A Year 12 student hands in a beautifully structured investigation. The argument is coherent, the vocabulary sophisticated and the conclusions persuasive. Yet when the teacher asks, “Which evidence changed your mind?”, the student struggles to explain. The document looks excellent. The learning behind it is uncertain.
This is the central assessment challenge for secondary schools in Aotearoa New Zealand. NCEA assessment in the age of AI should establish what a student can understand, justify, remember and apply independently—not merely whether they can submit a polished product. Schools can do that through clearly stated AI rules, evidence collected during learning, short explanatory conversations, carefully designed tasks and proportionate authenticity checks. All of these must operate within the requirements of the particular achievement standard and NZQA assessment conditions.
The question matters immediately in 2026, even as New Zealand prepares to replace NCEA with a new senior secondary qualification system over the coming years. Teachers cannot postpone the work of designing trustworthy assessment until the reform is finished.
What does NCEA actually require when students use generative AI?
The first distinction is between using AI to learn and submitting AI-generated content as evidence of one's own achievement. Those are not the same activity. A learner might use a tool to explore possible explanations during a lesson. That does not automatically mean the tool can draft or transform material submitted for assessment.
The New Zealand Ministry of Education's generative AI guidance makes three points essential for NCEA practice. Schools with consent to assess need an authenticity policy that addresses generative AI. Evidence submitted for assessment must be the student's own work. Generative AI is not permitted in NCEA external assessment. For internal assessment, teachers must check the particular standard's requirements, conditions and school policy rather than assume that every form of AI assistance is allowed.
NZQA's acceptable-use guidance recommends checkpoints, comparison with earlier work, source acknowledgement, follow-up questions and caution about AI detectors. These are authenticity safeguards, not a substitute for an assessment design that reveals learning from the beginning.
The key operational question for each task is: What specific performance may the assessor legitimately credit to the student? If AI performs the central reasoning, composition or evaluation that the achievement standard seeks to assess, a student cannot simply claim that performance as their own. If a tool is used within permitted boundaries during preparatory learning, the teacher still needs independent evidence of the assessed competence.
Internal and external assessment are not interchangeable
| Assessment situation | Appropriate starting point | What the teacher should verify |
|---|---|---|
| Classroom practice before assessment | AI may have a learning role where school policy, tool conditions and safeguarding allow | The student can explain and practise without copying the tool's answer |
| NCEA internal assessment | Check the specific standard, assessment conditions and school authenticity policy | All credited evidence is independently demonstrated by the student |
| NCEA external assessment | Generative AI is not permitted under current Ministry and NZQA guidance | Supervision, assistance and submitted material comply with official specifications |
| Ordinary formative feedback | Teachers can use appropriate questioning and feedback methods | Feedback supports understanding rather than writing the final assessed response |
This table is a planning aid, not a blanket permission list. The latest standard-specific specifications and school procedures govern individual assessments. Teachers should consult the Principal's Nominee or relevant assessment lead before implementing an unfamiliar AI arrangement.
Why is a polished answer no longer convincing evidence by itself?
Generative AI can produce plausible paragraphs, organize an investigation, suggest mathematical steps, summarise readings and critique an argument. Some outputs are useful. Others are mistaken or oversimplified. In either case, the quality of the output reveals little about who did the thinking.
Consider a student preparing a geography assessment about flood risk. The submitted report offers several sophisticated mitigation strategies. Without additional evidence, the teacher may not know whether the student understands the difference between hazard, exposure and vulnerability, whether the maps were interpreted correctly or whether the recommendations came from a chatbot.
Three distinct questions can restore clarity:
Understanding: Can the student define the key concepts and explain the logic connecting them?
Ownership: Can the student show how their argument developed, identify their evidence and justify their choices?
Transfer: Can the student use the same reasoning on a fresh but related flood-risk scenario without prohibited assistance?
A useful assessment system collects the evidence that matters, rather than multiplying bureaucratic paperwork. Students should not be asked to keep lengthy diaries solely to demonstrate innocence. The purpose is to see meaningful decisions and skills at appropriate points.
Nor is this problem confined to digital tools. A parent, tutor or teacher can over-scaffold an assessed task, leaving the final product beyond what the student could achieve alone. NZQA's guidance on authenticity identifies excessive guidance alongside copying and inappropriate AI use. Authenticity is a question of independent achievement, not simply software detection.
