How to Disclose AI-Assisted Grading to Learners: A Clear Notice and Teacher-Review Checklist

When AI contributes to a score, grade, or feedback, learners should not have to guess what happened between submission and return. A clear disclosure is not a long technology policy. It is a practical explanation of what the system did, what the teacher did, what information was used, and what learners can do if they believe something is wrong.

This matters because assessment is more than producing a mark. Learners use grades and feedback to decide what to improve next, whether to ask for help, and whether an outcome reflects their work. Clear documentation can support transparency, accountability, and human review processes—principles reflected in the NIST AI Risk Management Framework. UNESCO’s guidance for education also emphasizes a human-centred approach, including attention to human agency and data protection when generative AI is used in education.

For tutors, school owners, and L&D teams, the goal is simple: give learners useful information at the moment they need it. Put a short notice before submission, a specific label beside returned work, and a clear route for questions or review. Keep the teacher responsible for the educational decision.

Start by naming the AI’s actual role

“AI-assisted grading” can describe very different workflows. A disclosure should match the real workflow, not a broad product label. Before drafting any learner notice, identify which of these roles applies to the assessment.

Do not collapse these roles into one statement. A system that drafts comments is not doing the same job as one that recommends a rubric score. Likewise, a teacher who checks every returned grade is providing a different safeguard from a teacher who only reviews a sample.

Use precise verbs. Say “drafts,” “checks,” “suggests,” “flags,” “reviews,” or “confirms.” Avoid vague wording such as “AI supports assessment” when learners need to understand whether the tool can influence their result.

The five details learners should see before submitting work

A pre-submission notice should be short enough to read, but complete enough to support an informed choice where a choice exists. Place it next to the submission button, not only in a handbook or terms page.

1. What is being used. State that an AI-enabled tool may assist with checking, score recommendations, feedback drafting, or another named task.
2. What work it applies to. Identify the assessment, question types, rubric criteria, or feedback elements affected.
3. What information is processed. Explain in plain language what the tool receives, such as submitted answers, rubric criteria, or course-level performance information. Do not make privacy assurances unless they have been verified for the specific tool and contract.
4. Where human judgment applies. Name the person or role that reviews, confirms, edits, or can override the outcome. If there is no individual review, say so plainly.
5. How to ask a question or request review. Give a real contact route, the information learners should provide, and the expected next step.

For younger learners or multilingual groups, provide a shorter version and link it to a fuller explanation. Accessibility is not a finishing touch: make the notice readable with screen readers, avoid unexplained technical terms, and ensure the question route is available to learners who cannot easily use a particular platform.

Copy-and-customize learner disclosure pack

1. Pre-submission notice

AI-assisted assessment notice

For this assessment, we use [tool or system type] to [check objective answers / draft feedback / suggest rubric scores / identify items for teacher review]. It uses [submitted work and relevant assessment criteria].

Your final [grade / score / feedback] is [reviewed and confirmed by a teacher / produced automatically for the listed question types]. A teacher can review or change an outcome where appropriate. If you have a question about your result or believe it does not reflect your work, contact [role or contact route] and include [assessment name, question or criterion, and reason for review].

Adapt the bracketed text to the real process. If the system only checks multiple-choice questions, name that limitation. If a teacher reviews only scores outside a specified range, state that rather than implying universal review.

2. Returned-grade label

Assessment processing label: AI assisted with [specific task]. [Teacher role] [reviewed and confirmed / edited / did not individually review] this [score / feedback].

Place this label where learners see the grade or feedback. It should not be hidden behind an information icon. A learner should be able to tell whether the teacher reviewed the individual result without opening another page.

3. Teacher-review statement

Teacher review: I used the assessment criteria and the learner’s submitted work to review this outcome. Where the AI provided a suggestion, it did not replace my professional judgment. I can amend the score or feedback when the evidence supports a different decision.

Use this only when it accurately describes the process. Do not use a teacher-review statement for a workflow in which the teacher did not review the individual learner’s work or result.

4. Learner question or appeal message template

Subject: Request for review of AI-assisted assessment

I would like a review of my [assessment name] result. I am asking about [question, criterion, score, or feedback comment]. My reason is: [brief explanation]. The part of my submitted work I would like considered is: [location or excerpt]. Please let me know who will review this and what happens next.

A meaningful route does not require a formal appeal process for every low-stakes quiz. It does require that learners can reach a person, explain the issue, and receive a clear next step. For high-stakes assessments, align the route with your organisation’s established assessment and complaints procedures.

What “teacher review” should mean in practice

Teacher review is a workflow, not a slogan. Define it before publishing the disclosure. For example, it might mean a teacher checks every AI-recommended rubric score before release. It might mean a tutor reviews all flagged responses, borderline results, or learner requests. These are different safeguards and should be described differently.

• Review before release: A teacher checks the individual outcome before learners see it.
• Review after release: Learners receive an initial result, with a teacher available to investigate questions or challenges.
• Sampling: A teacher checks a portion of results for quality assurance, not every learner’s result.
• Override authority: A named teacher or assessor can change a score, feedback, or both, with a reason recorded where appropriate.

Keep an internal record of which version of the workflow applies to each assessment. NIST’s guidance describes documentation as a way to enhance transparency and support human review; a simple assessment register can make that principle operational. Record the assessment name, AI role, input data, review rule, escalation contact, and date last checked.

Pre-publication checklist for a real assessment workflow

Test the disclosure as a learner would experience it. Do not approve wording in isolation from the actual platform, gradebook, and support route.

• Can a learner see the notice before clicking submit?
• Does the notice name the AI task accurately and avoid inflated claims about what the tool can do?
• Does it distinguish automated checking, AI-generated feedback, score recommendations, and final decisions?
• Does the returned-grade screen show whether an individual teacher reviewed the result?
• Can the teacher override the outcome, and do staff know how to do so?
• Does the question route reach a named person or monitored inbox?
• Have you tested the route with a sample learner message and confirmed the response process?
• Have you checked that the information-processing description matches the provider agreement and your own privacy documentation?
• Have you checked the notice on mobile, with assistive technology where possible, and in the languages your learners need?
• Have you set a review date for the disclosure when the assessment, provider, model, rubric, or review workflow changes?

Make disclosure part of the assessment design

A disclosure cannot make an unsuitable grading workflow suitable. It can, however, make the workflow visible enough to test, improve, and challenge. When the notice, returned-grade label, and review route all match the real process, learners know what AI contributed and where a teacher remains accountable.

For small providers, start with one assessment type rather than rewriting every policy at once. Build the four templates into the course workflow, test them with staff and a small learner group, then revise the wording when the process changes. SubSchool can help education teams automate repetitive teaching tasks while teachers retain authorship and the final educational decision; use that principle as the standard for every learner-facing grading notice.

Practical next step: Choose one AI-assisted assessment this week, complete the five disclosure details, and ask a colleague to compare the published notice with the actual grading workflow. If the two do not match, fix the workflow or the wording before learners submit work.