My Own Writing Was Flagged 56% AI — I Had Written Every Word
On a Tuesday afternoon, I sat down with a mug of tea that had gone cold and a stack of student essays to grade. One essay, a personal narrative from a usually quiet ninth grader, had been submitted with a required AI-detector report attached. The tool flagged it as 84% AI-generated. The boy swore he hadn’t used anything except his brain and a Google Doc. I didn’t know what to believe.
So I decided to run a little experiment. I opened the same AI detector, copied the essay I’d written the night before for my own graduate course — a reflective paper about teaching reading comprehension to struggling learners — and hit submit. I hadn’t used AI. I’d typed it in a library, on a laptop with a sticky spacebar, muttering to myself as I revised the third paragraph.
The report came back: 56% AI-generated.
I laughed, then I didn’t. The essay was mine. It had come from lesson notes, a few books, and twenty years of classroom experience. And yet over half of it, machine-checked, looked like it had been written by a bot. That quiet kid in the second row? Maybe he was telling the truth too.
I’ve been thinking about that 56% ever since.
What does that percentage even mean? Many detectors compare patterns in writing — sentence length, vocabulary choices, predictability, transitions — against a database of human and AI-written text. If my sentences are too tidy, too structured, too “balanced,” the algorithm raises its eyebrows. Apparently, years of writing lesson plans and grading rubrics have trained me to sound like an AI. Or, to put it more accurately, AI has been trained to sound like a competent, organized teacher. We’ve built tools that mimic clear communication, and then we’re surprised when clear communication looks artificial.
The problem is we’re treating a probability score as if it were a DNA test. A 56% score sounds specific and scientific, but it’s really just an estimate. No tool can actually detect intention. It can only notice patterns. And patterns are everywhere.
One of my students writes like she speaks — full of fragments and false starts. She got flagged at 12%. Another student, who is simply a neat, formulaic writer, gets flagged at 90% on everything, including the poem he wrote about his grandfather. If I relied on the detector, I’d be punishing structure and rewarding chaos. That’s not pedagogy. That’s guesswork.
I started talking to my students about this openly. I told them about my 56%. A few looked relieved. One boy, the one with the 84% score, nodded and said, “That’s what I’m saying. I just wrote it the way you taught us.” And he wasn’t being defensive. He was genuinely confused. He’d followed every rule I’d ever given him about transitions, topic sentences, and clear organization. Somewhere along the way, he’d become formulaic enough to be mistaken for a machine.
That’s the part that really bothers me. We spend years teaching students to write well, and then we install surveillance software that flags exactly the skills we’re trying to build. Clear thesis? A healthy level of confidence? Smooth transitions? Suspicious. If you write with consistent voice and no spelling errors, you may be a cyborg.
I’m not saying AI detectors should be banned or ignored. But they need to be treated like the blunt instruments they are. A rough scan can point a teacher toward a conversation. It shouldn’t point toward a punishment.
What’s helped me lately is slowing down the process. Instead of running every paper through a detector, I’m asking students to show me a few moments in their writing where they struggled. A deleted sentence in their revision history, a messy paragraph they rewrote, a margin note they ignored. That takes more time, but it’s actual evidence. It’s harder to fake a thought process than a sentence.
I’ve also changed how I assign writing. I give more in-class timed pieces with no internet access. That doesn’t catch every case, but it gives me a baseline. I can see what a student produces under pressure, without a library, without help. When their at-home essay suddenly sounds like a completely different person, that’s worth investigating. And I use detectors sparingly, as a conversation starter, not a verdict.
For parents, my advice is similar. If your child comes home upset about an AI score, ask to see what they actually wrote. Ask them to walk you through their idea and where it came from. You don’t need to be a tech expert or a writing teacher. Just be curious about their thinking. Most kids are honest when they’re not being accused. And if they did use AI, you’ll likely find that out too, because they won’t be able to explain a single thing about the paper.
I haven’t resolved the bigger question. I don’t know how to fairly assess writing in a world where everyone has a fluent, invisible assistant in their pocket. But I know that a detector’s percentage isn’t the truth. It’s a clue, at best. And right now, my own writing is sitting in some database, read by an algorithm that thinks I might be one of them.
That’s a strange thought. Maybe we’re all just teaching children to write like machines, and then accusing them of being machines. Or maybe the machine just learned to write like us. Either way, the next time the software says something is 56% fake, I’m going to ask the student what they meant. Then I’ll ask them to tell me, in their own words, why they chose that first sentence. Because that’s something no algorithm can explain.
At least, not yet.
Please indicate: Thinking In Educating » My Own Writing Was Flagged 56% AI — I Had Written Every Word