Teaching writing is an act of faith. You assign an essay, thirty of them come back, and you spend your weekend writing 'Good effort! Watch your thesis statement' in thirty slightly different fonts of exhaustion. Somewhere around essay twenty-two you realize you're writing the exact comment you wrote on essay six, you know it didn't help essay six, and you write it anyway because it's eleven at night and you are a husk. I did this for years. I called it grading. It was closer to apologizing in advance.
'Watch your thesis' is not feedback, it's a vibe
The problem with 'watch your thesis' is that the student has no idea what 'watch' means. Watch it do what? What specifically is wrong with it? What would a better one even look like? Without answers, the kid revises blind — they change something random and hope, or they change nothing because they genuinely don't know where to start, and then I'm annoyed the next draft didn't improve. I created that outcome. My feedback was a riddle and I graded them on solving it.
Good feedback quotes the student's own words back at them and says exactly what those words are doing wrong. 'This sentence — "There were many causes of the Civil War" — makes a claim so broad it commits you to nothing; name the two causes you'll actually argue.' That's a comment a fourteen-year-old can act on tonight. It's also the comment I wanted to write for essay twenty-two and did not have the spoons for.
How Claude Opus 4.8 changed my Sunday
I now paste each essay into Claude Opus 4.8 with the assignment brief and grade level using the Give Specific, Quote-Backed Feedback on a Student Essay prompt. Opus is genuinely good at the careful-reading part — it pulls the student's actual sentences, names the specific issue, and suggests a fix without doing the writing for them, which was my hard line. I will not hand a kid a rewritten paragraph; that teaches them I write better than they do, which they already suspect, instead of teaching them to write. The prompt is built to coach, not replace, and it holds that line. I tested it on purpose once, practically daring it to just fix a weak paragraph for me, and it pointed at the problem and suggested an approach but left the writing to the imaginary student. Good.
The opening strength is sneakily the most important rule in the whole thing. The prompt forces it to start with a genuine, specifically-quoted thing the student did well — not 'nice job!' but 'the way you opened paragraph two with a concrete example pulls the reader in.' Fourteen-year-olds are braced for a beating when they see teacher comments. If the first thing they read is specific praise of their actual words, the defensiveness drops and they actually absorb the criticism that follows. I knew this in theory for years and forgot it every weekend out of exhaustion. The prompt remembers it on essay twenty-two as well as essay one, which is its real superpower: it doesn't get tired.
- Quoting the student's real line makes feedback impossible to wave off or misread
- The opening genuine strength (quoted, specific) changes how a defensive teenager receives everything after it
- 'Fix these three first' gives a priority instead of a demoralizing wall of every problem at once
- Weighting toward one feedback priority means the comments have a focus instead of being thirty equally-shallow notes
- It never rewrites the paragraph for them, so the student keeps the work and learns from doing it
What I still do myself, every time
This is where I get preachy, so bear with me. I read every essay. The AI does not understand what a particular student was reaching for — the kid who's clearly trying out a risky structure, the one whose 'error' is actually a voice I want to encourage. Twice the prompt flagged a 'run-on sentence' that was a deliberate, working rhetorical choice, and I cut that note because punishing a kid for the one bold thing they did would be educational malpractice. I also tweak the tone; the defaults can run a touch clinical for an anxious ninth-grader, so I'll soften a line by hand. The model drafts; I'm still the teacher.
I also have a rule that I never let the AI assign the grade. It writes the feedback; I decide the number, because the number carries weight and consequences and that decision is mine to own. The prompt is built for feedback, not grading, which fits how I think about it: the comment is the teaching, the grade is just the receipt. Keeping that line clear in my own head is what lets me use the tool without feeling like I've outsourced the part of the job that's actually mine.
The numbers, honestly: thirty essays used to eat a full weekend and leave me resentful. Now it's an evening, and — this is the part that matters — the comments are better than the ones I wrote when I had infinite time, because they're specific and quoted and prioritized instead of tired. My students revise more because they finally know what to revise, and a class that revises is a class that's actually learning to write instead of just producing one draft and moving on. When I want to turn a batch of common errors into a mini-lesson, I feed the patterns into my Build a Ready-to-Teach Lesson Plan from One Learning Objective prompt and teach the fix to the whole room. Grab the Give Specific, Quote-Backed Feedback on a Student Essay prompt on Prompt Dock and give your students the comment you've been too tired to write.