Grading essays is the slowest part of teaching. A class set of thirty essays can eat a whole weekend — and most of that time goes to writing the same comments over and over. AI can cut that time roughly in half, but only if you use it the right way: it drafts feedback, you decide the grade.
That is the rule this whole guide rests on. AI gives feedback. The teacher gives the grade. If you let AI decide grades, you will get fast grades — and wrong ones, with no way to explain them to a student or a parent.
What you need before you start (5 minutes)
Have these three things ready:
- A rubric — if you do not have one, Step 1 shows you how to make one with AI in two minutes. A rubric is what keeps AI feedback honest: without it, AI praises everything.
- The essays, without names — remove student names before you paste anything into an AI tool. Check your school’s AI policy first; some schools forbid putting student work into AI tools at all.
- ChatGPT or Claude — both work. Claude handles long essays well because it is built for long documents; ChatGPT is fine for shorter essays and for teachers who already use it.
Do not paste a whole class set into one prompt. AI quality drops sharply after the first few essays, and the feedback starts repeating itself.
Step 1: Write a clear rubric first (5 minutes)
If you already have a rubric, skip this step. If not, ask AI to draft one:
Read the rubric once and fix it before you use it. A vague rubric produces vague feedback — “good organization” means nothing unless you can say what good organization looks like in this essay.
Save the rubric. You will paste it into every feedback prompt, and you will reuse it for the next essay on the same skill.
Step 2: Get draft feedback, not grades (10 minutes per batch)
Work one essay at a time. Paste the essay (no name) and the rubric, then use this prompt:
Why “no grade”? Because the moment AI gives a number, you stop thinking — and the number is often wrong. Feedback you can check is useful. A score you cannot defend is not.
Two follow-ups that fix most problems:
- Too vague: “Rewrite the second improvement with a direct quote from the essay.”
- Too long: “Cut this feedback to 100 words. Keep the two strengths and the next step.”
Step 3: Check the feedback, then decide the grade yourself (5 minutes per essay)
Read the essay yourself — yes, the whole essay — and read the AI feedback against it:
- Are the quotes real? AI sometimes quotes sentences that are not in the essay. If a quote does not exist, the feedback is useless.
- Are the strengths real? AI praises fluent writing even when the ideas are weak. Check the ideas, not the grammar.
- Does the feedback match your rubric? If the rubric says “organization: developing” and the AI says the essay is well organized, one of them is wrong — and it is usually the AI.
- What grade would you give? Decide it from the rubric and your own reading, not from the AI’s tone. AI sounds confident about everything.
This step is where the time goes, and it is also where the accuracy comes from. Skipping it saves twenty minutes and costs you a grade you cannot explain.
Step 4: Turn feedback into comments students can use (5 minutes)
- Rewrite the AI feedback in your own words, in your normal tone. Students can tell when a comment was not written by their teacher.
- Keep each comment short: one strength, one improvement, one next step. A wall of text does not get read.
- Paste the comment into Classroom or your marking document, next to the grade you decided.
- Keep a note of the patterns. If ten essays all struggle with the same thing, that is tomorrow’s lesson — not ten identical comments.
A good comment tells a student exactly what to do differently next time. “Good work, add more detail” does not. “Your second paragraph needs one example from the text to support your claim” does.
A complete example
Here is the workflow filled in, so you can see what “specific” looks like:
Then, for each essay, use the feedback prompt above with the rubric and the essay text pasted in — no name attached. The AI returns two strengths, two improvements, and a next step. You read the essay, check the quotes, decide the grade from the rubric, and write the comment in your own words. The same two prompts work for any essay type — change the topic and the criteria, and the workflow stays the same.
Common mistakes to avoid
- Letting AI assign the grade. A grade is a professional judgement. If a student or parent asks why, “the AI gave you a B” is not an answer.
- Pasting student names or personal data into an AI tool. Remove names first, and check your school’s policy before you paste any student work.
- Grading a whole class set in one prompt. Quality drops after the first few essays and the feedback starts repeating. One essay at a time.
- Trusting AI on facts and quotes. AI invents quotes and misreads arguments. Check every example it gives against the actual essay.
- Using AI feedback word for word. Students know your voice. Feedback in someone else’s voice reads like a form letter — and gets ignored like one.
FAQ
Is it ethical to use AI for grading?
Yes — if you read the work yourself, decide the grade yourself, and follow your school’s policy on disclosure. It is not ethical to let AI decide grades, or to paste student work into AI tools when your school forbids it. When in doubt, ask your school before you start.
ChatGPT or Claude — which is better for grading essays?
Both work. Claude is strong with long documents, so it suits long essays. ChatGPT is fine for shorter essays and for teachers who already use it for lesson plans. The rubric matters more than the tool — a good rubric with either tool beats a vague prompt with the best tool.
Can AI detect whether a student used AI to write the essay?
Not reliably. AI detectors make mistakes in both directions, and a detector score is not proof. If you suspect AI-written work, talk to the student and ask them to explain their argument — that tells you far more than a score.
How much time does this actually save?
Roughly half, on the first pass. You still read every essay and decide every grade. What disappears is the slow part: staring at a blank comment box, trying to phrase the same feedback for the tenth time.
