I'm a first-year master's student, not a teacher, and I found the Give Specific, Quote-Backed Feedback on a Student Essay prompt while looking for something to grade my classmates' work. Instead I started running my own drafts through it, which is a slightly humbling thing to do voluntarily, like asking a brutally honest friend to read your diary. It turns out being your own toughest grader is easier when the grader is a model that has no stake in protecting your feelings and a rule against rewriting your work for you.
Why I can't edit my own thesis
Every writer knows their own argument too well to see where it's vague, because in their head it's perfectly clear. My thesis sentences read fine to me and like fog to everyone else, because I'm reading the version in my mind, not the version on the page. What I needed was a stranger who'd quote my actual sentence and ask 'what does this commit you to argue?' — and do it at midnight, for free, on a draft I was too embarrassed to show a human yet.
So I paste my draft into Claude Opus 4.8 with my own assignment brief and set the grade level to 'graduate seminar.' The feedback comes back with my own clunky sentences quoted at me. 'This claim — [my actual fuzzy sentence] — asserts something is important but never says to whom or why; specify the stakes.' Reading my own bad sentence in quotation marks, with the problem named, is a uniquely effective form of motivation. There's no arguing with it. It's right there in my voice.
What makes this work where 'AI, improve my essay' completely fails is the no-rewrite rule baked into the prompt. If I asked a model to fix my draft, it would hand me a polished paragraph in its voice, I'd paste it in, and I would have learned absolutely nothing except that I can paste. The thing I'm in grad school to get better at — arguing clearly — would still be exactly as weak. Because this prompt refuses to write for the student, and I'm treating myself as the student, it forces me to fix my own sentence. The fix is mine. The skill transfers. That distinction is the entire reason I use this instead of just asking for a rewrite, and it's the difference between a crutch and a coach.
The advanced move: priority-stacking
Here's the tweak that made this a real workflow instead of a novelty. The first time, I set the feedback priority to 'all areas equally' and got a thorough, overwhelming report that I mostly ignored because everything felt equally urgent, which means nothing did. Now I run it in passes. First pass, priority 'thesis and argument first' — I fix only that and nothing else. Second pass on the revised draft, priority 'structure and transitions.' Final pass, 'writing quality.' Each pass the 'fix these three first' list is short and I actually do all three, because three is a number a tired person can finish.
- Running it in priority passes beats one giant report — fix argument before you polish sentences you might cut
- Seeing my own weak sentence quoted back is more motivating than any abstract rubric
- The 'never rewrite for the student' rule means I keep my voice — it points, I write
- Setting the grade level to my actual program tunes the rigor up to where it should be
The caveat, because there's always one
It is not a substitute for a human reader who knows my field. Twice it 'corrected' a discipline-specific term I'd used correctly, because the convention in my subfield is weird, and a less confident version of me might've changed it. So I treat it as the relentless first reader who catches the obvious stuff, and I save my one human advisor's limited patience for the subtle stuff only a person in my field can see. That division of labor is the whole trick: let the model handle 'this sentence is vague,' let the human handle 'this idea is wrong.'
There's also a quieter benefit I didn't anticipate, which is dignity. Showing a half-baked draft to a professor whose respect you want is genuinely stressful, and that stress made me sit on drafts too long, polishing the same three paragraphs because I was scared to expose the weak ones. Running it past a model first means I show up to office hours with a draft that's already past its embarrassing phase. I've gotten over my hangup about asking it to read my worst work, because it has no opinion of me, it will never bring up paragraph four at a department party, and it answers at one in the morning when the panic actually strikes.
My drafts go to my advisor cleaner now, which means his feedback is about ideas instead of basic clarity, which means I learn more. When a draft is for a tricky concept I'm still figuring out, I'll first run the topic through my Explain Any Concept at Two Levels (Kid Version + The Honest Version) prompt to make sure I understand it before I try to argue it. Grab the Give Specific, Quote-Backed Feedback on a Student Essay prompt on Prompt Dock and point it at your own next draft — it's the honest reader you've been avoiding.