Join us on Wednesday 21st of October at 8.00pm (GMT/BST) with our guest contributors, Anna Mills, College of Marin (@annamillsoer.bsky.social) and Carl Whithaus, University of California, Davis. This #LTHEchat on Bluesky will explore how peer and AI feedback can strengthen students’ agency, critical AI literacy and evaluative judgement while keeping human voice and relationships at the heart of learning.
It is easy to feel that AI is out of our control and out of our students’ control. And in many ways, it is. But what if we could take back some control and help our students to do so? This is what we are attempting in the Peer & AI Review + Reflection (PAIRR) Project. We hope that educators find a sense of agency in defining a limited role for AI in supporting learning according to our values and knowledge. Our approach builds on established best practices in writing instruction by introducing critical reflection on AI feedback alongside peer and instructor feedback.
We hope to help students build a sense of agency as they discuss readings about AI, writing and ethics; as they chat back to AI feedback rather than simply accepting it; and as they engage in metacognitive reflection on peer and AI feedback. With PAIRR, AI becomes a tool for thinking about their writing rather than a shortcut. Students stay in control of their words as they strengthen their voice, judgment, and revision skills. One student commented, “My opinion has been changed since using AI in [this course]. I realised that I can use AI to give me feedback instead of directly editing my paper in order to get insight on how to make my paper better without losing my voice.”
Even as it builds a sense of agency in relation to AI, PAIRR also offers students extra writing support in the form of supplementary AI feedback. It helps students understand what human audiences need from their writing and what human readers are getting from what they have written. Comparing peer and AI feedback helps students see how they feel about human versus AI feedback, what each might offer them, and what each might lack. We emphasise that writing is human communication, and that AI cannot replace the human audiences or peer and teacher relationships that motivate us.
When one of our research articles won an award from the Computers and Composition journal, we felt that the judges’ words expressed what we were aspiring to. The judges wrote that the PAIRR model “resists both technological solutionism” and “reactionary refusal,” “offering a rigorously tested framework for integrating AI ethically while preserving the relational, contextual value of human expertise in the writing classroom.”
The PAIRR model
Where does PAIRR come from, and how does it work? Peer and AI Review + Reflection (PAIRR) is a pedagogical model that was developed and tested at the University of California, Davis in 2023–24. In 2025 we were awarded a $1.5 million grant from the California Education Learning Lab, and that is when other campuses joined. We are now working with eight public colleges and universities across California. Here are the five steps in our model:

1. We start by planning classes with a clear AI policy aligned with the instructor’s values, goals and learning objectives. The class discusses this AI policy, and students engage with selected readings intended to build critical AI literacy:
- The first reading comes from the open textbook AI and College Writing: An Orientation, and it introduces students to thoughtful use of AI feedback while retaining their voice and critical thinking.
- We share a statement that educates about our very real concerns about AI: environmental impact, labour, bias of all kinds, linguistic justice and equity, privacy, intellectual property and datafication. The statement explains why we choose to work with AI anyway and how engaging with AI might help us build capacity for civic engagement in shaping AI futures.
- We offer a linguistic justice and equity reading that helps students think about why they might want to privilege their own writing voice over the polished standard English voice that AI often promotes.
2. Then, once students have written a first draft of a major assignment, they complete peer review workshops in class. Peer review comes before AI because we believe humans belong at the centre of response. (Some instructors, particularly in online classes, have reversed this order and use AI feedback first. We recognise that the sequence of peer feedback first may not work best in all cases, and we want teachers to adapt PAIRR to make it work in their own courses. Still, we recommend peer review first, because at its base PAIRR – and peer review in general – is about the human interaction, about the writer receiving feedback from someone like their intended audience. AI can help with that, but can’t replace the value of the human).
3. Students then get AI feedback on their draft and chat with the AI about it. This can work on various platforms, but currently we use MyEssayFeedback, a not-for-profit platform designed by faculty that incorporates our approach in its user interface and reduces our administrative labor by syncing with the learning management system and giving participation credit automatically.
