Home › Education & learning › Make a grading rubric: how to do it with AI

Make a grading rubric: how to do it with AI

Last checked

From official vendor pages; selected tasks also include dated logged-out free-plan trials. Report an error.

Beginner Risk: Low 🎓 Education & learning

Build an analytic grading rubric from your assignment and learning goals: criteria, 3 or 4 performance levels and a clear descriptor in each cell. AI is good at a first draft and at making descriptors parallel; you check that it matches your curriculum and test it on real student work.

Who it’s for: Teachers, instructors, trainers and anyone who has to grade work consistently.

Why low risk: A rubric draft contains no student data, but a vague or misaligned rubric leads to unfair grading.

Best pick

ChatGPT has the highest score on our published rubric (94/100) of the 4 tools listed for this task. The free plan covers everyday text chats with limited uploads; Plus adds projects, scheduled tasks and custom GPTs (chatgpt.com/pricing, checked Oct 23, 2026). OpenAI's pricing page lists ChatGPT for Teachers, a free plan for verified U.S. K–12 educators through June 2027 (chatgpt.com/pricing, checked Oct 23, 2026).

Step by step

  1. Paste the assignment instructions and the curriculum expectations or learning goals it assesses.
  2. Say how many criteria and levels you want (for example 4 criteria × 4 levels) and the level labels your school uses.
  3. Ask for observable descriptors: what a marker would actually see in the work at each level.
  4. Ask for a student-friendly version in 'I can' language if you'll share it with the class.
  5. Test the rubric on 2 or 3 real pieces of work (without names) and fix any cell that's hard to apply.
  6. Save the final version as a template or reusable prompt for the next assignment.

AI prompts to make a grading rubric

Replace the [brackets] with your details. See all prompt examples

Draft the rubric
Create an analytic rubric as a table for this assignment. Criteria: [list, or 'suggest 4 based on the goals']. Levels: [Level 1–4 / Beginning–Exceeding]. Each cell should describe observable evidence in the work, not effort. Keep the language parallel across levels.

Assignment: [paste]
Learning goals: [paste]
Student version
Rewrite this rubric in 'I can' statements that a [grade] student can understand, keeping the same criteria and levels: [paste rubric]
Stress-test it
Here is my rubric. Point out any cells that overlap, use vague words (like 'good' or 'some'), or would be hard to tell apart when marking, and suggest clearer wording: [paste]
Single-point version
Turn this rubric into a single-point rubric: one column with the 'meets expectations' description per criterion, plus blank 'Strengths' and 'Next steps' columns: [paste]

Illustrative example

You paste

Grade 7 persuasive essay; goals: clear claim, evidence, counter-argument, conventions; 4 levels.

A good answer looks like

Criterion 'Evidence' – Level 4: uses 3+ relevant, accurate pieces of evidence and explains how each supports the claim. Level 3: uses 2 relevant pieces with some explanation. Level 2: evidence is present but loosely connected. Level 1: little or no evidence.

Illustrative example written for this guide, not output captured from a specific tool.

Which tool to pick, and why

Reasons come from the capabilities, free-plan details and prices we've checked on official pages (separate from any dated trial above). Sources: chatgpt.com/pricing · claude.com/pricing · gemini.google/subscriptions/ · www.microsoft.com/en-us/microsoft-copilot/for-individuals

Common mistakes

Privacy tip

A rubric itself has no personal data. If you paste student samples to test it, remove names and identifying details first.

ChatGPT (incl. Work & Codex) setting · Claude (incl. Claude Code) setting · Google Gemini setting · Sources: help.openai.com/en/articles/7730893-data-controls-in-chatgpt · privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings · support.google.com/gemini/answer/13594961

Related tasks

Report an error on this page