Students are increasingly using ChatGPT and other AI tools to cheat on writing assignments in the undergraduate classes I teach, and in academia overall. I’ve tried a variety of strategies to combat it, from improving my AI detection abilities to designing what I hope are more AI-resistant prompts and trying to change my mindset as an instructor to better adjust to the AI age.
Despite these efforts, inappropriate AI use has increased among my undergraduate students instead of decreasing. I’ve noticed what I believe is AI writing not just in formal assignments, but in discussion board posts and, perhaps most disturbing of all, in written responses to classmates and to me.
I’m not alone. Reliable data is still lacking on how prevalent AI writing is in education, but in casual conversations with colleagues, most have reported similar upticks, with several estimating that more than 50% of writing submissions are now AI-generated.
Recently, I attempted to fight fire with fire, and asked ChatGPT how I could prevent students from using it and other similar tools to write their work for them. ChatGPT responded with a range of suggestions. Some were standard but helpful, others were standard and I think ineffective, while there were also a few tips I haven’t considered implementing.
Here’s a closer look at ChatGPT’s advice for educators such as me who are, ironically, tired of reading and grading papers written by ChatGPT.
Asking ChatGPT How To Limit Student Cheating
For this experiment, I used GP4-4o model, which I get access to through my subscription to ChatGPT Plus.
I started with the following prompt:
Hi, I’m a college writing and English professor. My freshman writing students are frequently using ChatGPT and similar AI tools to write their papers and their online discussion board posts. What can I do to prevent them?
ChatGPT provided a series of suggestions. I then asked it to share an example of an AI-resistant assignment and look at one of my assignments, and followed up with other requests. The advice ChatGPT provided on preventing ChatGPT fell broadly into the following categories.
1. Make Prompts Personal and Specific
What ChatGPT advised: Ask students to connect readings or themes to personal experiences, class discussions, or campus/local events. Use hyper-local or course-specific references: “Refer to our in-class debate on March 5th” or “Discuss your experience attending the campus event on X.”
My take: This is really good advice. I teach online, so it’s not always applicable for me, but overall the principle is sound. Focus on assignments that require information that is not available on the web. This way, even an AI equipped with web-browsing capability can’t come up with satisfactory answers.
However, when ChatGPT provided examples of writing assignments it felt were AI resistant, ChatGPT didn’t quite follow its own advice. For instance, one assignment asked a student to include a personal reflection on what writing the assignment was like. In my experience, those types of questions are easy for AI to duplicate.
2. Require Process Documentation
What ChatGPT advised: Have students submit brainstorming notes, outlines, rough drafts, peer reviews, and reflections. Ask for screenshots of revision history if using Google Docs. Include a short oral presentation or in-class writing follow-up.
My take: This is sound advice. I’ve resisted implementing these types of documentation requirements in my own classes because it’s a cumbersome process for me and feels like an annoying extra step for those students who are not using AI to do their work. However, it seems increasingly necessary, and if structured properly, can help reinforce lessons around outlining and the overall writing process.
3. Use Multimodal or Interactive Assignments
What ChatGPT advised: [Use] podcasts, infographics, video essays, or collaborative wikis to make AI generation harder and student voice more visible.
My take: This is great advice and can make assignments more interesting and fun for students in general. For writing class this is a bit more challenging because my job is to teach students to write, not necessarily to create podcasts or audio content. However, I’m always looking for new and engaging ways to convey writing concepts, and this could do that with the added bonus of limiting the impact of AI.
4. Integrate AI Use Transparently
What ChatGPT advised: Instead of banning AI entirely, consider teaching responsible use: Scaffold AI literacy. Give students assignments that require them to experiment with AI tools and reflect on their usefulness and limits. Example: “Use ChatGPT to generate a rough draft of an introduction, then revise it critically. Highlight your changes and explain your reasoning.”
My take: I like this advice in theory, and teaching AI literacy and appropriate AI use cases is something I support. However, in my experience, it doesn’t work quite as smoothly as this suggestion makes it seem. My AI-approved assignments are just as likely to be AI overused as non-AI-based assignments. Once students are using AI to generate a rough draft, many also use AI to generate their reflection on the assignment, in my experience.
In this vein, ChatGPT did suggest an interesting assignment in which you have students read AI-generated work and student-generated work and guess which was which. The hope is that they will see why the AI-generated work is lacking. I like this suggestion even though the “encourage specific types of AI use,” has not been an effective strategy for me.
5. Use In-Class Writing to Anchor Authentic Work
What ChatGPT advised: Require timed, low-stakes in-class writing assignments or reflections. Use in-class work to build parts of a longer paper. This creates a clear benchmark for each student’s natural voice and fluency.
My take: I like this advice. Though it is hard to implement in the online classes I currently teach, I used to devote a good deal of time to in-class writing when I taught in person for other reasons. Working with students as they write, rather than criticizing them afterward, is a good pedagogical policy overall, I think. Here, it has the added benefit of allowing you to help cut back on AI use.
Overall Takeaways
Like many instructors these days, I spend a lot of time grading ChatGPT-produced work supposedly from students. I generally have a lot of complaints with these assignments, but in this instance, I think ChatGPT earned an A.
The advice it provided was a quick and accurate summary of what most instructors who have thought about these issues would advise, and gave me some food for thought on how I can change my approach to decrease AI use. For example, I will likely spend more time focused on activities such as in-class or synchronous writing and documentation.
The best AI counterstrategy, however, seems to me to be hyperlocalized assignments that require offline reporting or research. If this can be realized, it appears to be the most direct way to beat the machines without making things harder for us instructors or unnecessarily laborious for students.
Even ChatGPT agrees that’s a win-win.