Learn
Can teachers use ChatGPT to mark homework? What is fine in India, and a safe way to do it
In short: yes, a teacher can use ChatGPT or another chatbot to help with marking, but only as an assistant. Use it to draft feedback comments against your own marking scheme, never to decide a child's mark on its own, and never with a child's name or other details in what you paste. The mark is yours to give and yours to sign. This page covers marking and feedback only; for which free AI tool suits which teacher task, see AI tools for teachers in India.
What the rules say
| Rule | What it says | What it means for marking |
|---|---|---|
| CBSE | We found no CBSE circular on teachers using AI to mark or grade work (we read the titles of CBSE's 2026 academic circulars on cbseacademic.nic.in on 7 November 2026; the AI ones are about the CT and AI curriculum and an AI literacy course). | Follow your school's own policy; ask your head of school if there is none. |
| OpenAI's usage policies | Among the uses OpenAI does not allow: "automation of high-stakes decisions in sensitive areas without human review", with education listed as one of those areas (usage policies effective 29 October 2025; a new version from 20 November 2026 widens this to "high-stakes actions"). | A chatbot's mark must not stand without you reviewing it. |
| OpenAI's terms on age | A user must be at least 13 (or the minimum age in their country), and users under 18 need a parent's or guardian's permission. | This is about who uses the tool. Your own teacher account is a different matter from asking children to sign up. |
| DPDP Act, 2023 | "Personal data" is "any data about an individual who is identifiable by or in relation to such data", and a child is anyone under 18. | A child's named homework, test or photo is that child's personal data. |
On the DPDP Act: the Act asks companies to get a parent's verifiable consent before processing a child's personal data, and the detailed children's rules start in May 2027 (our guide to children's data under the DPDP Act explains the dates and exceptions). The DPDP Rules give educational institutions an exception, but only for tracking and behavioural monitoring "for the educational activities of such institution" or in the interests of the safety of children enrolled with it. That is the school's own processing; it is not a licence to send children's work to an outside chatbot company. The safe habit is simple: remove the name before you paste.
What a chatbot gets wrong when it marks
Studies that compared chatbot marks with human marks are still few, and most are preprints (not yet peer-reviewed). Two from 2024:
- A study comparing GPT-4 with human graders found its average marks matched the teachers', but its agreement with individual human graders was low, and it showed "a risk-averse grading pattern" (Lundgren, 2024).
- A study of essay grading found that chatbot scores "do not correlate well" with human scores, and that both models it tested generally gave lower scores than the human raters, ChatGPT most of all (Kundu and Barbosa, 2024).
In other words, a chatbot can look right on average and still be wrong for the child in front of you. The problems teachers are most likely to meet:
- It marks the answer, not the method. In maths, a correct final number with a wrong step, or a slip in one step of a correct method, needs a human eye. Our guide on checking an AI's maths answer gives quick tests.
- It does not know your marking scheme unless you give it, and it may drift from it across a long batch.
- It reads handwriting and Indian-language answers less reliably. A photo of a handwritten Hindi answer is easy for it to misread; check what it thinks the child wrote before you trust its comment.
- It sounds sure when it is wrong. A confident comment is not a correct one.
A safe workflow, step by step
- Remove names before you paste. Replace the name, roll number and section with a label such as "Student A". Do not paste photos that show a face, a name on the page or an ID card.
- Give it your marking scheme. Paste the question, the model answer and how marks are split (for example, 1 mark for the formula, 1 for the substitution, 1 for the answer with the unit).
- Ask for feedback, not a mark. Ask what is right, what is missing and one next step for the child, in simple words.
- You decide the mark. Read the answer yourself and give the mark. The chatbot's comment is a draft you may edit or discard.
- Spot-check your own process. Every so often, mark a few answers fully by hand first and then compare with what the chatbot suggested. If they often disagree, stop using it for that kind of work.
- Keep the child's work in school. Do not keep children's answers in your personal chat history longer than needed, and follow your school's rules on where student work is stored.
One reusable feedback prompt
Copy this, fill in the brackets, and paste one answer at a time with the name removed:
> You are helping a teacher in India give feedback on a Class [class] [subject] answer. Do not give a mark. Here is the question: [question]. Here is the model answer and marking scheme: [scheme]. Here is the student's answer: [answer, name removed]. In plain English (or [language]), write three short points: one thing the student did well, one thing that is missing or wrong (quote the student's words), and one next step. If you are not sure what the student wrote, say so instead of guessing.
The last line matters: it asks the chatbot to admit when it cannot read or follow the answer, instead of inventing a comment.
What to tell children and parents
If you use AI to help draft feedback, say so. Children learn more from knowing how their work was checked, and it matches what you ask of them: our guide for students on using ChatGPT for homework without cheating asks children to be open about AI help too. Tell parents that no child's name or personal details go into the tool, and that you decide every mark.
For classroom routines beyond marking, see our practical guide to AI in the Indian classroom.
Where CMDLearn fits
CMDLearn is building an AI tutor for Classes 1 to 10 in Hindi, Marathi and English, designed for government-school classrooms: it explains one step, checks the child's work and leaves the writing to the child. It is still being built and no school or child uses it yet. If you would like to hear when early access opens for teachers, join the waitlist.
Sources
All read on 7 November 2026.
- OpenAI, Usage Policies (effective 29 October 2025: "automation of high-stakes decisions in sensitive areas without human review", education listed; the version effective 20 November 2026 says "high-stakes actions"), read via Internet Archive copies of 15 October and 6 November 2026 because openai.com refuses our fetch: openai.com/policies/usage-policies
- OpenAI, Terms of Use (13 or the minimum age in your country; parent's or guardian's permission under 18): openai.com/policies/row-terms-of-use
- The Digital Personal Data Protection Act, 2023 (personal data in section 2(t); child in section 2(f); children's data in section 9): www.meity.gov.in/static/uploads/2024/06/2bf1f0e9f04e6fb4f8fef35e82c42aa5.pdf
- The Digital Personal Data Protection Rules, 2025 (commencement in rule 1; Fourth Schedule, Part A, item 3, educational institutions): www.meity.gov.in/static/uploads/2025/11/53450e6e5dc0bfa85ebd78686cadad39.pdf
- Lundgren, M., Large Language Models in Student Assessment: Comparing ChatGPT and Human Graders, arXiv preprint 2406.16510, 24 June 2024: arxiv.org/abs/2406.16510
- Kundu, A. and Barbosa, D., Are Large Language Models Good Essay Graders?, arXiv preprint 2409.13120, 19 September 2024: arxiv.org/abs/2409.13120
- CBSE academic circulars page (2026 circular titles read; none on AI in marking): cbseacademic.nic.in/circulars.html
This is general information for teachers, not legal advice. Follow your school's and state's rules on student data and assessment.