The most important question about artificial intelligence in New Zealand schools is not whether it can write a lesson plan or produce an impressive essay. It is whether students leave the classroom knowing, understanding and being able to do more than they could before.
Short answer: Yes, schools in Aotearoa New Zealand can use AI constructively—but they should embrace purposeful, safeguarded uses, not unrestricted access or automated teaching. The teacher remains responsible for learning, assessment and relationships. School leaders need a clear policy, privacy checks, equitable access, authentic evidence of student thinking and a way to stop a trial that is not working. The right test is not how often AI is used. It is what students learn without depending on it.
This is now a practical leadership decision, not a distant technology debate. In July 2026, New Zealand's Education Review Office (ERO) published a national review of how schools were responding to AI. Its findings reveal widespread use alongside uneven expectations, particularly around student work, safety and capability. The challenge is to move from isolated experimentation to a coherent approach that schools, teachers, students and whānau can understand.
What has changed for New Zealand schools in 2026?
AI tools can generate explanations, translate or simplify text, draft communications, create practice questions, produce images and offer feedback almost instantly. For an overloaded teacher, these capabilities are tempting. For a student facing an unfamiliar assignment, they can be equally tempting—but the two uses are not educationally equivalent.
The Education Review Office's July 2026 national review found that 93% of school leaders surveyed were using AI. The most common leadership uses included drafting communications, summarising information and preparing learning resources. The same review described concerns about critical thinking, assessment authenticity, and the absence of consistent school-wide expectations. These figures describe ERO's surveyed population and context; they are not proof that AI improves student achievement.
The Ministry of Education's generative AI guidance takes a practical approach: consider benefits and risks, check terms of use, avoid entering confidential or sensitive information, build policy with staff, and retain human responsibility for learning. The NZQA advice on AI and assessment adds an assessment-specific boundary: authenticity policies must address acceptable AI use where required, and teachers must verify that assessed work is genuinely the student's.
Together these sources suggest a reasonable position: a school may benefit from AI without treating every AI-generated output as trustworthy or every AI-supported task as learning.
The distinction that matters: assistance versus substitution
Imagine a Year 9 student writing an explanation of why a local river's water quality changes after heavy rain. The student might ask a teacher-approved AI tool for definitions of runoff and sediment, then compare the explanation with evidence from a class investigation. The student identifies an incorrect assumption, revises their argument and explains the findings in their own words.
Now consider a different student who asks the tool to write the complete explanation, pastes it into a document and edits a few sentences. Both students have produced readable work. Only the first has provided clear evidence of developing understanding.
This is the distinction between assistance, which helps the student do the intellectual work, and substitution, which does that work for the student. It is more useful than a blanket claim that AI is either good or bad.
Teachers can ask three questions before authorising a classroom use:
Who is making the important decisions? The learner should select evidence, weigh alternatives and justify conclusions.
What understanding will remain without the tool? Students should be able to explain, retrieve, apply or adapt what they learnt independently.
How will we see the process? Drafts, annotations, discussions, practical demonstrations and reflection should make learning visible.
If these questions cannot be answered, the task may need redesigning before AI is introduced.
Where can AI genuinely help teachers?
Planning and resource preparation
A teacher may use AI to generate possible examples at several reading levels, suggest discussion prompts, build a rough outline for a lesson or propose alternative ways of introducing a concept. However, the output must be checked for curriculum alignment, accuracy, cultural suitability, accessibility and age appropriateness. A convincing explanation can still contain an error.
For example, an intermediate teacher planning a lesson on fractions could ask for three everyday contexts in which students might compare quantities. The teacher then rejects unsuitable examples, checks the mathematics and chooses a context connected to the class. The tool has supported preparation; the teacher has designed the learning.
A school should distinguish time saved from educational value. Quicker worksheet production is not, by itself, evidence that students will remember, reason or transfer a skill more effectively.
Differentiation without lowering expectations
AI may help teachers draft vocabulary supports, clearer instructions or additional practice for students learning English, including multilingual learners. It should not automatically determine what a learner is capable of or place students into fixed ability groups.
