Anonymize the essay. Paste the rubric. Generate feedback that maps to your criteria, not generic platitudes. You keep the grade decision — AI just drafts the comments you would have written if you had three more hours.
Take one anonymized student essay and one of your existing rubrics, and produce specific, criterion-by-criterion feedback in under five minutes per essay — while keeping the actual grade decision in your professional judgment.
Every English and humanities teacher knows the math. You assigned a five-paragraph essay to 120 students. Each essay deserves at least a paragraph of specific, evidence-based feedback. That is 120 paragraphs of original prose written by you, on top of teaching, on top of planning, on top of life. The honest reality is most teachers cut corners somewhere — either by giving short comments, by recycling stock comments, or by skipping individual feedback for some students entirely.
The promise of AI feedback is that you can give every student a meaningful, criterion-aligned response without giving up your weekend. The risk is that AI without rubric grounding writes feedback that is fluent, encouraging, and structurally identical for every essay. That is worse than short comments because it pretends to be personalized while saying nothing.
Rubric-aligned feedback is the fix. When the AI tool is forced to evaluate the essay against your specific rubric criteria, the output stops being generic praise and starts being a paragraph that actually maps to where this essay sits on your rubric.
Before you do anything else, strip the essay of identifying information. The simplest workflow is a single search-and-replace in your word processor:
[STUDENT][NAME], [SCHOOL], [PLACE]This is non-negotiable. Under the Family Educational Rights and Privacy Act (20 U.S.C. § 1232g), the essay-with-name attached is part of the student's education record. Pasting that record into a consumer AI tool that has not signed a FERPA-compliant data-processing agreement with your district is a disclosure problem regardless of the tool's marketing claims. The anonymized essay, by contrast, is just text — no longer linked to an identifiable student.
Here is the prompt template. The trick is putting your actual rubric criteria into the prompt verbatim, not paraphrasing them.
You are an experienced [GRADE LEVEL] [SUBJECT] teacher giving feedback
on a student essay. The student is anonymous to you.
RUBRIC (verbatim from my classroom rubric):
1. Thesis: [paste your rubric language for thesis, all four levels]
2. Evidence: [paste rubric language for evidence]
3. Analysis: [paste rubric language for analysis]
4. Organization: [paste rubric language]
5. Conventions: [paste rubric language]
ESSAY (anonymized):
[Paste anonymized essay here.]
OUTPUT REQUIREMENTS:
- For each rubric criterion, write 2-3 sentences of specific, evidence-based
feedback referencing exact phrases or sentences from the essay.
- Do NOT assign a grade or rubric score. The teacher decides that.
- Do NOT use generic praise. Every comment must point to specific text.
- End with one "next steps" sentence: the single most important
revision the student should attempt before turning in their next essay.
- Tone: warm, direct, professional. Address the student as "you."
- No more than 250 words total.
Run this prompt with one essay. Read the output. Edit it. Compare it to what you would have written. The first few times you do this you will catch the model misreading evidence, overstating thesis quality, or being too gentle on conventions. That is fine — the editing pass is where your professional judgment shows up. After ten essays the editing pass shrinks dramatically because you have learned where the model is reliable and where it is not.
This matters. The grade decision is yours. The AI tool is drafting the prose; it is not deciding the score. There are pedagogical, legal, and union-contract reasons for this depending on your district. Pedagogically, you are the one who knows whether this student has been making progress. Legally, in most districts the teacher of record signs the gradebook, and assigning a grade by algorithm without your professional review may violate district grading policy or your collective-bargaining agreement.
The discipline here is to use AI as a draft engine for the comments and as a structured reader pointing at evidence in the essay — never as the grader. When you sit down with the AI-drafted feedback, you are the editor, not the consumer.
Take five anonymized essays from a recent assignment. Run the rubric-grounded prompt on all five. Compare the AI-drafted feedback to the feedback you actually wrote for those students. Where does the model see something you missed? Where does the model miss something you caught? The gap between the two is your honest report card on AI feedback for your subject and your students.
Day 3 takes the differentiation thread one level deeper: adapting content for students with IEPs.