On this page
Four steps for educators: read the work, match the task, check each comment against a passage, then edit and decide whether to share it.
Keep the student’s work, the assignment and your own judgment in the review.

Review one feedback comment before sharing it

Use this private working record to review one draft comment against the assignment, the learner’s actual work and your own assessment. It records your checks; it does not grade, approve or validate feedback for you.

Do not enter a student name, ID or full submission. Follow your institution’s rules before putting any student work into an AI service. This page does not send or save these notes; print the record if you need a copy.

Name the relevant task and rubric criterion, if one applies.

Use a paragraph, line or section reference; avoid copying the whole submission.

Remove identifying details before recording a comment here.

Record the exact feature that supports or contradicts the comment.

Check each point against the work and assignment

State what you will keep, change or remove, and the action the learner can take.

Record your human decision

Verify every claim against a locatable part of the writing

For each observation, locate the paragraph, sentence, example or missing step it refers to. Ask whether the comment accurately describes the student’s work and the assignment instructions. A comment about “more evidence” is not useful when it cannot name which claim needs support or what kind of support the prompt asks for. Correct a false observation before deciding whether any suggested revision is relevant.

Compare the comment with the actual criterion, not an AI-written interpretation of a rubric. Check that it addresses the intended skill, uses course terminology correctly and does not introduce a requirement that the assignment never set. A peer-reviewed rubric study of AI feedback on student writing identifies accuracy, constructiveness, clarity, recognition of strengths and preservation of the student’s original text as distinct review dimensions. Its sample came from specific biology and chemistry courses; use those dimensions as questions, not as a universal score or proof that a comment is valid.

Supporting reading: [1] [2] [3]

Make the next step specific without writing the student’s answer

Replace a vague judgment such as “develop this more” with a question or action tied to the learner’s own text: identify the claim that needs a reason, compare the two examples, or explain how the cited passage supports the conclusion. Preserve a genuine strength so the learner knows what to keep. Make the instruction short enough to follow and appropriate to the learner’s current task.

Watch for feedback that supplies a finished sentence, changes the student’s meaning or solves the very reasoning task they need to practise. The published writing-feedback rubric treats preservation of original text separately from accuracy and clarity. A comment can be grammatically polished yet still remove the learner’s opportunity to make the revision. The educator decides which prompt or explanation will support learning in this context.

Supporting reading: [2] [3]

Check tone, context, consistency and student choice

Read the comment as the learner will receive it. Remove unsupported assumptions about effort, ability or intent; check the tone and language level; and look for a concrete next action. If reviewing a set of comments, compare similar cases for consistency without assuming that identical wording is fair when students have different instructions, accommodations or learning needs. Ask a colleague to review an uncertain or consequential comment through your normal process.

Explain when AI may assist with feedback and how a learner can ask a question or receive an equivalent route when one is needed. The University of Minnesota recommends considering student agency, transparency, privacy and alternative opportunities. These are planning questions, not a claim that every institution requires the same opt-out policy. Follow the rules that apply to your course and jurisdiction.

Supporting reading: [1] [2]

Set the privacy boundary and make the release decision yourself

Before any student work enters an external service, check the institution’s approved tools, data terms, consent or notice requirements, retention settings and permitted account route. Remove names and other identifiers where policy allows de-identified work, but do not assume that removing a name alone makes a submission safe to share. If the approved route is unclear, do not upload the work; ask the relevant school or privacy contact. James Cook University’s published staff guide sets rules for its own environment, including a specified institutional account. Those account instructions are JCU-specific and should not be generalized to another institution.

Review the final comment beside the original writing, rubric and assignment, then decide whether to edit, hold or share it. Record the specific edit and reason when that helps explain the decision later. The checklist below stores nothing on our server and does not decide whether the comment is ready. A human educator remains responsible for following local policy and approving any feedback they send.

Supporting reading: [1] [2]

Worked example: keep the useful observation, remove an unsupported prescription

Constructed teaching example: a fictional assignment asks a learner to use a small voluntary campus poll and discuss who its results can represent. The fictional paragraph says, “Eighteen volunteers from the evening class preferred the later bus. This suggests the evening respondents may value a later service.” A draft comment says, “Your evidence is not representative; collect at least 500 randomly selected responses and rewrite the conclusion.” No real student, course, survey, AI session or product output is represented here.

The first part of the comment raises a relevant question: the poll includes volunteers from one evening class, so it cannot automatically describe every campus student. The required sample size of 500 is unsupported by the assignment and by the constructed information. A more bounded teacher-edited comment might say: “Your sentence identifies the 18 evening-class volunteers. What limits should readers keep in mind before extending their preference to the whole campus? Add one sentence explaining who was and was not included.” The educator would still check this wording against the actual task and rubric before sharing it.

Which part of the draft comment can be retained, and what must be checked or changed?

Retain the question about how far the result can be generalized only if that matches the assignment criterion. The passage itself identifies a small voluntary group from one class, so the limitation is locatable. Remove the invented requirement for 500 random responses: no sample-size calculation, course rule or evidence supports that number in this exercise. Then make the next step specific and let the learner write the explanation.

This is a constructed illustration of editorial judgment, not a claim about an AI tool’s output or an observed classroom outcome.

Supporting reading: [2] [3]

Questions about this topic

Does this checklist grade student writing?

No. It helps an educator review one feedback comment. It does not assign a score, evaluate a student or approve a course grade.

Can a completed checklist prove that feedback is accurate?

No. You must inspect the exact student passage, assignment and applicable rubric. The checklist records those checks but cannot perform them.

Should I upload a student’s full essay to get feedback?

Only if the institution’s approved process and applicable rules allow that exact use. Check the account, data terms and disclosure route first; when uncertain, do not upload the work.

Are my entries saved by this page?

No. The fields are ordinary browser inputs with no automatic saving or server submission. Print the page if you need a copy, and avoid entering identifying details.

Sources and editorial notes

The supporting reading below informs our editorial guidance. Exercises and scenarios illustrate a method; they do not report a completed product test or measured learning outcome. Read our editorial policy or send a correction with a supporting source to info@ai-essay.com.

  1. University of Minnesota Teaching Support: Assessing Student Work with GenAI Tools
  2. James Cook University: AI Guides, AI for Feedback
  3. Bauer et al., Journal of Microbiology & Biology Education: A rubric to assess generative AI-based feedback on student writing assignments