What ELA Teachers Should Know About AI Detectors

Every secondary teacher has seen a student turn in work that the student didn’t write themselves. Generative AI is the newest tool for students who want a shortcut, and this has led many teachers to seek a definitive method for determining whether or not a student used artificial intelligence to write for them. We commonly refer to these tools as AI detectors.

As a high school English teacher myself, I really wish AI detectors were the answer we all want. It seems only fair that since students can easily use AI to write for them, teachers should have equally accessible tools to detect AI-generated work. Unfortunately, as with everything in education, the reality is more complicated.

What AI detectors are and how they work

An AI detector is a tool that claims to be able to tell you what percentage of a piece of writing was generated by AI. There is a misconception that AI detectors work by recognizing writing that was previously created by AI, as if all AI-generated writing were stored in some vast database that the detector could consult to see if the piece was cataloged there.

The reality is that AI detectors are built using the same systems that generate AI writing. The detector is going to analyze the writing, looking for statistical patterns of word choice, sentence variation, structure, transitions, and complexity. It is guessing about the percentage of the writing that is AI. This is similar to the way scholars can take a piece of writing or music from hundreds of years ago and by statistical analysis suggest that the writing or music is an unknown work of Shakespeare or Mozart. AI detection is at best a probability and never a certainty.

AI detection is an unreliable and risky proposition, especially in high-stakes situations. If your school or institution is running all student writing through an AI detector, you have likely encountered some false positives. The high-achieving students I work with are genuinely and rightfully concerned that their original writing may be flagged as AI. False positives also happen when a student is relying on a grammar aid or translation assistance to make their original writing more polished. These kinds of issues can disproportionately impact our second language and special education students. Finally, there are false negatives: Just because writing passes an AI detector does not mean the student actually wrote it.

promoting authentic student writing

Promoting authentic writing for our students begins long before they get their writing assignment. Through surveys of my own students, I have found that self-efficacy is very highly correlated with not using AI to cheat. In other words, making sure my students have the knowledge and skills they need to approach the writing task can help ensure that they do the writing themselves. This might look like having students write smaller segments of the assignment in low-stakes ways, sometimes on paper, before they find out those bits of writing will become part of a larger piece. It means making sure students have access to strong models and, if they need them, scaffolds like sentence frames.

I ask all of my students to do all the parts of their writing process in a Google Doc that I provide for the assignment. This way I can check their progress at any time and review the version history if I have any doubts about their writing. Peer review, collaboration, choice of topics, requiring evidence of note-taking and outlining, and asking students to write about unique content are all means to increase authenticity in student writing.

When my students turn in writing assignments, I look for signs of authenticity. I use a Chrome extension called Revision History. It puts a light brown bar at the top of the Google Doc with some key statistics that are easy to glance at. This bar tells me how much time the student spent working on the document, how many edits, how many large copy-and-pastes happened. If I need it, Revision History offers a “Details” button that will play back all of the writing the student did on the document. I also look for the comments from their peer review, and of course I read the student work.

If an AI detector is a regular part of your process, or something that happens automatically through a product that your school is using, you should disclose that to students in your course policies.

Sometimes, even with all of our best practices for promoting authentic writing, a student will still try to pass off AI writing as their own. Some students also still try to turn in work written by a peer or copied from the internet. Even if we had a perfect AI detector, that still wouldn’t stop a small percentage of students from trying to cheat themselves out of an education.