Approaching Writing Assignments with AI in Mind
Civics of Technology Announcements
Annual Conference: We are holding our 5th Annual Conference on August 6th-7th, 2026! Our keynotes this year are Dr. Meredith Broussard (data journalist & author of More than a Glitch among other books) and Natasha Singer (NYT edtech reporter & author of the forthcoming book, "Coding Kids: Big Tech's Battle to Remake Public Schools”). Learn more here, register here, and submit proposals here.
Book Club in June: Techno-Negative: A Long History of Refusing the Machine by Thomas Dekeyser. Join us on Wednesday, June 24th at 7:00 PM Eastern Time.Register here.
Post by Jacob Pleasants
A little more than a week ago, I ran a workshop through our university’s Center for Faculty Excellence on “Redesigning Major Writing Assignments with AI in Mind.” This is an area of real need at my institution (and, I suspect, many others). Faculty are feeling immense pain and frustration. They don’t want to walk away from what they know is a foundational part of students’ education, but wading through AI-generated swill has become all but untenable.
Importantly: These faculty are not (by and large) teaching “writing” courses. They are teaching courses in History, Music, Journalism, Engineering, Architecture, Public Health — all of which involve writing and, often, researching (another AI minefield).
So, I put together a workshop. No easy answers were promised. But I did promise to present ideas and possibilities that people could apply to a specific writing assignment and move it in a better direction.
It was a half-day session that featured some pretty deep collaborative work, and I cannot share everything we did. But I figured I would share some of the big ideas that others might find useful either in their own teaching practice or in their faculty/teacher development work.
Foundational Principles
I began the session by laying out a few guiding principles for the session, beginning with some points affirming the value of writing (and writing assignments). Not that I needed to convince my attendees of this; if they wanted to abandon their writing assignments, they would not have come to the session! But nevertheless, these seemed like good places to start:
Writing is not reducible to the “production of text”
Writing is not merely the inscription of pre-existing “ideas in our heads”
Writing is thinking
Machines do not write
Note: I credit John Warner for these bullets
I had attendees collaboratively articulate what they want students to learn from the process of writing, and what they are generally trying to assess through students’ writing. Consensus pretty quickly formed around some fundamental purposes. We want students to be able to communicate, formulate arguments, critically engage with sources, and utilize ideas from their courses. Writing is pretty great for all those things! And generative AI poses a significant challenge to all of those things.
In contemplating what to do about this, I put forward one more guiding principle: There are many paths one can take as an instructor, but I for one want to take a path that avoids battling my students. That may seem obvious, but many approaches actually adopt a combative frame. Set a clear policy, monitor for compliance, vigorously enforce against noncompliance. That sounds like a battle to me, and one I don’t really want to fight. Not only is it deeply taxing for both instructor and students, it’s also not a fight where our tools are particularly good. AI detectors remain weak, although I did mention Pangram, which is certainly an interesting technical project.
Understanding Students
Before getting into specific approaches, it’s also worth spending some time trying to understand students’ experiences and decisions. Below are just a few of the points I highlighted, and come from conversations with lots of different students and individuals who work closely with student writers (including folks at our campus Writing Center). It wasn’t published at the time of the workshop, but I would also strongly recommend this great piece on this topic.
Students experience a whole lot of anxiety and fear. Even for those who refuse to use generative AI, the fear of accusation is real. Some students report intentionally writing differently (e.g., inserting mechanical errors, aggressively avoiding em-dashes) to avoid even being suspected. These anxieties are especially real and costly for international students.
Many students write using a wild amalgam of different software. They do some voice dictation, copy it into a Google Doc to expand, move it to Grammarly for some refining, then back to something else for final submission. Very few students do a full-scale copy/paste job from an AI chatbot. But plenty of AI gets used (most often Grammarly) within their complicated writing workflow.
Students reach for generative AI when they hit “pain points” in the process: places of uncertainty or difficulty where the stakes are high and self-efficacy is low. This is what decades of research on cheating behaviors tell us. We cannot eliminate pain points from writing, but understanding this will help us anticipate where we need to direct our attention.
Approaching the Writing Assignment: 3 Levers
In preparing this workshop, I tried to organize a wide array of different ways of approaching writing assignments, and landed on three main categories. They’re not totally separate, but I found the scheme helpful for thinking about what sorts of things we can try to do as instructors. I think of them as “levers” that we can operate, though which ones are available and what they will do depends greatly on context. A lever that works well for a small in-person graduate-level course will probably not work so well in a large undergraduate gen ed survey or an asynchronous online course. The key here is that none of these are “solutions” and we should not expect any of them to eliminate problematic uses of AI. The goal is more modest: make things better.