A five-part evidence model for authentic learning
Schools can use five complementary forms of evidence. No achievement standard requires this exact framework, and teachers should not mechanically add all five to every task. Rather, select the smallest combination that validly reveals the knowledge and skills required by the standard.
1. A starting-point demonstration
Before students receive extensive feedback or use permitted digital support, ask them to complete a brief independent task. A science class might interpret a short results table; a history class might distinguish a primary source from a later interpretation. This creates a baseline.
A baseline is not a trap or a prediction of final grades. It tells the teacher where teaching is needed, and it provides context when reviewing subsequent work. Allow the accommodations and assessment arrangements required for that learner.
2. Visible decisions during the process
A good checkpoint focuses on one consequential choice, rather than collecting every keystroke. The student might explain why they rejected a source, revised a hypothesis or changed the order of an argument.
For a long investigation, two brief check-ins may be enough: one when the research question is refined, another when evidence is interpreted. The number should reflect task length, risk and workload. Record enough to support an honest decision; avoid surveillance for its own sake.
3. A final product aligned to the standard
A final report, practical performance, portfolio or solution can still matter profoundly. The mistake is to treat presentation quality as proof of every underlying skill. Mark the criteria actually described by the achievement standard, not the student's willingness to add attractive formatting or elaborate language.
A restrained rubric asks: Which claims, calculations, designs or arguments are evidence of the intended learning? Which features are incidental? Which forms of assistance were permitted?
4. A brief explanation from the learner
A two-minute conversation can expose understanding that a finished page cannot. A teacher might ask: “Show me the evidence for that claim,” “What alternative did you reject?” or “Which step would fail if this assumption changed?”
Such questioning is valuable for formative diagnosis and can inform proportionate authenticity checks. It should not silently become an additional assessed component or change published conditions. If further evidence is required for a regulated assessment, follow the relevant NZQA and school procedures and ensure fairness across learners.
5. An appropriate transfer opportunity
Ask the learner to tackle a new example that requires the same principle. A student who understands proportional reasoning should recognize it when the numbers, units or context change. A student who grasps causal argument should distinguish correlation from causation in a new source.
Transfer tasks are particularly informative during teaching, revision or permitted verification. Their difficulty should be comparable, and they should not introduce unnecessary language, cultural or accessibility barriers. The goal is evidence of flexible understanding, not a surprise second examination.
What could authentic assessment look like in different NCEA subjects?
There is no universal “AI-proof assignment.” Tasks differ because standards assess different capabilities. The following adaptations are classroom design illustrations; they must be checked against current standard-specific conditions before use for assessment.
English: from the polished essay to defensible interpretation
Students study a text and prepare an interpretation. Before the final writing stage, the teacher asks each learner to identify a passage, explain a choice of language and defend one inference. During permitted learning activities, classmates can challenge alternative readings.
What becomes visible is not merely elegant prose but whether the learner can connect a claim to the text. If AI is used for practice when permitted, it may help generate counterarguments; the learner must still make and justify the interpretive decisions that the standard assesses.
Mathematics: from answer production to mathematical reasoning
An AI tool can supply a solution that looks plausible even when an assumption is wrong. During lessons, present students with two different solutions and ask which is valid and why. Then ask for an independent solution to a new problem.
The assessment task itself must comply with calculator, digital-tool and other conditions. The classroom habit worth building is to identify the method, check constraints, test the result and explain an error. A correct number is useful evidence, but it may be incomplete evidence of reasoning.
Science: from attractive reports to evidence-based conclusions
A group performs an experiment and records observations. Students independently identify anomalous results, discuss limits of the method and connect conclusions to the data. Those decisions can be observed during ordinary teaching and at appropriate authenticity checkpoints.
Do not use generative AI to fabricate experimental observations or to create a student's assessed conclusion. Where AI is allowed during exploration, it can be critiqued against measured data; it cannot replace data collection or student interpretation.
Social sciences: from generic reports to context-sensitive judgment
A student investigates a local planning decision. Their argument should reflect the place, the people affected, the sources available and the limitations of each claim. A teacher asks why a local source deserves weight and which perspective has been left out.
An AI-generated overview may sound authoritative while flattening New Zealand-specific context. This makes critical checking especially important when work involves Māori perspectives, local histories, mātauranga Māori or Pacific communities.
A practical NCEA authenticity workflow for school leaders
A school does not need a complicated new software stack to improve its evidence. It needs a shared operating routine that teachers, ākonga and whānau understand.
