Last updated: 13 October 2025
Estimated read time: 8–10 minutes
TL;DR: Generic feedback wastes effort and doesn’t move learning forward. In this guide you’ll get copy-ready prompts, subject-specific examples, and a starter comment bank so ChatGPT can produce feedback that is Specific, Actionable, and Kind — aligned to your rubric and your teacher voice.
Why AI feedback often sounds generic
The SAK principles of great feedback
Set up your rubric → structured outputs
Copy-ready prompt patterns
Real examples (before → after)
Teacher frameworks you already use
Make it sound like you
Quality control & ethics
Downloadables
FAQ
Most teachers’ first try with ChatGPT yields feedback that is… fine. Polite, wordy, and vague. That happens because the model defaults to safe language unless you constrain it with the right ingredients:
Missing rubric: No success criteria = no precision.
No grade/subject context: It can’t pitch advice at the right level.
No framework: Without WWW/EBI or similar, feedback becomes mushy.
No length/tone constraints: Outputs drift long and formal.
No evidence hook: If it doesn’t reference the student’s actual words, it’s generic.
The fix: give ChatGPT a rubric, a framework, your voice, a length limit, and the student’s answer verbatim.
Keep these three words front and centre. They’re what separate “nice” from “useful.”
Specific — Names what the student did (quote, step, number).
Actionable — Says how to improve on the next attempt.
Kind — Uses encouraging, plain language that sustains motivation.
Pro tip: Ask for 2 bullet-point next steps tailored to the student’s exact response. It forces the model to be concrete.
You already have success criteria in your head. Lift them out into bullets. Three to six is perfect.
Example rubric bullets (English GCSE literature paragraph)
Clear thesis and topic sentences
Evidence selection and word-level analysis
Logical paragraph structure and link-backs
Formal tone and precise vocabulary
Micro-checklist for each run
Subject + grade (+ exam board if relevant)
3–6 rubric bullets
Framework (e.g., WWW/EBI)
Tone + length (e.g., 80–120 words)
Student answer (verbatim paste)
You are a [subject & grade] teacher. Using the rubric below, write SPECIFIC, ACTIONABLE, and KIND feedback.
Rubric:
- [3–6 bullets]
Framework: WWW/EBI (What Went Well / Even Better If).
Voice & Tone: [supportive, concise, UK spelling]. Sounds like [a calm, practical teacher].
Length: 80–120 words.
Include: 2 bullet-point next steps tailored to THIS student’s answer.
Quote the student's wording or data where helpful.
Student Answer:
"""
[paste the student's answer verbatim]"""
Organise feedback by rubric criteria (1–3). Under each, provide:
- one What Went Well,
- one Even Better If,
- one concrete Next Step.
Keep total length to 120 words.
Mimic this teacher voice (style, not wording):
"[paste 2–3 short, authentic comments you've written]"
Match sentence rhythm and warmth. Use UK spelling. Avoid filler phrases.
Student answer (extract)
Macbeth is ambitious and that makes him do bad things. The quote “vaulting ambition” shows he wants power. He listens to Lady Macbeth and then kills Duncan which proves ambition is dangerous.
Before (generic feedback)
“Good understanding. Try to analyse the quote more and add more detail. Watch repetition.”
After (SAK-aligned feedback)
WWW: Your choice of “vaulting ambition” is relevant and links to Macbeth’s actions.
EBI: Analyse the wording: “vaulting” suggests ambition that overreaches and collapses, foreshadowing Macbeth’s loss of control. Reduce plot summary; push into how language creates meaning.
Next steps:
Add a sentence starting “The verb ‘vaulting’ implies…” and connect it to consequences.
Integrate one more quotation and analyse a key word to deepen interpretation.
Student answer (extract)
The chip gains mass in water because of osmosis.
Before (generic feedback)
“Right idea. Add more detail.”
After (SAK-aligned feedback)
WWW: You correctly identify osmosis as the process.
EBI: Sequence the steps and use precise terms (water potential, partially permeable membrane).
Next steps:
Write a 3-step explanation: (1) higher water potential in solution than potato cells; (2) water moves down its water potential gradient by osmosis through a partially permeable membrane; (3) cells gain water, so mass increases.
Cite one value from your results (e.g., “mass increased by 0.8 g”) to evidence your claim.
Question
Solve: 2x + 5 = 17
Student answer
(No workings, just “x = 6”)
Before (generic feedback)
“Correct answer but show working.”
After (SAK-aligned feedback)
WWW: You reached the correct solution.
EBI: Show each inverse operation on a new line and check by substitution.
Next steps:
Write: 2x + 5 = 17 → 2x = 12 → x = 6.
Substitute: 2(6) + 5 = 17 ✓. Add units/context if given.
Put familiar structures in the prompt so ChatGPT sticks to them.
| Framework | Use it to… | Prompt nudge |
|---|---|---|
| WWW/EBI (What Went Well / Even Better If) | Balance praise and direction | “Give 1 WWW, 1 EBI, and 2 next steps tied to the student’s words.” |
| TAG (Tell / Ask / Give) | Encourage reflection | “Tell what was done well, ask a clarifying question, give one improvement.” |
| SBI (Situation / Behaviour / Impact) | Practicals and presentations | “Describe the situation, the specific behaviour, and the impact on the result.” |
AI is most effective when it sounds like your classroom voice.
Share 2–3 real comments you’ve written.
Specify tone (e.g., “warm, succinct, plain English”).
Set boundaries (e.g., “avoid clichés like ‘in conclusion’”).
Keep it short (e.g., “100 words maximum”).
Voice snippet example you can paste
I value clarity over jargon.
I praise evidence use, not length.
I give one precise next step, not three vague ones.
Hallucination guardrails: Require the model to quote student wording or cite data (“Use one exact figure from the graph”).
Bias & fairness: Tie every judgement to the rubric; avoid subjective labels.
Privacy: Do not paste personal identifiers; check your school/department AI policy.
Human in the loop: You approve, trim, and personalise before students see it.
Versioning: Keep a short record of original AI output + your edit (helps moderation and consistency).
Cross-subject WWW/EBI/Next Steps you can import into your comment bank or LMS.
Columns included: subject, level, skill, what_went_well, even_better_if, next_steps, prompt_snippet
Subjects covered: English, Biology, Maths, History, Geography, Chemistry, Physics, Business, French, Spanish, Computer Science.
Can ChatGPT align feedback to my rubric?
Yes — paste 3–6 rubric bullets and require feedback organised by criteria. Ask for WWW/EBI and 2 concrete next steps.
How do I stop AI giving generic feedback?
Constrain length, require references to the student’s actual wording/data, and ask for actionable next steps, not vague advice.
Is AI-generated feedback allowed at my school?
Check your policy. Many schools allow AI as a drafting assistant with teacher review. Never paste personal data.
Can I keep my teacher voice?
Provide 2–3 authentic comments as a style reference. Ask the model to match rhythm and tone, not copy phrases.
Can I bulk-generate feedback for a whole class?
Yes. Use a spreadsheet export of answers, apply the same prompt structure, and keep a teacher review step. Alternatively, try a tool like Marking.ai to do this for you.
Prefer this done for you? Marking.ai generates class-wide, rubric-aligned feedback in minutes — while you stay in control.
Start free and keep your workflow: upload assessment, paste rubric, review outputs, publish to students.