Lever 1: Control the Environment
What are the conditions under which you want students to do their writing? To what extent do you want to try to control/monitor that? Traditionally, we give students a writing assignment and ask them to do it… however they want. But that’s not the only option we have available. Some alternatives:
The Tech Solution of “Process Tracking:” You can require that students write in something with a version history that you can then review. Process Feedback integrates into Google Docs and can give you a summary report on how a document was produced. It’s a way to “invigilate” the assignment. But then again, do you really want to surveil students in this way? It also cannot guard against all of the uses of AI that we might want students to avoid. For instance, they can still ask the AI chatbot for the outline, key ideas, etc. That said, there might be some contexts where this really is an appropriate way to go. One interesting idea that came up during the workshop: This could actually be quite useful for group projects so that students can hold one another accountable for the work.
Bring Writing Into the Classroom: This one is time consuming, but people have been trying this and noting positive results. I’ve done it with a graduate course that I taught recently. Turns out, a lot of students really appreciate writing without the constant distractions that they usually have to contend with at home. So yes, it’s an invigilation method. But it’s not merely that.
Lever 2: Design the Assignment
What exactly are you asking students to do in this assignment? What is the product, and how will it be graded? There is a lot that can be done here to foster greater student buy-in and alleviate some of those pain points. A few possibilities:
The Product is Not Only a Piece of Writing: Perhaps it’s also a short presentation or conversation with the instructor. Maybe the product isn’t just the end point of the writing process but some sort of documentation of the process itself. The writing still needs to happen (we’re not abandoning writing!), but adding on some additional components has multiple benefits. Most obviously, some of these will quickly reveal whether a student has offloaded the mental work. But also, you might find that students have actually thought more than their written work would indicate. We’re not just trying to catch students with this, but providing additional ways to express their thinking.
Weigh the Process More Heavily: If you’re only grading the final product, that communicates to students that the end point is the only thing that really matters. But if you value the journey as much (if not more) than the destination, that should probably be reflected in how you grade it. And that means the assignment itself is not just to create the final thing, but go through a sequence of smaller steps.
Include Critical Engagement with AI: Maybe you don’t want to take AI off the table entirely, and maybe you’d like to add something about AI to the learning goals for this assignment. The most obvious version of this is to require students to generate some AI output during the assignment and critique it (perhaps a chunk of text, an outline, or maybe incorporate it into the search for sources) . They will need to engage with an AI system to some degree if you want them to critically reflect on the experience.
Lever 3: Provide Instructional Support
A lot of my college writing assignments didn’t really have much in the way of instructional support behind them. I was assigned them and expected to just kind of figure it out. The most familiar instructional support is feedback on drafts, either from the instructor or peers. But if we think about those “pain points” for students, it might be wise to include more structured guidance. When students believe they can do things themselves, they are more likely to do so.
Support the Outlining: We know this is a pain point and that students will turn to AI for this. We know that’s a pretty bad idea. Don’t have them do this on their own. Bring it into the classroom and support the process.
Using Sources: It is easy to assume that students can digest primary sources and use them in their writing. This is not an easy thing to do. Teaching students how to read and make sense of disciplinary texts is worth our time. Most students will not just pick it up on their own. This is another pain point as well, and it’s really tempting to just get the AI-generated summary of a challenging text.
Peer Feedback? This is a mainstay for writing assignments, but there are plenty of times when peer feedback does not live up to its potential (will peers actually give decent feedback). Getting feedback from an AI is something that plenty of people have tried, but there is good reason to be doubtful. The PAIRR approach incorporates AI into the process in ways that seem to be productive.
Put it Into Practice
We now look at our particular teaching contexts and the specific assignments we want to ask students to do. What are we trying to achieve with those assignments? What do we want students to get out of them? What are we trying to assess? And which levers do we want to use to achieve those goals?
I asked attendees to work through these questions using this organizer, in partnership with a fellow attendee who served as a “critical friend.” I sat in on a lot of deep and interesting conversations, and everyone in attendance landed on some new ideas that they planned to put into action. Lots of optimism! I will be following up with my attendees during the coming academic year (nothing ever quite goes according to plan). What levers do you plan to use?