AI has a place in their writing work.

Looking inside the writing process

In our study, 20 undergraduate students completed a 20-minute writing session in response to the following instruction:

People spend a lot of time trying to achieve perfection in their personal or professional lives. They often demand perfection from others, creating expectations that are difficult to meet. Conversely, some people believe that perfection is neither attainable nor desirable.

The task was to write a thesis and argumentative paragraphs based on evidence to defend their position on the value of the quest for perfection. The students knew that they were not supposed to finish their text, but rather to move forward in their writing process towards a completed text. They were also told that there was no right or wrong way to use AI, and that they should use generative AI exactly as they normally would when writing.

Rather than directly observing students, the study relied on screenshots taken after the session and analysis of the students’ descriptions of their own writing process. Collecting this data – their actions on the computer and transcriptions of voice recordings – allowed the researchers to analyze the process without interrupting it.

To reduce the risk that students would change their behavior if they felt they were being observed, the researchers started a timer and then left the room during the writing session. The objective was to limit the Hawthorne effect, a phenomenon by which individuals modify their behavior because they know that they are being observed.

What we observed

Across the transcriptions, a few qualitative trends emerged repeatedly in how students collaborated with the AI ​​during writing. First, many participants turned to these tools early in the writing process to generate ideas or outline a thesis.

We then see these students using the proposals produced by the AI ​​to stimulate and structure their own ideas. One student describes this strategy as follows: “After [avoir généré quelques idées]I usually just use this [résultat] as a starting point. HAS”

In those moments, the AI ​​functioned less as a definitive answer and more as a brainstorming tool helping students overcome the anxiety of the blank page.

Moreover, students often continued writing independently after generating these first ideas. Many transcripts contain phrases like “I think my thesis should be…” or “Let me write this part,” which suggests that some students retained control of their argument.

Fix the bot

Another strong pattern that emerges from the transcriptions is that students rarely accept the texts produced by AI without modifying them. Rather, they actively revise the generated language. As one student described it, the AI ​​“rewrites” its first instructions, then the student in turn rewrites the AI’s response. This allows him to claim “authorship and control” of the final version.

Another participant also redirected the tool’s response when it did not match the instructions: “The AI ​​is not following the instructions… try again.” HAS”

These moments show that students are critically evaluating the AI’s answers and treating it almost like a debate partner, rather than simply copying them.

We also found that some students completely rejected the AI’s suggestions. In several writing sessions, participants explicitly decided not to use AI-generated responses. One student commented on this choice as he wrote: “I don’t really use AI for my research. HAS”

Other transcripts show students returning to their own writing when the AI’s responses seemed too generic or disconnected from their argument. These moments indicate that they are not only collaborating with AI: they are also drawing limits on the place it can occupy in their writing process.

Finally, several transcripts show that students turned to AI in moments of uncertainty or when they felt stuck. As one participant explained: “I used AI a lot because I was struggling. Even in these cases, however, students often used him as support while writing their essay, rather than directly copying and pasting his answers.

What this says about AI and writing

Our analysis suggests that generative AI fits into student writing not as a complete replacement for the human author, but as a form of negotiated collaboration. The results indicate that it occurs most often during idea generation, revision and when a feeling of blockage appears, while students retain control over the choice of arguments, the way of writing and the final wording.

Understanding how decisions to use AI unfold during the writing process – not just what appears in the final essay – can help teachers design assignments and rules that clearly keep the human in charge.

As these initial results come from a pilot group of 20 undergraduate students, they should be interpreted with caution. To check whether these trends are confirmed on a larger scale, the research team is currently extending the study to 100 participants. This expanded phase will also examine how neurodivergent students interact with generative AI during the writing process, an area still largely unexplored by research.

Undergraduate student researchers at Kennesaw State contributed to the preliminary analysis presented in this article: Kylee Johnson, Vara Nath, Ruth Sikhamani and Kaylee Ward.