The crucial safeguard is preserving the underlying learning goal. If the objective is to analyse an argument, a shorter text may lower an unnecessary reading barrier, but students should still evaluate claims and evidence. Staff should check AI-generated translations and any language or cultural nuances with appropriate expertise.
Feedback that invites thinking
A useful AI-assisted question might be: “What evidence supports your conclusion?” or “Can you identify a different explanation?” An unhelpful use is allowing a system to silently rewrite a student's submission into polished prose.
Teachers can use AI to draft feedback options, but the teacher must decide what is accurate, relevant and respectful. Feedback should lead to another student action: a correction, explanation, revision, worked example or question.
Administration with careful boundaries
Drafting a generic newsletter or formatting a meeting agenda may be lower risk than analysing individual learners. Even administrative work can involve sensitive records. Never assume that removing a name makes a prompt anonymous: contextual details can identify students, families or teachers.
The Office of the Privacy Commissioner's education-sector guidance on digital technologies cautions against putting learners' personal information into free AI tools and recommends due diligence for approved systems too.
Where should schools draw a firm line?
Some uses warrant a clear prohibition, especially before a tool has passed procurement, privacy and safeguarding review.
| Situation | Recommended starting position | Reason |
|---|---|---|
| Teacher drafts generic lesson examples with no student data | Consider with human checking | Lower risk, potentially useful |
| Student uses an approved tool to brainstorm, then explains ideas independently | Conditional, supervised permission | Can support thinking when evidence remains visible |
| Student submits AI-generated writing as their own | Not permitted | Misrepresents authorship and learning |
| Staff paste identifiable student records into an unapproved chatbot | Do not permit | Serious privacy and security risk |
| AI automatically assigns final grades without teacher judgement | Do not delegate | Professional accountability remains human |
| Students upload photographs or voice recordings without permission | Do not permit | Consent, privacy and safeguarding concerns |
| AI detector alone is used to accuse a student of misconduct | Do not rely on it | False positives and procedural fairness risks |
These are prudent school-operating recommendations, not a substitute for an individual school's legal duties or current assessment rules. For externally assessed work, staff must consult the relevant NZQA assessment specifications rather than assuming that permission for a classroom activity extends to an assessment.
What does New Zealand guidance mean for NCEA and authentic assessment?
At senior secondary level, academic integrity cannot be reduced to the question, “Did a detector say this was AI-written?” NZQA describes a combination of stronger assessment design, milestones, observation, comparisons with earlier work, referencing, follow-up questioning and appropriate declarations. It also warns teachers not to rely on AI detectors alone.
For an internally assessed project, a teacher might collect a short proposal, a source log, a draft with decisions explained, an in-class progress check and a brief discussion of the final work. These do not guarantee authenticity, but they provide a more defensible view of what the learner actually knows.
External assessment has its own constraints. NZQA's guidance says that chatbots, generative AI and paraphrasing tools cannot be used to generate content submitted as a candidate's own work in external assessments. Teachers should check the specific assessment rules in force at the time rather than generalising from this sentence to every context.
For a deeper treatment of NCEA internal and external authenticity, the related Learning Coach guide, NCEA in the Age of AI: How Can New Zealand Schools Assess Real Learning?, examines assessment-specific evidence and responsibilities. This article focuses instead on the whole school's adoption decisions, from privacy and classroom design to leadership and evaluation.
Why student privacy needs to come before procurement
A polished demonstration can conceal difficult questions. What data does the supplier collect? Where are prompts and outputs stored? Can the supplier use them to train a model? What happens when the account is deleted? How are under-age users protected? Can families and students understand the system's use?
New Zealand's Privacy Act 2020 and subsequent changes create obligations that schools cannot hand over to a technology vendor. In particular, the new indirect-collection notification principle, IPP 3A, took effect on 1 May 2026, according to the Ministry of Justice. Whether the principle applies to a particular workflow requires a factual assessment of how personal information is obtained; it should not be invoked as a blanket rule for every classroom prompt.
