The Best AI Prompts for Giving Feedback
A strong feedback prompt names four things a vague one skips: the actual student work being responded to, the specific criteria or skill in focus, the balance of strength and growth points, and a tone that stays specific rather than generically encouraging. Feedback isn't a grade — it's the part of assessment meant to change what a student does next.
Hattie and Timperley's (2007) influential review of feedback research organized effective feedback around three questions: Where am I going? How am I doing? Where to next? Most weak feedback — AI-generated or otherwise — answers only the middle question, leaving a student with a rating but no direction.
Quick Answer: The best feedback-generation prompts include the actual student work, name the specific criteria or skill being addressed, request a balance of one strength and one growth point, and ask for language that points toward a next step rather than just a rating. A working template: "Give feedback on this student work for [specific skill/criteria]. Note one strength, one area to improve, and one specific next step." Paste the real work — feedback prompts fail without it.
What Makes a Feedback Prompt Different From a Grading Prompt
A grading prompt asks for a score. A feedback prompt asks for something that has to reference the actual piece of work in front of the student — generic feedback that could apply to any submission isn't feedback at all, it's a template wearing feedback's name.
Feedback vs. a Grade: Two Different Jobs
A grade summarizes; feedback guides. A rubric score of "3 out of 4" tells a student where they landed, but it doesn't tell them what to do differently on the next attempt — that's the job feedback is actually built for. Both matter, but conflating the two in a single prompt tends to produce a response that does neither job especially well.
- A grade answers: "How did I do, overall?"
- Feedback answers: "What should I do differently next time, specifically?"
Why Feedback Must Reference the Actual Work
A feedback prompt without the real student text attached tends to return generic encouragement dressed up as specific advice — "great effort, keep refining your ideas" could apply to almost any piece of writing, which is exactly the problem. Pasting the actual work into the prompt is the single biggest factor separating useful AI-generated feedback from filler.
This holds regardless of subject or assignment type. A math solution, a lab write-up, and a persuasive paragraph all need the same underlying discipline — the actual work in the prompt, not a description of what the assignment was asking for.
The Four Levels of Feedback
Hattie and Timperley's (2007) research identified four distinct levels feedback can target, and naming the level explicitly in a prompt produces noticeably more useful results than leaving it unstated. Most single-sentence feedback prompts default to the task level alone, which is useful but incomplete — it tells a student what was right or wrong without touching how they approached the work or how they might monitor it themselves next time.
Table: Four Feedback Levels
| Level | What It Addresses | Example |
|---|---|---|
| Task | Correctness of the specific work | "Your third paragraph is missing a topic sentence." |
| Process | The strategy or method used | "Try outlining your argument before drafting next time." |
| Self-regulation | The student's own monitoring of their work | "Before submitting, check whether each paragraph answers the prompt." |
| Self | The student as a person, not the work | "Great job!" |
Why "Great Job" Is the Least Useful Feedback Type
Self-level feedback — praise directed at the student rather than the work — is consistently the least effective category in Hattie and Timperley's (2007) analysis, precisely because it carries no information a student can act on. "Great job!" and "Nice try" both feel encouraging, but neither tells a student what to repeat or what to change.
The most useful feedback usually blends task-level specificity with process-level or self-regulation-level guidance — what was right or wrong, plus how to think about it differently next time.
Requesting a Specific Level in the Prompt
- "Give task-level feedback on whether this math solution's steps are correct, and process-level feedback on the strategy used."
- "Give self-regulation feedback: what should this student check for themselves before their next submission?"
The Core Elements of a Strong Feedback Prompt
Every reliable feedback prompt is built from the same four decisions, the same underlying discipline covered more generally in AI Prompting & Content Workflows for Teachers (2026 Guide).
| Prompt Element | What It Controls | Vague Version | Specific Version |
|---|---|---|---|
| Student work | What the feedback actually responds to | (not attached) | "[Paste the actual paragraph or solution]" |
| Criteria/skill | What the feedback focuses on | "Is this good?" | "Focus on thesis clarity and evidence use" |
| Strength/growth balance | How the feedback is structured | (unstated) | "One strength, one growth area, one next step" |
| Tone | How the feedback reads to the student | (unstated) | "Specific and encouraging, not just positive" |
Prompts for Feedback on Written Work
Written feedback is where AI-assisted feedback prompts see the heaviest use, and it's also where the "paste the actual work" rule matters most.
A Base Template for Written Feedback
- "Give feedback on this student paragraph, focused on [specific skill, e.g., topic sentences]. Note one specific strength, one specific area to improve, and one concrete next step the student could try. Grade [X] reading level for the feedback language itself. [Paste paragraph]"
Balancing Strengths and Growth Areas
Feedback that's all correction reads as discouraging regardless of how accurate it is; feedback that's all praise gives a student nothing to actually change. Requesting both explicitly, in a fixed ratio, keeps a generated response from drifting too far toward either extreme.