After they read the initial feedback, students are required to ask at least two follow-up questions. They can “chat back” however they like, but we also offer suggestions. Dropdown menus show them types of requests and sentence starters they could use. Here are a few of our suggestions:
● Be frank and push back if you’re not sure you agree
● Ask for clarification on a specific point
● Ask it to help you explore what you’re not sure about
● Ask about an assignment requirement it didn’t mention
● Ask about your writing style’s unique strengths and how to bring in more of your voice
The aim is for students to treat AI as a thinking partner rather than an answer machine. These options and more can be found in the PAIRR packet.
4. After that, students compare the peer and AI feedback, reflect on what fits their goals for the assignment and what doesn’t, and make a plan for revision. We share reflection questions that invite them to step back and consider what they agree with and what they don’t, how the feedback affects their sense of their voice, and how they plan to revise.
5. Finally, students revise and write a revision reflection, deepening their metacognitive awareness of how feedback affects them and their strategic thinking about how to engage with it in future. This reflection is absolutely key. It encourages students to think critically about their own writing and revising processes and about AI’s role within those processes. It’s empowering because students have agency not only on this particular assignment, but in terms of thinking about when they do and do not want to use AI when they are working on writing tasks.
Opting out
We respect students who do not want to use AI, and we offer them an alternative assignment. We encourage them to schedule an additional writing centre or tutor meeting, or we may offer them the option to talk through their draft with a friend or family member. Then they write a reflection on the different kinds of feedback they received.
Results
In our 2023–24 pilot, we used the method with 654 students in a range of courses, from first-year writing to writing-intensive courses in other disciplines, including STEM. 57.6% of students preferred combined AI and peer feedback to AI feedback alone. Our takeaway was that AI works best when paired with human feedback, not on its own. AI and peer feedback often reinforced each other, and when they differed, they tended to offer complementary perspectives. Students found that AI could be helpful, but they still valued human feedback the most. As one University of California, Davis student put it, “I think the value of peer review is definitely important to understand how other people, who don’t have access to all your mind’s information, perceive your writing.”
The PAIRR feedback prompts
The quality of AI feedback depends on which AI system you use, but it also depends on how you direct the system. In PAIRR, when students request AI feedback, we make sure the instructor’s assignment and rubric are included to give context about what students are asked to do. We also describe the kind of feedback we as writing instructors think will be most helpful to students. Through two years of collaborative design and testing, we developed two core feedback prompts:
· The Reader-Response Prompt instructs the AI system to emphasise the intrinsic meaning of writing as human communication. It describes strengths and areas for consideration in terms of how a human reader might respond: what value a reader might find in the writing, where they might be confused, where they might need something different. It uses phrases like “… might resonate with readers.”
· The Criteria-Based Prompt is more directive: both prompts respond to the assignment criteria, but the Criteria-Based Prompt centres three to four direct suggestions for improvement instead of describing two areas for consideration in terms of likely reader reactions.
Both prompts give feedback on strengths as well as considerations for revision. Both include quotations and details from the draft to support each comment. And both strongly discourage the chatbot from writing for the student.
We are currently experimenting with a version of each prompt that gives contradictory suggestions for revision. Our framework already encourages students to question all the feedback they get. But we recognise that as these systems get more sophisticated, their errors are harder to spot. Students still need practice identifying flaws and experiencing those moments when they realise, “No, that’s not what I wanted to do,” or “It’s not getting this part right.” Contradictory but plausible feedback provides that practice: students can’t take every suggestion, so they have to recognise that AI advice can sound good and still be wrong for them.
We’re hoping to develop a full prompt library based on the work that instructors using PAIRR are doing. We want these prompts to be rigorously tested, and we want them to help students develop as writers. PAIRR is very much a shared endeavour by faculty hoping to shape AI technologies in ways that benefit students and encourage them to think about how they are writing and revising.