Principals and boards should involve their privacy officer or qualified adviser in reviewing the actual product and configuration. Before any pilot, ask the vendor to explain data retention, access, security, deletion, age controls, staff accounts, safeguards and incident reporting in writing.
A strong default is simple: do not put identifiable learner information into a general-purpose tool unless the school has confirmed that the specific use is lawful, necessary, safe and approved. Test lesson-design use cases with fictionalised or generic examples instead.
Should primary and secondary schools take the same approach?
Not necessarily. The developmental stage of the learner changes the risk-benefit balance.
For younger children, the priority is foundational language, reading, writing, number knowledge, discussion, attention and real-world exploration. Adult-mediated demonstrations may be more appropriate than independent conversations with open-ended AI systems. If the tool does most of the decoding, composing or calculating that the lesson intends to teach, the child loses necessary practice.
ERO's practical guide for primary school leaders notes that pupil AI use is generally more tightly supervised than at secondary level, and encourages age-appropriate policies, staff learning and clear expectations.
At secondary school, students are more likely to encounter these tools outside school. A strict silence about AI is unlikely to create good judgement. Learners need direct instruction on hallucinations, source checking, attribution, privacy, bias, and the difference between an explanation they understand and an answer they merely possess.
In both settings, schools should provide learning activities without AI so students practise independent thinking. Removing technology at the right moment can be as intentional as using it.
What about te ao Māori, te reo Māori and local community knowledge?
Aotearoa's schools operate in different communities and cultural contexts. An AI-generated answer may be confident about a Māori term, local history, whakapapa or tikanga while lacking the authority, context or nuance to be reliable. The availability of fluent-looking text does not establish cultural accuracy.
Where learning involves Māori knowledge or language, teachers should use relevant curricula, authoritative resources and local expertise. Consult mana whenua and appropriate whānau or community partners when knowledge, tikanga or local narratives call for it. Do not place sensitive cultural information into external systems merely to obtain a summary.
AI can be one aid to drafting an accessible classroom resource, but it should not replace relationships with knowledge holders, community consultation or the agency of Māori learners. The same care applies to Pacific languages, migration histories and locally held stories. Cultural responsiveness requires human listening and accountability, not only a more elaborate prompt.
A practical school AI policy: seven decisions to make
The Ministry's guidance recommends a purposeful policy covering use, risk, professional development, privacy and review. The following seven decisions can turn that advice into a document staff can actually use.
1. State the educational purpose
Describe the learning problems the school wants to address. Avoid defining success as the number of accounts activated. A better aim is stronger feedback, clearer explanations, lower preparation burden where appropriate, and greater learner independence.
2. Define approved and unapproved uses
Separate staff preparation, administrative work, classroom demonstrations, supervised student use and assessment. Specify what is never uploaded. Name the process for requesting approval of a new tool; do not treat every website calling itself “educational” as vetted.
3. Explain student authorship
Provide concrete examples of allowed assistance, such as generating revision questions, and disallowed substitution, such as submitting generated paragraphs unchanged. Expectations may differ by age, subject, assignment and assessment conditions. Require acknowledgement when relevant.
4. Preserve teacher responsibility
Teachers approve resources, judge evidence, communicate feedback and decide when additional support is needed. An AI-produced mark, label or risk score is not a professional diagnosis. Provide a way to question and correct the system's output.
5. Put privacy and safety controls into practice
Identify accounts, age requirements, access permissions, approved data categories, vendor assurances, reporting channels and review dates. Make the rules usable on a busy Tuesday morning, not just legally impressive in a policy folder.
6. Address access and inclusion
Ask which students have devices, connectivity, accessible tools and support at home. A classroom requirement should not penalise a learner who cannot use a paid or age-restricted service. Offer equivalent pathways without AI and check usability for students with disabilities.
7. Set review and withdrawal criteria
Decide who owns the policy, how complaints are heard and when evidence will be reviewed. If a pilot increases privacy risk, confusion, dependency or workload without sufficient benefit, the school should be able to modify or end it.
