What does NZQA authenticity mean for AI in NCEA assessment?

What does NZQA actually mean by authenticity?

Authenticity is the plainest rule in the NCEA system. The work a student submits has to be their own. NZQA puts the obligation on the assessor: you must verify that work submitted for assessment has been produced by the learner. That was the rule long before generative AI arrived. AI has not changed the rule. It has changed how much work it takes to satisfy it.

The Ministry of Education states the consequence bluntly. If generative AI is used to produce evidence that is assessed as the student’s work, then the assessment is invalid. Not marked down. Invalid. The question is never whether the writing is good. It is whether the evidence in front of you was processed and produced by the student.

This lands hardest on internal assessment. External assessment has an exam room and a supervisor. Internal assessment has you and whatever record you kept while the task was running.

Is AI allowed in NCEA internal assessment?

Take external assessment first, because it is the simple half. NZQA is explicit that the use of chatbots, generative AI, paraphrasing tools, or other tools that can automatically generate content is not permitted for external assessment. There is no school-level discretion there and nothing to interpret.

Internal assessment is judged standard by standard, not by a blanket rule. The Ministry’s own FAQs on GenAI in NCEA assessment put it plainly: limited use may be appropriate in the context of some achievement standards, depending on what is being assessed. If a standard assesses the student’s ideation or planning, then using a tool to ideate or plan is out, because that is the thing being assessed.

Two conditions have to hold before a student can rely on any permission. Kura and schools with consent to assess listed standards must have an authenticity policy, and that policy must cover the acceptable use of AI. Then any use has to comply with that standard’s Conditions of Assessment and must not scaffold or guide the student to the point where the work is no longer their own. Neither is a national default.

NZQA also expects the instruction to reach task level, not just policy level. Schools are asked to give explicit authentication instructions for each assessment task, including how generative AI can and cannot be used for each standard assessed. In practice that means a line on the task sheet, standard by standard, that a Year 12 student can read and act on. A cross-reference to the school policy is not that line. If those lines do not exist yet, that is the first job, ahead of any decision about tools.

How do you evidence authenticity for a task where AI exists?

NZQA’s advice on managing authenticity is process advice, not detection advice. It points to monitoring progress through submission of plans and drafts, logs, version histories, feedback, conferencing, planned next steps and regular checkpoints. Its assessor guidance on authenticity adds oral questioning where there is doubt, requiring plans and source material with the final piece, modifying publicly available tasks, and supervising research through checkpoints.

Almost all of that is evidence you generate while the task is running about how the work came to exist. NZQA does list plagiarism software and AI detection tools among its strategies, but they come last and carry the heaviest caveat on the page. A version history plus a fifteen minute conference at the halfway point will hold up at moderation. A detector score on its own will not.

On detectors NZQA is direct: be very mindful of false positives, and do not rely on this alone. Putting a false positive to a student who wrote every word is a serious thing to do, and neither of you can prove it either way afterwards. Treat any score as a reason to talk, never as a finding. Students leaning on tutors, study guides or a chatbot usually show up in the process record well before any tool flags them.

Does authenticity cover the teacher's marking, or only the student's work?

Only the student’s work. Authenticity is a rule about the evidence, and the evidence is what the student produced. The NZQA guidance on acceptable AI use is written about students generating content for assessment. It does not set out which tools a teacher may use when reading and marking that evidence.

The teacher-side constraints come from moderation instead, and they are firm. Internal moderation requires verification of grade judgements before results are reported, and NZQA specifies that verification must be undertaken by a subject specialist with standard-specific knowledge, working from samples chosen around grade boundaries rather than picked at random. Schools also have to retain annotated samples of student work and evidence of internal moderation. A grade that no named subject specialist can stand behind does not meet that bar, whatever produced it.

The Ministry of Education’s position on teachers and AI fits the same shape: the use of AI tools for marking should support, not replace, the professional judgement of teachers. So the useful question about any tool is whether a human assessor still makes the judgement, owns it, and can defend it at verification. If a tool drafts a judgement and you check it against the standard, you are the assessor. If a tool issues a grade and you forward it unread, you are not.

What do the resubmission rules mean for AI-drafted feedback?

This is where teachers get caught, and it has little to do with AI. NZQA’s rules on resubmission are tight. Only one resubmission can be offered for each assessment of a standard. The student has access to an Achieved grade only. Feedback must be non-specific, must not overdirect the student, and no further teaching or learning can have taken place before the resubmission.

Now put a marking assistant into that picture. A tool that produces detailed, specific, improvement-focused comments is doing exactly the right thing for formative work and exactly the wrong thing at a resubmission point. The tool has no idea which stage of the standard you are at. You do. Anything it generates has to be read against those rules before it reaches the student.

Most departments settle on a split. Use generated feedback freely on formative tasks, where specificity is the whole point, and to speed up your own reading of summative work. Write resubmission feedback yourself, so you can be certain it stays non-specific enough to survive a moderation query.

What should you check before letting a tool near NCEA work?

Start with privacy, because it is the check most likely to stop you. Student writing is personal information under the Privacy Act 2020. Ministry of Education guidance to schools is blunt about it: do not use personal data when you use AI tools, because what you put in may be reused. The Office of the Privacy Commissioner’s position is that the Privacy Act applies to everyone using AI tools in New Zealand, that a privacy impact assessment should come first, and that if in doubt you should not use AI tools to handle personal information at all. It also expects the use to fit the purpose the information was collected for; training a model on student writing does not.

After privacy, the questions are about assessment integrity. All of them are answerable in an afternoon.

Purpose-built marking tools are built around these constraints in a way an open chatbot is not. Jeddle is one example: JeddAI drafts against the rubric and success criteria you have already uploaded rather than a generic scale, and every comment sits in front of the teacher for review and editing before a student sees it. That is the shape to look for: the tool does the reading and the drafting, and a named assessor still owns the call. Teachers weighing up expert essay marking should judge any product on whether it keeps that division of labour.

None of this removes the work of authentication. It moves your time. Hours saved reading a class set are hours you can put into checkpoints, conferencing and the process record, which is what NZQA asks you to produce.

• Does the task sheet carry explicit, standard-specific instructions on how AI may and may not be used?
• Can you produce a process record for each student: plans, drafts, version history, conference notes?
• Does a named subject specialist verify grades around the boundaries before results are reported?
• Is student writing retained by the vendor, and can it be used to train a model?
• Does the tool mark against your rubric and the achievement standard wording, or a generic scale?
• Can you edit or discard every comment before a student sees it?