What Teaching Writing With AI Reveals About Student Voice

After exchanging peer feedback on their first drafts, students in Ana Castellon’s “Introduction to Academic Literacies” class uploaded their work to ChatGPT for another round of feedback. The exercise was part of the statewide research study Peer & AI Review + Reflection (PAIRR), led by UC Davis researchers Drs. Carl Whithaus and Marit MacArthur. The study explores how AI feedback on student writing can help students improve their drafts and influence their decision-making.

Castellon and fellow Ph.D. student Casey Manogue, both instructors for the UC Davis Writing Center, were approached by the PAIRR research team to collect data in their classrooms. Over the course of one quarter, they and other participating instructors implemented the PAIRR curriculum. Students discussed readings about AI and language equity, completed peer and AI reviews of their drafts, compared the feedback, and decided which recommendations to incorporate into their final work. For Castellon and Manogue, teaching writing with AI means helping students make those decisions for themselves.

Early study results suggest that students found value in combining peer and AI feedback, which Castellon and Manogue also observed in their classrooms. But their students expressed reservations about AI in the classroom, particularly the reasoning behind its recommendations and its tendency to flatten the humanity behind writing. “This study is human-centered,” Manogue said. “It’s showing that, even though this technology can be very beneficial, there’s still a lot of value in the human perspective.”

Teaching Writers to Question AI

Teaching writing with AI begins with questions in Castellon’s and Manogue’s courses. Before using the technology on their own work, students examine its effectiveness, limitations, and implications for language and education. Through readings and classroom conversations, they consider why AI was created and how it can influence social, political, and ethical spheres—including their own studies.

Castellon and Manogue hope that open conversations about AI in the classroom will show students that they don’t need to be secretive when using it. Rather than asking students to embrace or avoid the technology, they encourage them to consider when it supports the goals of an assignment and when it interferes with them. “When you tell students not to do something, they often respond by saying, ‘Why not? Let’s see what happens,’” Castellon said. “But if you have these conversations with them, you prompt students to stop and think about why AI actually makes sense for their assignments. You provide a different lens for thinking about how they can find real benefit in these tools.”

Putting Peer and AI Feedback Side by Side

Students then put that critical approach into practice. Castellon and Manogue help their classes write prompts that align with the assignment rubric and ask specific questions about tone, structure, and audience. Students compare ChatGPT’s responses with feedback from their peers and decide what to incorporate into their work.

An earlier published PAIRR study shows the value of comparing peer and AI feedback. When a peer and ChatGPT flagged the same issue, students felt more confident that it needed attention. When the feedback differed, they gained another perspective on their drafts. Overall, 58% preferred receiving both forms of feedback, while only 6% preferred ChatGPT alone.

The two sources offered different benefits. AI feedback on student writing was immediate and tied to the rubric, while peers understood the course and responded as authentic audiences. Comparing them required students to decide which feedback they trusted and whether it supported what they were trying to communicate.

Protecting Voice and Learning

Students also identified AI’s limitations. Its advice could be generic or irrelevant, and its responses did not always align with their prompts. Rather than allowing AI to correct or rewrite their work, students used it selectively to outline papers, clarify main ideas, or refine language and syntax.

“AI can be very deceiving. You have to be aware that not everything it’s giving you is 100% truthful or logical,” Castellon said. “If you’re not questioning where the information is coming from and checking that it’s accurate, you’re losing out on critical thinking opportunities about your own writing.”

Castellon and Manogue were surprised by how working with AI made students more protective of their voices. Instead of assuming its version was stronger, students questioned what had changed and whether the revision still reflected what they wanted to say. “My students said, ‘AI takes what I say and gives me a robotic spiel. I don’t think I’m in there,’” Castellon said. “They realized that AI just translates everything, and they chose not to lose themselves in that process.”

For Castellon, preserving students’ voices is inseparable from the learning that takes place during writing. As a second-language learner, she recognizes how AI can make coursework easier for multilingual students. She also believes that practicing language while drafting essays, without relying on AI to do that work, is part of becoming a better writer and learner.

“I’m honest with my students and I say, ‘If I had access to artificial intelligence during my studies, I would have done myself a disservice, because you have to practice language,’” Castellon said. “Yes, it’s easier to feed your writing into AI. But then you take away the most beautiful thing, which is your own learning.”

Choice and Consent in AI Research

Student agency extends beyond deciding which feedback to use. It also requires meaningful choices about whether their work and experiences become part of AI research. PAIRR deepened Castellon’s and Manogue’s awareness of their responsibilities as scholars studying emerging technology. As writing instructors and graduate student researchers, they had to consider transparency and informed consent with greater care.

“It’s really important to make sure that students understand what they’re getting themselves into when they sign up for a research study like this,” Manogue said. “Since I’m carrying out research for PAIRR, I’ve had to really think about how to be as transparent as possible with them and make sure they know what they’re consenting to with large language models and research studies.”

Castellon and Manogue don’t know how significantly artificial intelligence will change education, but their students’ responses offer guidance for navigating that change. Their approach to teaching writing with AI centers on helping students question its recommendations, make deliberate choices, and recognize the value of what they—not the technology—bring to their writing.