- Strength-only feedback risk: feels nice, teaches nothing.
- Correction-only feedback risk: technically accurate, discourages revision.
- Balanced request: "One specific strength, one specific growth area, one next step" — a simple three-part structure that works across nearly any assignment type.
Feedback on Math and Problem-Solving Work
Math and problem-solving feedback needs a different emphasis than essay feedback, since the strength and growth area often live in the steps shown, not just the final answer.
- "Give feedback on this student's solution, focused on whether each step logically follows from the last. Note where the reasoning is sound, where it breaks down, and one specific strategy to try on a similar problem."
Feedback that only addresses whether the final answer is correct misses the more useful information: which specific step introduced the error, and what strategy might prevent the same mistake next time.
Feedback that works well for a persuasive essay doesn't automatically transfer to every subject. See how the same three-question structure adapts for a world-language classroom in How to Write AI Prompts for Spanish, or for applied money-math work in How to Write AI Prompts for Financial Literacy.
Applying This to a Grade 5 Persuasive Essay Draft
Say a Grade 5 class has just turned in first drafts of a persuasive essay, and the assignment's focus is using evidence to support a claim. Pasting one student's actual paragraph into a feedback prompt, rather than describing the assignment generically, changes what comes back.
A prompt built from the core elements above — the actual paragraph, the specific criteria (evidence use), a balanced strength-and-growth structure, and grade-appropriate tone — might return something like: "You clearly state your opinion in the first sentence, which is a strong start. Right now, your reasons don't include a specific example — try adding one detail from the article you read that supports your first reason."
- Strength named: clear opinion statement in the first sentence.
- Growth area named: reasons lack a specific supporting example.
- Next step given: add one detail from a source the student already read.
That response touches all three of Hattie and Timperley's (2007) questions without stating them explicitly: it confirms what the goal was, names what's working and what's missing, and gives one concrete next step. Reviewing it before sending — for tone and reading level especially — remains a required step regardless of how well-built the prompt was.
Once a pattern of gaps shows up across feedback on a full class set, How to Generate 50 Quiz Questions in 5 Minutes With AI covers turning that pattern into a quick, targeted comprehension check.
Prompts Built Around the Three Feedback Questions
Hattie and Timperley's (2007) three questions give a feedback prompt a ready-made structure, whether the task is a full essay or a single math problem.
Where Am I Going?
Feedback works better when a student can see it against a clear target. "Restate the goal of this assignment in one sentence before giving feedback" keeps a generated response anchored to what the work was actually supposed to do.
How Am I Doing?
This is the level most feedback already covers — what's working and what isn't, specific to the actual submission. The risk here is staying too vague: "good effort" answers this question in name only, without any real content.
Where to Next?
This is the question generic feedback skips most often. Explicitly requesting "one specific, doable next step" forces the prompt to produce forward-looking guidance instead of stopping at a diagnosis.
Table: The Three Questions as Prompt Instructions
| Question | Prompt Instruction |
|---|---|
| Where am I going? | "Restate the goal or standard being assessed in one sentence." |
| How am I doing? | "Note one specific strength and one specific gap, tied to the actual work." |
| Where to next? | "Give one concrete, doable next step — not a vague general encouragement." |
Prompts for Different Feedback Moments
Not all feedback happens the same way — a quick verbal comment during independent work looks nothing like a written comment on a submitted essay, and a prompt should match the moment.
Table: Feedback Format by Moment
| Moment | Format | Prompt Addition |
|---|---|---|
| In-class, real-time | Short verbal script | "Keep it under 15 words, phrased as something a teacher could say out loud" |
| Async written comment | Full three-part comment | "Written for a student to read independently, no verbal delivery" |
| Peer feedback | Sentence starters | "Write 3 sentence starters a classmate could use, not the feedback itself" |
| Report-card style | Summary across a period | "Summarize a pattern across the term, not a single assignment" |
Sentence Starters for Peer Feedback
Peer feedback often needs scaffolding more than the feedback itself, since students are still learning to give feedback, not just receive it: "I noticed you ___," "One question I have is ___," "Next time, you could try ___" are reusable starters an AI prompt can generate once and apply across an entire unit. The same sentence-starter scaffold works well for visual-art peer critique, too — see How to Write AI Prompts for Art for how it adapts to a gallery-walk or critique setting.
Adjusting Feedback Tone and Complexity by Grade Band
Table: Feedback Defaults by Grade Band
| Grade Band | Tone | Structure |
|---|---|---|
| K-2 | Very concrete, simple vocabulary | One strength, one simple next step |
| 3-5 | Concrete with light abstraction | Strength, growth area, next step |
| 6-9 | More analytical, standards-referenced | Full three-question structure |
Younger students generally respond better to feedback anchored in something they can immediately picture doing again — "add one more detail like you did here" — rather than an abstract instruction like "elaborate more."