A 30-day implementation plan for principals
A manageable pilot is better than a school-wide announcement with no shared practice.
| Period | Leadership action | Evidence to gather |
|---|---|---|
| Week 1: Listen and map | Survey current staff practice, ask students and whānau about use, identify existing risks | Baseline practices, known tools and concerns |
| Week 2: Set boundaries | Draft permitted-use guidance, check privacy, appoint an AI lead and select a low-risk use | Approved pilot brief, red lines and responsible people |
| Week 3: Pilot deliberately | Support a small teacher group; keep a comparable non-AI pathway | Teacher workload notes, quality checks and student work samples |
| Week 4: Review and decide | Inspect accuracy, accessibility, privacy, engagement and independent learning | Evidence-based decision to continue, change or stop |
Start with a use case that has a clear benefit and limited exposure: perhaps drafting alternative practice questions for teacher review. Avoid launching a personalised system that collects extensive learner data before the school has an appropriate approval process.
ERO's 2026 guidance suggests that an AI champion can help teachers build literacy and share safe practices. The person needs time, training and support; they should not become a single point of failure. Over time, capability should spread among staff rather than sitting with one enthusiast.
Three classroom routines that keep the learner in charge
Routine one: Think first, ask second, decide third
Before students consult AI, require an initial independent attempt: a prediction, explanation, sketch or question. After receiving an AI answer, students annotate what they accept, what they challenge and why. Finally, they produce their own synthesis without copying.
In a science lesson, this means students commit to a hypothesis before using a tool to generate alternatives. They then compare explanations against observed results rather than treating the most fluent answer as automatically correct.
Routine two: Explain without the screen
After an AI-supported activity, invite learners to explain a key concept with their device closed. Ask for an example or a transfer question: “How would this change if the situation were different?” This helps teachers notice the gap between familiarity and understanding.
An unexpected difficulty is not evidence that AI is inherently harmful. It is a teaching signal: the student may need more worked examples, retrieval practice, discussion or time to reason.
Routine three: Keep a decision trail
Ask students to keep short notes of where the tool helped, what they checked and what they changed. A decision trail can contain prompts or screenshots only when allowed by school privacy policy. The aim is reflective judgement rather than surveillance.
These routines also illustrate a core principle of learning coaching: adults can scaffold the process without permanently owning the learner's decisions. As competence grows, support should be reduced, not endlessly automated.
How should principals measure whether AI is working?
Schools often collect the wrong metrics first: number of prompts, teacher logins, generated worksheets or minutes supposedly saved. These can show adoption or convenience, but not meaningful learning.
Choose a small evaluation set before a pilot begins:
Independent performance: Can students explain and apply the target skill when AI is unavailable?
Quality of reasoning: Are students checking evidence, noticing errors and defending choices?
Teacher workload: Has time shifted towards useful teaching, or towards checking machine output?
Equity: Are students with fewer resources or additional learning needs benefiting, participating or being excluded?
Trust and safety: Are privacy concerns, incidents and complaints reported and addressed?
Compare work against an appropriate baseline, account for differences between groups and avoid announcing improvement on the strength of anecdotes alone. A short pre-task and post-task in comparable conditions can be informative; rigorous causal claims require stronger design.
An effective pilot might show modest time savings but significant gains in teacher feedback quality. Another might generate attractive materials while students become more passive. Both findings are useful. Evidence—not the vendor's demonstration—should determine the decision.
Communicating with parents and whānau
Families are more likely to support a policy they can understand. Schools should tell them which tools are approved, why they are being considered, what information is protected, what students may submit as their own work, and what alternatives are available.
A useful message avoids both panic and salesmanship:
Our school is exploring carefully selected AI uses to support teaching and learning. Teachers remain responsible for instruction, assessment and student wellbeing. We will not require students to share personal information with unapproved tools, and we will teach learners to question AI output rather than copy it. Families can contact the school with concerns or accessibility needs.
