Here’s the math that used to keep me up at night. Thirty-two students. Five paragraphs each. Two essay assignments a month. That’s three hundred and twenty paragraphs of feedback I was supposed to write — thoughtfully, specifically, in a way that actually helped each student improve.
I wasn’t doing it. Nobody was doing it. We were all writing “good analysis” and “needs more evidence” and calling it feedback.
AI changed that for me. Not by writing the feedback, but by making real feedback possible again.
The Problem Wasn’t Laziness
Before I get into the how, I want to be clear about the why. Teachers aren’t giving shallow feedback because they don’t care. They’re giving shallow feedback because the math doesn’t work. There are only so many hours in a day, and a thoughtful written response to a student essay takes real time — time most teachers don’t have when they’re managing courses, planning lessons, and doing everything else that doesn’t show up in a job description.
The result is a slow erosion. You start with good intentions, you have detailed feedback on draft one, and by draft three of a different unit you’re scrawling a check-plus and moving on. Students get the impression their writing isn’t that important — because the feedback they receive suggests it isn’t.
This isn’t a character flaw. It’s a structural problem. And AI can help solve structural problems.
What I Actually Do Now
My process isn’t complicated, but it took a few iterations to get right.
I paste a student’s paragraph (or a full short essay) into a conversation with an AI model — Claude, in my case — and I give it a specific prompt. Not “give feedback on this essay.” Something much more targeted: Identify the strongest analytical move this student makes, and identify the one place where their reasoning loses me. Be specific — quote the actual sentences.
Then I read what the AI produces. I agree with it, disagree with it, add to it, or throw it out entirely and write my own. Usually I’m editing and extending. Sometimes the AI catches something I’d missed in a stack of thirty. Sometimes it misunderstands what the student was trying to do, and I correct that in my final response.
The feedback the student receives is still mine. It reflects my read of their work and my knowledge of where they are in the year. But it’s more specific than what I would have written at 10pm with thirty papers left to go.
The Voice Problem
The thing teachers worry about, and rightly, is losing their voice in feedback. Students can feel the difference between a comment written by someone who knows them and a generic observation that could apply to anyone. If feedback starts feeling boilerplate — even well-written boilerplate — it loses its effect.
This is why I don’t use AI to generate final feedback. I use it to generate a first pass that I then make my own. The key is staying in the driver’s seat. AI should help you think faster, not think instead of you.
A few things that help with this: I always add something specific to the student’s situation — a reference to our class discussion, a callback to a previous draft, something that shows I actually read their work. I cut anything that sounds generic. And I read the feedback aloud before sending it. If it doesn’t sound like me, I rewrite it.
The Deeper Point
What this changed, beyond the hours, is the quality of what I can afford to assign. When feedback is a burden, you unconsciously start assigning things that are easy to grade — multiple choice, fill-in-the-blank, formulaic responses. You stop assigning the messy, open-ended writing that actually develops thinking, because the grading will wreck you.
When feedback becomes manageable again, you can assign the hard stuff. You can give students a genuine writing task and give them genuine feedback on it. That loop — write, get real response, revise, grow — is what actually develops writers.
The scale problem was always a structural obstacle to doing the thing we knew was right. AI didn’t change what good feedback looks like. It changed what’s possible.
This post draws on ideas from The AI Doesn’t Know Your Students, Chapter 9: “Thirty-Two Essays.” The book is available now.
Hi would you like to purchase the domain name shouldiuse.com?
No, thank you.