You maintain assessment authenticity by designing tasks that are hard to outsource and by gathering evidence of the process, not just the product. Combine in-class writing, drafting checkpoints, oral verification and clear AI-use expectations so you can be genuinely confident the work reflects each student's own thinking and voice.
What does assessment authenticity actually mean under NCEA?
Authenticity means the assessor can be confident the work presented is genuinely the student’s own. Under NCEA, NZQA expects teachers to manage authenticity for every internal assessment, so results reflect what each student can actually do. Generative AI does not change that principle — it raises the stakes of getting it right.
The shift is subtle but important. Authenticity used to be mostly about detecting copied text or shared answers. Now a fluent, original-looking response can be produced in seconds, so a clean plagiarism report no longer proves much. The question moves from is this text copied? to is this thinking the student’s own?
That reframing is helpful because it points to solutions. If you can see reasoning develop — early notes, revised drafts, a student explaining a choice — you have far stronger evidence of authenticity than any single finished artefact can give you.
Which tasks are most vulnerable to undisclosed AI use?
The most vulnerable tasks are unsupervised, take-home written responses to generic prompts. A standalone essay on a common text, a report on a well-covered topic, or a reflection with no personal anchor can all be generated with little trace. If a chatbot can produce a passable answer from the prompt alone, the task carries real authenticity risk.
Externally assessed standards sat under exam conditions are largely protected, because the writing happens in a supervised setting. The pressure sits on internal assessment, where students often work over several days at home. This is exactly where NZQA’s conditions of assessment and your own checks have to do the heavy lifting.
It is worth naming the flip side too. Tasks that already draw on a student’s own data — a personal investigation, a response to material generated in class, or an analysis of their own performance — are naturally more resistant. Auditing an assessment programme for these two extremes tells you quickly where redesign effort will pay off.
- Generic essay or report prompts a model can answer without the student's input
- Take-home tasks completed with no supervised or in-class component
- Reflections or analyses with no personal, local or class-specific anchor
- Assessments where only the final product is collected, with no process evidence
How can you redesign tasks so the evidence stays genuine?
Redesign tasks so a meaningful part of the evidence is produced where you can see it, and so the prompt demands something only that student can supply. A short supervised writing session, a task grounded in class discussion or local context, or a requirement to reference their own earlier draft all make undisclosed outsourcing far harder.
You rarely need to rebuild a whole assessment. Often it is enough to add a supervised checkpoint, ask for annotated planning, or connect the task to a shared class experience the model has never seen. These changes keep the standard’s intent intact while making authenticity easier to defend at moderation.
- Add a short in-class writing or planning component completed under supervision
- Anchor the prompt in class discussion, a field trip or a local issue
- Require students to build on and reference their own annotated drafts
- Ask for a brief personal rationale explaining the choices they made
What in-process checks confirm the work is a student's own?
In-process checks confirm authenticity by capturing thinking as it develops rather than judging only the finished piece. Drafting checkpoints, brief conferencing and a working knowledge of each student’s usual voice let you notice when a submission suddenly does not match the writer you have watched all term.
A two-minute verification conversation is one of the most effective tools available. Ask the student to explain a decision, define a term they used, or extend an argument on the spot. Students who did the thinking answer easily; those who did not tend to struggle, and you learn a great deal without any accusation.
Keep this proportionate and consistent. Apply the same checks across the class, document your process, and treat a mismatch as a prompt for a conversation, not proof of misconduct. That protects both integrity and each student’s right to fair treatment.
How should you talk with students about acceptable AI use?
Be explicit about what is and is not acceptable for each task, because ambiguity is where most problems start. State clearly whether AI may be used for brainstorming, for checking spelling, or not at all, and ask students to disclose any tools they used. Clear expectations stop honest students crossing a line they did not know existed.
This is also a teaching opportunity. Helping students understand why authenticity matters — that the assessment measures their learning, not a tool’s output — builds judgement they will need beyond school. Framing AI literacy as a skill, rather than only a threat, tends to earn more genuine cooperation than surveillance alone.
Put the expectations in writing and revisit them. A short, shared statement on each task — what is allowed, what must be disclosed, and how the work will be verified — removes guesswork for students and gives you a consistent reference if a question about authenticity later arises.
How can a tool like JeddAI help you keep assessment fair?
A tool like JeddAI helps you apply your criteria consistently across a whole class, so every student’s evidence is judged against the same standard. Teachers connect or upload student work, and JeddAI drafts feedback and marking aligned to their own rubric, success criteria and comment banks — then the teacher reviews and edits every judgement.
That consistency matters most when you are weighing authenticity, because it frees your attention for the human checks — conferencing, verifying voice, reviewing drafts — that AI cannot do for you. The teacher stays in control of every decision while spending less time on repetitive marking. Get started with JeddAI to see how it fits your assessment workflow.
Frequently asked questions
Can AI-detection software prove a student used AI?
No. AI detectors produce both false positives and false negatives and are not reliable enough to stand as sole proof. Treat any flag as a reason to look at process evidence and talk with the student, not as a verdict.
Does NZQA allow students to use AI in NCEA assessment?
It depends on the standard and its conditions of assessment. Some tasks permit certain tools; others require independent, supervised work. Always check the specific conditions and make your expectations explicit to students.
What should I do if I suspect undisclosed AI use?
Follow your school's authenticity and academic-integrity process. Gather process evidence, hold a verification conversation, and document what you find before drawing any conclusion.
Is it fair to ban AI entirely for an assessment?
Yes, if the standard requires independent work and you state it clearly in advance. Fairness comes from consistent, well-communicated conditions applied to the whole class.
How much process evidence should I keep?
Enough to be confident the work is genuine and to justify your judgement at moderation — typically dated drafts, planning notes and any in-class writing. Keep it proportionate and consistent across students.
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.
