As anyone who has encountered an AI tool in a writing space can attest, there is an allure to accepting the numerous revisions that it offers.
Whether that’s clicking “improve” on a sentence full of red squiggly lines or typing a brief description of what you want to write about in an empty prompt box (that will, of course, automatically turn into a full draft), there is a wide array of ways to get assistance for your writing.
For adult writers with a lifetime of experience and sophisticated composing processes, allowing AI to assist in the moment might be advantageous. But, for students who are still developing their writing skills and voice, using AI without guidance might be detrimental to their learning. It is too easy for students to slip into “cognitive surrender”—where “the decision maker no longer constructs an answer, but adopts one generated by an external system” (Shaw & Nave, 2026)—and lose “productive friction,” or what the WestEd Digital Fluency group and the National Writing Project describe as “the mental, social, and emotional efforts that make learning meaningful” (2025).
Put another way, a key aspect of learning to write is moving through a process in which the struggle is the goal. Words become sentences. Sentences transform into paragraphs. Entire arguments or stories emerge. Friction creates better writing. If we allow AI to continually subvert that process for our students, we are robbing them of the metacognitive awareness they need, not to mention the actual mental work they need to do as writers.
Still, as Kristen Hawley Turner and I have noted (2026), writers, even “solo” authors, are always in conversation with others, whether that’s responding to another piece of writing, being inspired by other ideas, or specifically working with another author to collaboratively produce texts. As collaborators, “[C]o-writers often complement each other’s strengths and weaknesses, making the final product richer and more polished than it might have been if created individually.”
If we look at AI not as a replacement for a student’s ideas and voice—or as a replacement for teacher feedback and guidance—but as yet another co-writer or collaborator, and if we understand that our students are already experimenting with AI (whether we want them to or not), then it is our job as educators to explicitly model for and teach students how to co-write with AI.
PAIR: A Protocol for Co-Writing with AI
First, an acknowledgment. There are significant ethical, environmental, economic, and educational consequences for any of us who choose to use AI to assist with writing.
There are also benefits. It almost goes without saying that inviting writers to use AI is a complicated issue with many nuances. In my work with teachers, I no longer talk about AI use as a “yes/no” dichotomy or even a “red, yellow, or green light” decision. Our discussions have shifted more toward how to make critical and creative decisions about who our writers are, what they need, and why they are using—or not using—AI in any given moment.
In my work with teachers, I no longer talk about AI use as a “yes/no” dichotomy or even a “red, yellow, or green light” decision. What matters is why students are using—or not using—AI in any given moment.
Similarly, I recommend that writing teachers ask a series of questions—individually in their own personal reflection and in conversation with colleagues—at many stages of the instructional cycle. These questions help us understand the affordances and limitations of AI as a coauthor, promote student autonomy, and recognize that co-writing is, in the end, a part of any writer’s process:
- When should I teach my students to use AI at any given stage of the writing process—only during brainstorming, or at any point in the composing process when they are struggling and are truly stuck?
- Why should I teach my students to use AI with this particular piece—is it meant to be a coach and a coauthor in a deliberative sense, or is it simply trying to get ahead of the game by saving time and is a case of “cognitive offloading?”
- How should my students use AI—is this a case where they can turn on a grammar checker and get some quick tips, barely slowing down their writing process and improving overall style? Or, will they choose to engage in an extended dialogue with a chatbot to prime their thinking and get feedback?
- To what extent will I teach my students to choose to use AI—by relying on the pre-prompted questions that the AI offers? Or will they craft their own prompts to ask questions by thinking metacognitively about the work at hand?
These are not easy questions, nor are they ones that maintain a consistent answer in all contexts. These questions, too, are nuanced and incomplete, even in the initial asking. They could—and likely should—lead to even more questions.