Before assessment: The curriculum leader and subject team review the standard, current NZQA guidance, assessment conditions and school policy. They identify the exact knowledge or performance being credited. Students receive written information explaining whether AI is prohibited, permitted in a specified preparatory way or irrelevant because the task is completed under controlled conditions.
At the beginning: The teacher demonstrates acceptable referencing and explains why acknowledged use does not make prohibited AI-generated evidence valid. Students see a concrete example of the difference between learning support and having a tool complete the task.
During the task: Teachers choose meaningful checkpoints, retain a small sample of student reasoning and notice whether the learner can discuss the work naturally. Feedback helps students improve within the support limits set by the standard.
Before a decision: The assessor considers the submitted evidence and any genuine concerns proportionately. Unusual vocabulary or a shift in style might prompt a conversation, but it is not proof of misconduct. AI detectors are fallible and should not determine an allegation on their own.
When a concern arises: Follow the school's documented process, preserve the evidence, give the student an appropriate opportunity to respond and comply with NZQA requirements. Avoid ad hoc penalties imposed by an individual teacher.
NZQA's information on authenticity processes for externally assessed standards emphasises clear expectations, checkpoints, records and timely communication with students and families. These measures work best when they are part of ordinary teaching rather than an emergency search for evidence after submission.
The one-page assessment design canvas
Before releasing a senior assessment task, a teaching team can answer these six questions together.
| Design question | Decision to record |
|---|---|
| What does the specific standard assess? | Name the knowledge, skill, process or performance—not just the finished artefact |
| What assistance is permitted? | Specify AI, peers, teachers, sources and technology according to the official conditions |
| Where will independent understanding be visible? | Choose a baseline, decision checkpoint, appropriate performance or explanation |
| How will accessibility be protected? | Confirm language, disability, technology and other approved assessment supports |
| What will happen if authenticity is uncertain? | Use the school's fair process and any NZQA-required escalation |
| What can the student learn from the experience? | Plan feedback that strengthens strategies for the next task, without altering the current assessment rules |
Teams can complete this canvas in an existing department meeting. The resulting instruction sheet should be short enough for students and whānau to read, but precise enough that two teachers interpret it consistently.
A practical wording for permitted practice might be: “You may discuss concepts with an approved learning tool during the teaching phase. You may not submit generated text, analysis or images as your own assessed evidence. Your internal assessment follows the conditions provided for this standard.” A department must adapt that language to its actual policy and task; the example is not a national rule.
Why AI detection is a weak foundation for trust
When schools are anxious about academic integrity, it is tempting to buy an AI detector and treat its score as objective proof. That approach is fragile. NZQA itself advises care because of false positives. A detector typically examines characteristics of text; it does not witness the student's learning process.
There is a human cost as well. A student writing in additional-language English may be treated suspiciously for stylistic reasons unrelated to dishonesty. A diligent student who has improved their writing may be challenged because the result is unexpectedly polished. Conversely, an AI-generated passage can evade a detector.
A better investigation begins with evidence that is directly connected to the learning: drafts where appropriate, dated checkpoints, comparison with earlier work and respectful, targeted questioning. This does not eliminate misconduct, and schools still need clear rules and consequences. It makes decisions fairer and more defensible.
Trust is also educational. Students should know what independent achievement means before they are assessed. They need opportunities to practise paraphrasing, citing, checking AI claims and identifying the point at which assistance stops being assistance.
Learner agency and culturally responsive assessment in Aotearoa
AI assessment policy must not treat the class as a group of identical writers with identical access to technology. Rural connectivity, device availability, home support, language proficiency and approved accommodations can change the conditions under which students work. A school that relies on expensive home access to show progress risks confusing resources with capability.
The Ministry also cautions that AI models may be weak on mātauranga Māori, te reo Māori and Pacific languages. A fluent AI summary of a local tradition is not necessarily a reliable account, and students should not be encouraged to upload restricted, personal or culturally sensitive material into a public tool.
The response is not to turn cultural knowledge into a generic AI literacy exercise. Consult relevant people and authoritative resources; respect local protocols; use appropriate language and contexts; and retain human educational judgment.
Learner agency also means helping ākonga understand the quality of their own evidence. After a low-stakes learning activity, invite them to reflect on what they can now explain without prompts and which part they still need to practise. Reflection should guide learning, not become a theatrical statement that every sentence was entirely unaided.
