How I convinced my students that I could tell if they used AI

Part of preventing students from using generative AI to do their work for them is showing them ways it can be helpful without violating academic integrity. I wrote about that process here. 

But another big part of keeping students from using the AI unethically, is convincing them that I will be able to tell they "cheated." This is how I showed my 9th graders that I could tell when they used AI to respond to a question. 

One of my colleagues came to me because she thought one of her students had used ChatGPT for one of the short paragraphs about a book we were reading as part of a novel study. We talked about the student, why they might have done that, why she thought it was AI, etc. And then I ran it through an AI detector. As predicted by this expert educator colleague of mine, it was flagged as entirely AI written. She pursued her own actions at that point, but I saw an opportunity. 

The Opportunity: 

I took the AI generated response and put it on a page with two other responses I knew my students had actually written. I printed out 18 copies of the page with the three paragraphs and gave it to my students in pairs. It seemed important to do this on paper.

This is the part where I was exceptionally clever. At first I did not tell my students why I really gave them this paper. I just said, "Here are some student answers I want you to review."

Then, I told them I wanted them to score those answers. I use a four point grading scale and students are always asking what makes an answer a 4. So, they thought they were working to understand what made one of those paragraphs better than the others. My students worked on this diligently for several minutes while I circulated and took an interest in how they scored various paragraphs. 

Finding the AI Paragraph:

Eventually, they wanted to know what scores I gave these paragraphs. I admitted that this activity was not really about scoring the paragraphs, and I asked them to look again and see if they could tell which was written by AI. I asked them not to say anything, but just to mark that one on their paper. Then I circulated again and noticed which paragraph each partner group had picked. All of them had correctly, quickly, and decisively identified the second paragraph as AI generated. 

They were quite proud of themselves when I said that all of them had picked the AI paragraph correctly. 

The Message: 

When they stopped congratulating each other and I had their attention again, I made my point very simply. I said, "If you can tell which one is AI that easily...so can I."

The room went quiet. I don't think very many of my 9th graders were planning to use ChatGPT, but for those who were considering it, I knew this would be a memorable moment. 

You Can Try This At Home (or at school):

You don't have to wait for a student to turn in work done by AI. You can use ChatGPT to generate an AI response to a short assignment and recreate what I did above, mixing two or three pieces of real student work with the AI writing. The trick is to make sure the AI version is grammatically perfect, but not that great of an answer to the assignment. You want students to be able to easily identify the AI part once you tell them that's the goal. Don't make that hard by using lots of real work examples. Three samples will make it easy for them to unanimously find the AI version. 

Note: 

I think I am pretty good at knowing if a student plagiarized, or used AI to do their writing for them. But for this activity my goal was to show them that I could spot it easily. Their perception mattered more than my actual skill in this case.