Trying or adapting our approach
To see what kind of feedback you get, you can copy and paste our PAIRR prompts into any chatbot with a draft and assignment. You can also use the PAIRR custom bot we built on the PlayLab platform (PlayLab is a nonprofit that offers these for free and does not use the data to train AI systems) or try one of our contradictory feedback bots. If you don’t have a student essay you have permission to use, generate an AI sample draft to get feedback on. You are welcome to comment on, edit, or use an excerpt from one of our prompts; the CC BY NC 4.0 licence allows for all of that!
An invitation to engage
Please connect and let us know what you think! Everything we create is an open educational resource. We want to keep developing this model with input from others and share it as part of larger discussions of how educators can best navigate AI in writing.
AI Use Statement
We came up with the ideas and drafted the piece. We used AI for editing and style suggestions as well as global feedback and a reference list. We reviewed all AI suggestions.
Acknowledgements
PAIRR is the work of a large team across nine campuses. We thank the whole team and especially Marit MacArthur, Julie Gamberg and Sophia Minnillo, who presented the August 2026 PAIRR webinar this post draws on. PAIRR is supported by the California Education Learning Lab, with additional support from the Association for Writing Across the Curriculum, the National Endowment for the Humanities, and the UC Davis Center for Artificial Intelligence and Experimental Futures (CAIEF).
References
Gamberg, J., MacArthur, M., Mills, A., Sinha, A. and Sperber, L. (n.d.) AI is problematic. Why does PAIRR use it? An ethics statement for the Peer & AI Review + Reflection Project. UC Davis University Writing Program.
Mills, A. AI and College Writing: An Orientation. LibreTexts.
Minnillo, S., Gamberg, J. and Whithaus, C. ‘A prompt that nudges AI to speculate on human readers’ responses to a draft’, in Gupta, A. and Gogan, B. (eds) Writing Studies Prompt Library: Expert Approaches to Generative Artificial Intelligence Prompting. Writing Across the Curriculum Repository. WAC Clearinghouse.
PAIRR Project (n.d.) PAIRR feedback prompts.
PAIRR Project (n.d.) PAIRR packet.
Sperber, L. (2026) ‘Now that we have AI, we can ditch human feedback! ✨ AI feedback should be used in a human-centered process, with peer feedback’, in Bad Ideas About AI and Writing. WAC Clearinghouse.
Sperber, L., MacArthur, M., Minnillo, S., Stillman, N. and Whithaus, C. (2025) ‘Peer and AI Review + Reflection (PAIRR): A human-centered approach to formative assessment’, Computers and Composition, 76, 102921.
Bios

Anna Mills is a leading voice on AI in higher education across platforms like LinkedIn, Substack, and BlueSky (@annamillsoer.bsky.social). She has taught writing in California community colleges for 20 years, authored two popular OER textbooks, and serves on the Modern Language Association Task Force on AI in Research and Teaching. Her writing appears in The Chronicle of Higher Education, Inside Higher Ed, and Computers and Composition. As a volunteer advisor, she has helped shape MyEssayFeedback.ai, and she serves as co-Principal Investigator for the Peer & AI Review + Reflection (PAIRR) project.

Carl Whithaus is a Professor of Writing and Rhetoric at the University of California, Davis. He studies writing technologies and digital cultures and works on a variety of projects related to writing in the sciences, engineering, and agriculture. His books include Swarms, Viral Writing, and the Local (University of Pittsburgh Press, 2025), Multimodal Literacies and Emerging Genres (University of Pittsburgh Press, 2013), Writing Across Distances and Disciplines (Routledge, 2008) and Teaching and Evaluating Writing in the Age of Computers and High-Stakes Testing (Erlbaum, 2005). He serves as a Principal Investigator (PI) for splash! milk science update and for the Peer and AI Review + Reflection (PAIRR) Project.
