As teachers consider these questions, we then need to help students consider the same types of productive, ethical decision-making in their own work with AI as a coauthor. To help in that process, I’ve developed the PAIR protocol, which can encourage writers to critically examine their own choices about when, why, how, and to what extent to use AI—or not—in their writing process. Teachers can use this protocol in various scenarios and at various grade levels to mentor students through a co-writing process with AI, which will help them learn how to intentionally, ethically, and strategically integrate these tools into their work. The PAIR protocol looks like this:
- Primary Problem and Purpose for Using AI: Before choosing to use any form of AI, a student writer should be able to articulate what they are struggling with and want to accomplish with a specific writing task. Subsequently, they can determine whether it is appropriate to use AI to help in a way that would support their growth as a writer, and not merely as a shortcut for efficiency.
- Asking AI for Assistance: Once a writer has identified a problem, they should determine the limited, specific need that they have in the moment. The request for assistance should be targeted, ensuring that the writer is only taken away from the flow of their own writing process as long as necessary. If the first step in the protocol asks them to decide whether and why to use AI, this step moves into the what and the how. For instance, they should consider exactly what they want AI to do—and put guardrails both on the AI and on themselves—to maintain productive struggle and preserve their writerly autonomy. For example, a student could ask AI to provide feedback on a paragraph they’ve written but specifically ask that it not rewrite any sentences for them.
- Interacting Intentionally with AI: The next step involves being critical about AI suggestions. Before accepting any suggestions, a writer should consider what is being offered and how it aligns with their own writing style and goals. This is where teacher mentorship and modeling are paramount, showing what to accept or consider outright, what to reject, and what could be used as inspiration for other ideas or writing.
- Reflecting and Documenting Uses of AI: Finally, as writers wrap up their work and prepare assignments for submission, they should prepare a brief statement at the end of their writing that describes the decisions they made to use AI, as well as why those uses of AI were particularly useful in their writing process. This allows them to consider their writerly decisions, with and without AI, in a metacognitive manner.
To illustrate how PAIR might be used to guide students in different contexts, let’s look at two examples, one elementary and one secondary.
AI as a Co-Writer in Elementary Grades: Word Choices
Ms. Garcia wants her 3rd grade students to use AI in a purposeful, yet age-appropriate manner to improve the sensory details in their personal narrative writing. Having recently been introduced to a school-approved, enterprise-level AI chatbot builder, she designed her own chatbot for this purpose. During a mini-lesson, Ms. Garcia explains that the chatbot will identify details students have provided, encourage them to add more “showing, not telling” details, and not write anything for them. She then shares a rough draft of her own personal narrative as an example:
It was a warm, sunny summer day. As I licked my ice cream cone, it began drip, drip, dripping all over my hands, and then my chin, and soon my whole face was covered. It was the best treat ever!
Before entering anything into AI, Ms. Garcia asks the class what they think is strong about the details she’s added and what they think she could do better. After a brief discussion where students give suggestions (“Say more about where you were when you got the ice cream cone!”, “What kind of ice cream was it?”, “Were there toppings?”), she copies her text into the chatbot and asks it to provide suggestions about adding sensory details.
Note that, in this example, Ms. Garcia has already modeled the first two elements of the PAIR framework: she identified for her students the primary purpose (P) for using AI (helping with sensory language) and has pre-programmed the AI for a limited form of assistance, providing guardrails for students and asking for specific help on using a metaphor (A). She is also creating space for students to pause and activate their own thinking before seeing what AI produces for them.
She shows the students the AI response. Here’s an excerpt:
You have a fun scene started here—the rhythm of “drip, drip, dripping” gives a real sense of motion. Let’s look at how we can push the imagery even further, so the reader feels like they are standing right next to you on that sidewalk.Let’s take it sentence by sentence.Sentence 1: “It was a warm, sunny summer day.”The Challenge: “Warm” and “sunny” tell the reader the weather, but they don’t let the reader feel it yet.Guiding Questions:• What did the heat actually do to the physical world around you? Was the sidewalk radiating heat through the soles of your shoes? Was the sun beating down on the back of your neck?• Did you hear cicadas buzzing, or see heat waves shimmering off parked cars?Your Turn: How can you rewrite this first sentence by describing one specific physical sensation or visual detail that proves how hot the day was without using the words “warm” or “sunny”?