What does the coming NCEA replacement change?
This discussion is not only about existing NCEA standards. In 2026, the Ministry announced plans for a new subject-based senior secondary qualification structure. According to its published qualifications update, the transition includes a new foundational qualification for Year 11 from 2028, the New Zealand Certificate of Education (NZCE) for Year 12 in 2029, and the New Zealand Advanced Certificate of Education (NZACE) for Year 13 in 2030. Schools should consult current Ministry and NZQA implementation guidance because transitional details can evolve.
The reform creates an opportunity to improve assessment design, but it does not automatically solve the authenticity problem. Subject-based systems can still reward sophisticated-looking work that obscures who reasoned, wrote or solved. The underlying principle survives the change of qualifications: assess the learner's demonstrable competence, not the convenience of the technology around them.
For school leaders, a sensible 2026–2027 priority is to improve the quality of ordinary classroom evidence now, train staff in consistent authenticity decisions and make the resulting routines portable to the new framework. Avoid building elaborate AI rules that must be rewritten with every product release.
From checking work to coaching the learning process
Assessment safeguards prevent false claims about achievement. They do not, by themselves, help a student become a better learner. This is where a learning-coaching approach adds a useful second question.
Instead of asking only, “Is this student's answer authentic?”, the teacher can ask, “What does the student need to learn next so they can produce and defend the answer independently?”
Imagine a Year 11 learner who can choose strong quotes for an essay but struggles to connect them to a claim. Simply showing an excellent AI-generated paragraph may create temporary progress without teaching the missing skill. A teacher working as a learning coach can model one claim–evidence–reasoning connection, invite the learner to construct another, offer specific feedback and gradually withdraw the support.
This is consistent with a broader educational distinction: receiving information is not the same as constructing usable understanding. Learning Coach's educational approach emphasises observing where comprehension breaks down, asking diagnostic questions and helping learners progressively take responsibility for their own learning. That is a classroom role, not the same as an instructional coach employed to coach other teachers.
For educators exploring this shift, the book Stop Teaching. Start Coaching. examines the difference between delivering explanations and developing independent learners in an AI-rich world. Teachers and leaders interested in building learning-coaching skills can also explore the Learning Coach training pathway, checking the current speciality options and admission information directly rather than assuming immediate availability.
The purpose is not to replace explicit teaching, NZQA authority or professional assessment judgment. It is to make independent understanding a daily educational objective rather than a last-minute authenticity test.
Frequently asked questions
Can NCEA students use ChatGPT to prepare for assessments?
They may use appropriate tools in learning activities only where school policy, tool age limits and other requirements allow. The rules for work submitted for NCEA assessment are more restrictive. Teachers must check the particular standard and conditions; generative AI is not permitted in external assessment.
Can an AI-generated paragraph be submitted if the student cites the tool?
Attribution is important, but citation does not make an otherwise prohibited use permissible. A student cannot present AI-produced evidence as their own achievement. Assessment conditions govern what is acceptable, not a generic disclosure convention.
Should every NCEA internal assessment include an oral defence?
No. A brief conversation may help a teacher understand an individual student's work or check authenticity, but it must not become an unauthorized new assessment criterion. Select checks that suit the standard and follow school and NZQA requirements.
Are AI-detection tools enough to prove cheating?
No. NZQA cautions against relying on detectors alone because they can produce false positives. Decisions need fair procedures and evidence relevant to how the work was produced.
Will replacing NCEA remove the problem?
No. Qualification structure may change, but valid assessment will still need to show what the learner can know and do independently, with suitable fairness, accessibility and authenticity safeguards.
The real test of learning
A confident student should be able to say more than “I finished it.” They should be able to show the source that mattered, explain the decision they made, identify the step they would change and demonstrate the core idea again in an appropriate new situation.
New Zealand schools cannot control every new AI model that appears. They can control the clarity of assessment conditions, the opportunities students have to learn independently and the care used when judging evidence. That is a more durable response to the AI era—and a better educational ambition than trying to catch every suspicious sentence.
Resources and official guidance
New Zealand Ministry of Education — Generative AI: school practice and NCEA.
NZQA — Guidance on the acceptable use of artificial intelligence.
NZQA — Authenticity processes for externally assessed standards.
New Zealand Ministry of Education — More details about new senior secondary qualifications.
Information and policy links reviewed 27 October 2026. School-specific policies, achievement-standard specifications and later official changes take precedence.
