You keep the teacher in control by treating AI as a drafting assistant, never the final marker: it proposes feedback against your rubric or achievement standards, and you review, edit and approve every comment before a student sees it. Human judgement, moderation and professional accountability stay with the teacher throughout.
What does keeping the teacher in control actually mean?
Keeping the teacher in control means AI drafts feedback while you make every assessment decision. The tool reads student work and proposes comments, but nothing reaches a learner until you have read, edited and approved it. AI becomes a fast first-drafter of routine feedback, not an autonomous marker, so your professional judgement, and your accountability for the mark, stays exactly where it belongs.
This is often called a human-in-the-loop model. The loop is the review step you never remove: the AI suggests, you check it against what you actually meant, and you decide what stays. In practice that looks like scanning a drafted comment, tightening the wording, correcting anything the model misread, and only then releasing it. The student experiences your voice and your standards, simply delivered faster.
Why does human-in-the-loop matter for marking and moderation?
Human-in-the-loop matters because assessment carries professional and moderation obligations that cannot be delegated to software. In New Zealand, NCEA judgements against achievement standards are subject to internal and external moderation through NZQA, and a teacher must be able to justify every decision. An AI suggestion is evidence you consider, never a verdict you outsource; the person signing off the grade remains responsible for its defensibility.
There is also a quality risk to manage. AI can misread a student’s intent, miss context from earlier in a programme, or write fluent feedback that is subtly wrong. Left unchecked, that erodes trust with learners and whānau over time. Keeping a person in the loop catches these errors before they land, and it protects the relationship at the centre of good feedback: a teacher who clearly knows this particular student’s work.
Keeping a person in the loop also creates a clear record of professional reasoning. When you edit and approve each comment, the final feedback reflects a decision you can explain to a moderator, a colleague or a parent. That audit trail matters most for borderline grades and for standards where consistency across a cohort is scrutinised.
How do you set up AI feedback to reflect your judgement?
You set up AI feedback to reflect your judgement by anchoring it to your own criteria before it writes a single comment. Give the tool your rubric, success criteria, exemplars and any comment bank you already use, so its suggestions echo your expectations rather than a generic idea of good writing. The more clearly you define what you are looking for, the less editing you do afterwards.
It also helps to constrain the scope of each request. Ask the AI to focus on the two or three things you are actually assessing in this task, not everything at once. Specify the tone you want, more encouraging for an early draft and more exacting near a final submission, and the year level, so the language lands appropriately anywhere across Years 1 to 13.
- Upload or link your rubric and success criteria so feedback maps to the standard being assessed.
- Provide a comment bank or a few marked exemplars to model your voice and expectations.
- Name the two or three focus points for this task so feedback stays targeted.
- Set the tone and year level so wording suits the student and the stage of the work.
What should you check before approving AI-drafted comments?
Before approving any AI-drafted comment, check it for accuracy, fairness and tone against the actual piece of work. Confirm the feedback describes what the student really did, that any strengths and next steps are correct, and that the language is respectful and specific. If a comment is generic, wrong, or reads like it could apply to any essay, edit it or delete it, because a short accurate comment beats a long plausible one.
Pay particular attention to grades and next steps. Verify that any suggested standard judgement is one you can defend under moderation, and that improvement advice is genuinely actionable for this learner. Watch for invented detail, such as references to a text or event that are not in the submission, and for bias that could disadvantage a student writing in a second language or a different dialect.
- Accuracy: does the comment match what the student actually wrote?
- Standard alignment: can you defend any judgement against the achievement standard and moderation?
- Actionability: is the next step specific and achievable for this learner?
- Tone and fairness: is the language respectful and free of bias against EAL or dialect differences?
- Invented detail: have you removed any reference the AI added that is not in the work?
How do you save time without giving up control?
You save time without giving up control by letting AI absorb the repetitive drafting while you concentrate on the judgement only a teacher can make. Routine, high-volume feedback, such as surface errors, structural prompts and standard encouragement, is where AI clears the most time. You then spend your attention on the borderline calls, the individualised guidance and the moderation-critical decisions that genuinely need a professional.
A practical rhythm is draft, skim, refine, release: generate a batch of feedback, read quickly for anything off, adjust the few comments that need it, and approve the rest. Across a class set this can turn hours into a focused review session without lowering the bar. The goal is not less teacher input overall, but teacher input aimed squarely where it changes student outcomes.
It also helps to set your own review threshold before you begin. Decide, for example, that you will read every comment on a summative task but spot-check formative drafts, and that any AI suggestion touching a grade always gets full scrutiny. A clear rule keeps the time saving honest and stops the review step quietly shrinking under pressure.
How can a tool like Jeddle help you stay in control?
A tool like Jeddle is built around the review step rather than around removing it. JeddAI drafts feedback and marking aligned to your own rubric, success criteria and comment banks, then hands every comment back to you to edit, approve or discard. It applies your criteria consistently across a class while leaving each final decision, and the professional accountability that comes with it, firmly with you.
That combination is what keeps AI genuinely useful for feedback: faster drafting, consistent application of your standards, and a teacher who stays in charge of what students receive. If you want to see that workflow in practice, you can Get started with JeddAI and try it against your own marking, keeping the review step and your judgement at the centre.
| Stage | What AI can do | What the teacher owns |
|---|---|---|
| Reading the work | Scan the submission and surface patterns quickly | Deciding what actually matters for this task and learner |
| Drafting comments | Propose feedback aligned to your rubric and comment bank | Checking accuracy, tone and fairness before anything is sent |
| Judging against a standard | Suggest a possible grade or level | Making the defensible judgement and meeting moderation |
| Releasing to students | Prepare comments in your voice at speed | Approving, editing or discarding every comment |
Frequently asked questions
Is it acceptable to use AI feedback for NCEA assessment?
Yes, as a drafting aid, provided you make and can defend the final judgement. NCEA decisions must stand up to internal and external moderation, so the teacher remains responsible for every grade against an achievement standard.
Should I tell students I use AI to help with feedback?
Transparency builds trust. Let students and colleagues know AI helps you draft feedback and that you review and approve every comment, so learners understand their work is still read and judged by a teacher.
What is the biggest risk of relying too heavily on AI feedback?
The main risk is fluent but inaccurate feedback slipping through unchecked, including invented detail or bias. A consistent review step, where you verify each comment against the actual work, is what prevents this.
Does keeping a teacher in the loop cancel out the time saved?
No. Reviewing and editing drafted feedback is far faster than writing every comment from scratch, so you keep most of the time saving while retaining full control of quality and accountability.
Get started with Jeddle
Jeddle gives teachers and students instant, syllabus-aligned feedback powered by JeddAI.
Looking for study material? Browse Jeddle's Australian-English subject resources, or explore more articles on AI in Education.
