Artificial Intelligence is rapidly changing education, but not always for the better. One of the most troubling developments is the reliance on AI detection tools to assess whether students have used AI to complete their assignments. The problem? These tools are often wildly inaccurate, inconsistent, and, ironically, create the very environment they seek to prevent—one where students and educators alike must cater their work to the whims of an unthinking algorithm.

A Real-World Example of AI Detector Failures

Recently, my assistant principal (AP), who is working on another degree, called me for advice after being falsely accused of using AI to write a paper. His professor informed him that their institution’s third-party AI detection software flagged his submission as 68% AI-generated.

The issue? My AP is not only a principled educator but also a talented writer. He knew with certainty that he had not used AI to generate any portion of the paper. What he had done, however, was follow the exemplar report given to the class and had accepted a few grammatical suggestions from Microsoft’s built-in tools—like replacing “in spite of” with “despite.” That was the extent of the so-called “AI generation.”

Out of curiosity, he asked if I could run the same paper through my school’s AI detector. The result? It flagged his paper as only 8% AI-generated. For comparison, the exemplar report from his professor’s institution scored 13% AI-generated—a higher percentage than his own paper!

How is it possible that two separate AI detection tools produced such drastically different results? The answer is simple: these tools are not reliable.

A Flawed Approach to Academic Integrity

Given the discrepancy, the professor had a few options. She chose to allow my AP to resubmit his paper with a 15% penalty, provided that it passed another AI detection scan. He refused. To do so would be an admission of wrongdoing when none had occurred.

Ultimately, since the paper was part of a multi-stage assignment, the professor decided to “let it go,” with the caveat that if future submissions were flagged, further action would be taken.

So, let’s get this straight: the professor was comfortable relying entirely on an AI algorithm to do her job because she suspected a student had used an AI algorithm to do his. The irony is staggering.

The Harmful Effects of AI Detection Overreliance

This situation highlights the growing problem with AI detectors in education:

• They shift the focus away from learning. Instead of focusing on improving his work, my AP is now more concerned about avoiding AI detection flags.
• They encourage poor writing. One way to avoid being flagged is to write less fluently and more mechanically—an absurd incentive in an educational setting.
• They force students to cater their writing to flawed algorithms. If one AI detector flags a paper at 68% AI-generated and another at 8%, which one should students trust? The obvious answer is neither.

What Should Educators Do Instead?

Some suggest comparing flagged assignments to prior student writing samples to verify authenticity. This is a reasonable approach—but it requires effort on the part of the professor. The reality is that as AI-assisted writing becomes more ubiquitous, educators may not have purely student-generated samples for comparison.

Instead of relying solely on flawed AI detection software, educators must rethink how they assess student writing. Some effective strategies include:

1. Process-Based Writing Assignments – Requiring drafts, revisions, and in-class writing samples makes it easier to verify authentic student work.
2. Oral Defenses & Discussions – Having students verbally explain their arguments can confirm whether they understand their own writing.
3. Updated Academic Integrity Policies – Institutions must recognize that AI is here to stay and shift the conversation from detection to ethical use of AI tools.

A Final Thought

The reliance on AI detection tools to catch academic dishonesty is not just impractical—it is counterproductive. These tools lack consistency, produce false positives, and ultimately create an adversarial dynamic between students and educators. If we do not address this issue now, we risk creating a system where students are forced to write not for their professors, but for the approval of an algorithm.

And that is not education.