Tools for Turning These Prompts Into Classroom-Ready Feedback
A general AI chatbot can run every template in this guide, and pasting a real piece of student work each time is the fastest way to judge whether a given feedback style actually lands.
EduGenius can generate structured, three-part feedback directly tied to a class profile's grade level and a session's stored context, which is designed to keep tone and specificity consistent across a full stack of submissions rather than drifting piece to piece. Session history with feedback tracking also means a pattern across a student's submissions over time — not just one assignment — can inform what "next step" actually gets suggested.
- A general chatbot works well for occasional feedback or testing a tone before applying it broadly.
- A classroom content platform helps once structured feedback becomes a routine part of grading a full stack of work.
- Pasting the actual student work stays required either way — no tool substitutes for that.
New EduGenius accounts start with 25 welcome credits, and the Starter plan runs $7.99 a month for 500 credits, which is designed to cover a full class set of structured feedback rather than a handful of one-off comments.
Once feedback patterns are established, The Best AI Prompts for Creating Rubrics covers building the criteria feedback should reference, and An AI Workflow for Building Vocabulary Lists covers a related but distinct task — building the terms students need before they can act on feedback that uses them.
Pro Tips for Better Feedback Prompts
- Always paste the actual student work. Generic feedback prompts produce generic feedback, no matter how well-designed the rest of the prompt is.
- Request one next step, not three. A student who receives five suggestions at once often acts on none of them; one specific, doable step is more likely to actually get tried.
- Ask for feedback phrased as a question sometimes, not just a statement — "What might happen if you moved this sentence to the introduction?" can prompt more independent thinking than a direct instruction.
- Keep a bank of reusable sentence starters for peer feedback, since students often need the scaffolding more than the content of the feedback itself.
- Review AI-generated feedback for tone before sending it, especially for younger students — language that reads as neutral to an adult can land as harsh to a nine-year-old.
- Track which "next steps" a student actually tries over a few assignments, and let that inform what future feedback suggests.
- Save a base template per assignment type, adjusting only the criteria and pasted work each time rather than rebuilding the prompt structure from scratch.
- Read generated feedback out loud once before sending it. Wording that scans fine silently can sound blunter or colder than intended when actually spoken.
What to Avoid When Generating Feedback With AI
- Generating feedback without the actual student work attached. Without it, a prompt returns generic encouragement that could apply to almost any submission.
- Sticking to self-level praise like "great job." It's the least actionable feedback category — pair it with task- or process-level guidance instead.
- Piling on too many suggestions at once. One specific next step is more likely to get used than five competing ones.
- Skipping a tone check before sending feedback to younger students. Language that reads as neutral or direct to an adult can land very differently for a young child.
Key Takeaways
- Feedback and a grade do different jobs — a grade summarizes performance, while feedback should point toward a specific next action.
- Pasting the actual student work into the prompt is the single biggest factor separating useful AI-generated feedback from generic filler.
- Hattie and Timperley's (2007) three questions — Where am I going? How am I doing? Where to next? — give a feedback prompt a ready-made, research-backed structure.
- Self-level praise ("great job") is the least actionable feedback type — pair it with task- or process-level specifics instead.
- Request a fixed structure, such as one strength, one growth area, and one next step, to keep generated feedback balanced.
- Match format to moment — a short verbal script for real-time feedback, a full written comment for async review, sentence starters for peer feedback.
Frequently Asked Questions
What's the most important thing to include in a feedback-generation prompt?
The actual student work. Without it, even a well-structured prompt tends to return generic encouragement that could apply to almost any submission. Pasting the real text or solution is what lets the feedback reference something specific.
What are the four levels of feedback from Hattie and Timperley's research?
Task-level (correctness of the specific work), process-level (the strategy or method used), self-regulation-level (how a student monitors their own work), and self-level (praise directed at the student rather than the work). Self-level praise, like "great job," is consistently the least actionable of the four.
How do I get AI feedback that doesn't feel generic?
Paste the actual student work, name the specific skill or criteria in focus, and request a fixed structure — one strength, one growth area, one concrete next step — rather than an open-ended "give feedback" request. Generic prompts produce generic feedback almost every time.
Should AI-generated feedback be different for younger students?
Yes. Younger students generally respond better to concrete, simple language and a single clear next step, while older students can handle more analytical, standards-referenced feedback and the full three-question structure. Reviewing tone before sending feedback to younger students matters too, since language that reads as neutral to an adult can land more harshly for a child.
Can AI give useful feedback on math work, not just writing?
Yes, but the prompt needs a different focus than essay feedback: ask specifically whether each step logically follows from the last, not just whether the final answer is correct. Feedback that only checks the final answer misses the more useful information — exactly which step introduced the error and what strategy might prevent it next time.
Related Reading
References
- Hattie, J., and Timperley, H. (2007). The Power of Feedback. Review of Educational Research.
- Wiliam, D. — research on formative assessment and feedback loops.
- Dweck, C. — research on growth mindset and feedback framing.