My student sounds like AI, what do you think?

Text Authenticity
, ,
1

I’m seeing more “AI detector score” drama in my classes and I’m trying to handle it without turning into the writing police.

Here’s the scenario (and yes, it’s messy). A student turned in a short lit response that *sounds* polished. Another staff member ran it through a detector and it came back “high likelihood AI.” The student swears they wrote it, but also admits they used a grammar checker and rewrote the intro like 6 times. I don’t want to punish someone for being… careful?

“The author’s choices create a quiet pressure that builds across the narrative, nudging the reader toward empathy without demanding it outright. The imagery works less as decoration and more as a kind of moral weather.”

If you saw that out of context, would you assume AI? I honestly can’t tell anymore. And I’m not convinced “sounds like AI” is fair evidence.

So, practically: how do teachers check for ai in a way that’s defensible? If a student asks me “how to check if something was written by ai,” what’s the least harmful answer?

Also curious what you’d do if a student asks, flat-out: “is this text ai generated?” Do you treat that as a learning moment, or as a red flag?

2

That snippet could be a try-hard human.
I’d be nervous submitting it either way.

3

If you want something you can actually defend, don’t anchor on “the text feels model-ish.” That’s basically style prejudice, and the incentive it creates is weird: students learn to add clunky mistakes and random tangents to look “human.”

What worked better in my org (different context, same problem) was process-based checks that don’t accuse anyone by default. Like: require a short drafting trail, or ask for a 90-second explanation of two sentences and why they’re there. If they can explain the choices, point to where the idea came from, and answer a couple follow-ups, you’re verifying authorship without pretending a detector is a lie detector.

And if they did use tools? That becomes a policy + disclosure convo, not a “gotcha.”

4

The “moral weather” line reads like someone who’s been revising too much and is trying to land on one memorable phrase.

AI can do it, yeah, but I wouldn’t convict on that.

5

Ask them what they meant by that metaphor.
Two questions. Done.