Ms. Garcia begins querying her students to move into the Interacting Intentionally with AI (I) part of the protocol: “What do we like about this feedback? What does the AI get ‘right’ about what we should revise? What don’t we like? What might we change in our writing?”
She then invites students to make revisions based on the AI feedback, and ends with asking them what worked well in the process, what felt awkward or unhelpful, and what they learned about the skills of adding more sensory details to their writing (the R step of the PAIR protocol).
Before accepting any suggestions, a writer should consider what is being offered and how it aligns with their own writing style and goals.
AI as Co-Writer in Secondary Grades: Honing an Argument
For Mr. Thompson, asking students to identify a clear, reasonable alternative perspective had always been a challenge in teaching AP Civics. With his students old enough to use the district’s approved AI chatbot on their own, he saw an opportunity to help them learn counterarguments by engaging in dialogue and defending their ideas.
After students drafted their arguments, Mr. Thompson asked them to think about three approaches to counterarguments that they had been discussing in class.
“You have your own claim, evidence, and reasoning, and now you need to understand what an opponent would have for their own,” he explained. “To help think through the strongest approach to dealing with your opponent, we are going to chat with AI.” (P)
He then provides them with a structured prompt that they can fill in as they begin their dialogue (A):
“Attached, find a draft of my argument essay about [FILL IN YOUR TOPIC]. Based on the claims, evidence, and reasoning that I have supplied, I want to engage in a dialogue with an opponent. In three rounds, ask me to articulate and then provide me feedback that will help me improve: (1) a concession and pivot; (2) a direct refutation; (3) a rebuttal with acknowledgment and further articulation of my own argument. If I provide answers that are too simplistic, encourage me to more fully state my ideas.”
He then models the process, showing how he might interpret and intentionally integrate AI feedback (I). Mr. Thompson uses a sample essay he has been writing about how local action can stop climate change. As one example of the exchange, the AI chatbot returns the following:
Opponent’s Challenge: “Your focus on localized solutions ignores the fundamental nature of atmospheric chemistry. Carbon emissions don’t stop at state borders. A state can enact the strictest emission standards in the country, but if its neighboring state continues to burn cheap coal, the air quality and global temperature rise affect both equally. A fragmented patchwork of 50 different state regulations creates a ‘race to the bottom’ where polluting industries simply relocate across state lines to avoid local mandates. Only a uniform, centralized federal mandate can enforce accountability across all jurisdictions.”Your Task: Articulate a concession and pivot.• Concession: Genuinely acknowledge the opponent’s valid point regarding transboundary pollution or regulatory inconsistency.• Pivot: Immediately redirect the argument back to why local governance and regional specialization remain essential and more effective in addressing the crisis.
Finally, after they have made some revisions, Mr. Thompson asks students to reflect on the interchange with AI. “What new perspectives did AI offer you that you had not considered before? What did AI offer that you had already considered? Finally, which move did you decide to make in your writing after chatting with AI and why?” (R).
Through this process, Mr. Thompson helps his students to more clearly understand the evidence being presented in their own argument and to further refine their argumentation skills. They also get targeted feedback specific to their topic. With this use of AI, Mr. Thompson has kept the lesson moving forward, getting the class over what the WestEd Digital Fluency group and the National Writing Project describe as an “access barrier” (understanding the complicated nature of argument writing) and improving their “activation energy” (by providing them with options for next steps in revision) (2025).
Incorporating AI: When, Why, How, and to What Extent?
We know that using AI has the potential to make writing “easy,” entirely removing the productive struggle. As educators, we can teach our students how to partner with AI as a co-writer, ensuring that they maintain their authorial agency and can articulate when, why, how, and to what extent they have used AI in a world that will continue to demand these skills of them. By using the PAIR protocol, students can learn to get the help they need and preserve their own writer choices. The goal of using AI is not to remove the productive struggle; if anything, it should be for writers to choose the productive struggles that matter most for them.