Offer practical examples at parent evenings. A ten-minute demonstration of how an AI tool can be inaccurate may teach more than a brochure promising personalisation. Invite perspectives from students and whānau, including those who are less comfortable with technology.
How learning coaching can help schools move beyond the AI debate
There is an important professional shift beneath these tools. A teacher who knows when to explain and when to ask a student to think aloud can create better learning conditions than a system that constantly supplies complete answers.
Learning coaching begins with observation. What does the student understand already? Where does attention drift? Which task is difficult because of missing prior knowledge, and which because the instructions are unclear? What would allow the student to take the next step without adult or machine dependence?
Consider a student who repeatedly asks AI to simplify every paragraph. A learning-coaching response is not to remove the tool abruptly or permit endless simplification. It is to identify the barrier: unfamiliar vocabulary, background knowledge, comprehension strategy or confidence. The teacher models one strategy, asks the learner to try it, checks the explanation and gradually releases responsibility.
The Learning Coach approach emphasises the difference between isolated information and meaningful understanding, and between alertness and sustained, directed attention. Those distinctions are especially relevant when AI makes information abundant and distraction effortless. They should be applied thoughtfully, not treated as guarantees of improved outcomes.
Educators interested in this professional posture can explore Learning Coach training and its specialisations. The public offering describes an Enseignant Coach / Teacher Coach pathway, while programme availability, language delivery and admission arrangements should be confirmed directly before applying. The English book Stop Teaching. Start Coaching. also explores why learner independence matters in an age of instant answers.
Frequently asked questions
Should schools ban ChatGPT and other generative AI?
A school can restrict particular tools or uses on safety, privacy, age or assessment grounds. A blanket ban alone, however, does not teach students how to evaluate AI claims outside school. Clear boundaries plus explicit AI literacy are usually more educationally useful than silence.
Can teachers use AI to mark student work?
The Ministry's guidance supports appropriate AI assistance under human oversight, with teacher responsibility and professional judgement preserved. Schools must check current rules, approved tools, confidentiality, bias and the type of assessment. An automated suggestion is not a final professional decision.
Is AI suitable for primary pupils?
A narrow, adult-mediated use can sometimes support a lesson, but independent student access raises age, safety and learning concerns. Start with curriculum purpose, developmental needs, approved settings and the ability to deliver the lesson successfully without AI.
How can teachers tell whether a student used AI?
No single AI detector is a reliable proof. Better evidence includes observed work, drafts, in-class tasks, authentic discussion, previous samples and proportionate follow-up, with fair procedures when misconduct is suspected.
What is the first step for a principal?
Audit what staff and students are already doing, especially unapproved tools and personal-data practices. Then agree on a limited educational purpose, safeguards and a reviewable pilot before purchasing or mandating a product.
The question to take into the next staff meeting
A school that says “yes to AI” without defining what students must still do for themselves is making a technology decision, not an educational one. A school that says “no” without teaching critical evaluation is leaving an important part of modern literacy untouched.
The better choice is intentional: protect learners, support teachers, use evidence, involve communities and design tasks that make thinking visible. In Aotearoa New Zealand, embracing AI should mean strengthening the human work of education—not making students more dependent on fluent answers.
Resources and references
New Zealand Ministry of Education — Generative AI: guidance for schools and teachers. Official guidance on school policy, teaching, privacy and responsible use; consulted 27 October 2026.
Education Review Office — Ready or not: How are schools responding to Artificial Intelligence? Summary Report. National review, published 30 July 2026; consulted 27 October 2026.
Education Review Office — Insights for Primary School Leaders and Teachers. Practical sector guidance, 30 July 2026.
NZQA — Guidance on the acceptable use of Artificial Intelligence. Official school assessment and authenticity guidance, consulted 27 October 2026.
Office of the Privacy Commissioner — Digital technologies: education-sector privacy guidance. Official guidance on AI tools and children's personal information.
New Zealand Ministry of Justice — Enhancing the Privacy Act. Official explanation of privacy law amendments effective in 2026.
