A Year 9 student in Sydney can write a remarkably sophisticated prompt. She can ask an AI system to compare historical accounts, adopt a particular tone, structure a persuasive essay and critique a draft against assessment criteria. She may even know how to obtain a better response by supplying more context. But when her teacher closes the laptop and asks, “Which claim do you actually agree with, and why?”, the answer is less fluent.
That moment reveals a question Australian education cannot afford to avoid: does prompt proficiency demonstrate independent thinking, or merely proficiency at directing a machine?
Prompting is a useful digital literacy skill. It is not the same as knowing a subject, evaluating evidence, constructing an argument or solving a novel problem. The real educational challenge is to help students use AI when it contributes to learning without allowing an apparently impressive final product to conceal missing understanding.
In October 2026, the Australian Education Research Organisation (AERO) emphasised the limited robust evidence about AI's impact on learning in schools and advised educators to prioritise established teaching practices, clarify expectations, monitor risks, rethink assessment and remain culturally responsive. That is not an argument for panic. It is an invitation to teach the human part of learning more deliberately.
The difference between a clever prompt and an independent mind
Independent thinking is the capacity to form a question, mobilise relevant knowledge, weigh alternatives, make and justify a decision, notice uncertainty, and revise one's position when evidence warrants it. It does not mean thinking without teachers, peers, books or technology. Independence means owning the reasoning, including the decision to seek appropriate assistance.
A good prompt is an instruction to a digital system. It may show that a student understands the task, audience and desired output. Yet the quality of the response still depends on what the model supplies, and the student may accept a plausible result without being able to detect a missing premise.
Consider three actions in a Year 8 geography task on flood risk. The student could ask, “Write a paragraph explaining why some Australian communities experience higher flood risk.” The model may produce smooth prose. Alternatively, the student could propose three explanatory factors first and ask the model to challenge their reasoning. Or the student could evaluate local maps and authoritative material, then use AI to suggest a counterargument.
Each action involves prompting. Only the second and third clearly require an initial student contribution, though neither by itself proves enduring learning. A later explanation and unfamiliar application provide stronger evidence.
This distinction is increasingly important. The OECD Digital Education Outlook 2026 cautions that better immediate task performance with a general-purpose chatbot does not automatically translate into learning gains. It supports purposeful, educationally designed AI use rather than treating faster completion as success.
What Australia's curriculum actually asks students to develop
Australian schools do not need to invent independent thinking as a new learning outcome. The Australian Curriculum already connects AI learning to digital literacy, critical and creative thinking, literacy and other capabilities.
The Australian Curriculum's AI connection identifies understanding AI systems, engaging with them critically and exploring their ethical implications as connected educational work. It distinguishes generative AI from a search engine and calls attention to questions of bias, reliability and purposeful language.
This offers a practical interpretation for schools: teaching students to write prompts is only one part of a larger curriculum sequence.
A student should be able to explain what kind of system they are using, why a response might sound authoritative yet be mistaken, what evidence could verify its claims, whose perspectives may be absent, and whether using it is appropriate for the assignment. These questions are more demanding than memorising a collection of prompt templates.
Australia's national Framework for Generative AI in Schools also centres teaching and learning, human and social wellbeing, transparency, fairness, accountability, and privacy, security and safety. The framework is national guidance, while local policies and assessment rules still determine what a school can permit.
The goal, then, is neither “every student should use AI” nor “every student should avoid it”. It is to make the learning objective determine the tool decision.
Why the answer on the screen is a poor measure of what students know
A finished essay, correct maths answer or polished oral presentation is evidence of a product. It is not complete evidence of the processes that produced it.
This is sometimes called a performance–learning gap: a learner may perform well with extensive assistance but retain little capacity to reproduce or adapt the work independently. It can happen with answer keys, over-helpful adults and step-by-step templates, not only AI. Generative AI makes assistance immediate, persuasive and extremely easy to hide.
A 2025 study by Bastani and colleagues in Proceedings of the National Academy of Sciences examined AI assistance in high-school mathematics. Students using less-constrained AI could perform better during practice yet perform worse on a later unaided assessment than students without that assistance; a more carefully designed tutoring condition reduced this problem. The study does not prove that every AI tool harms every learner. It shows why the structure of assistance matters and why teachers should check learning after that assistance disappears.
Australian teachers should be particularly careful not to leap from a student's impressive AI-assisted submission to a judgement about lasting comprehension.
One useful question changes the conversation: what intellectual work would still be visible if the chatbot were unavailable?
A classroom routine: Think, Prompt, Challenge, Own
The following four-part routine is an editorial teaching proposal, not a validated research instrument or an official Australian government model. Its purpose is to keep substantive reasoning with the student while allowing limited, supervised AI assistance where school policy permits.
1. Think: make a first commitment before opening AI
Begin with a short independent attempt: a hypothesis, quick concept map, outline, proposed solution or list of questions. The response need not be correct. What matters is that the learner places some of their own knowledge on the table.
In a Year 7 science lesson, for example, students predict what happens to the temperature of a sealed container exposed to sunlight and explain their reasoning in two sentences. The teacher can then see the student's initial model and misconceptions.
Without this first step, an AI-generated explanation may become the learner's starting point rather than something to interrogate. For novices, the teacher should first supply necessary content and a worked example; asking students to struggle without foundational knowledge is not sound pedagogy.
2. Prompt: ask for a contribution, not the completed assignment
When AI use is appropriate and permitted, students can request an alternative explanation, a counterexample, a diagnostic question or feedback on the logic of their draft. A prompt such as “Ask me two questions that would test whether my explanation is complete, but do not give me the answer” generally keeps more thinking with the learner than “Write the full answer for me”.
The key decision is not how sophisticated the prompt sounds. It is which steps the student still must do. An AI system might supply an analogy for a difficult concept, but the learner should select and justify which analogy is accurate.
Teachers should not assume that an AI service will always respect the instruction to withhold an answer. Students need a reliable fallback: stop, close the chat and return to class materials or human help if it starts replacing the task.
3. Challenge: interrogate the response with external evidence
Students compare the output with the textbook, trusted documents, teacher modelling, data, primary sources or observations. They identify unsupported statements, ambiguous terms, omitted viewpoints and assumptions.
For a Year 10 history task, a model might offer a neat causal explanation for a major policy decision. Students should mark which claims are documented, distinguish fact from interpretation and use credible sources to test the conclusion. They must be able to reject an AI response even when it is grammatically elegant.
Critical thinking is not automatically achieved by typing “Are you sure?” into the same chatbot. A model's revised answer is another claim to examine, not independent proof.
4. Own: demonstrate the learning without the system
Finally, students explain the conclusion in their own words and apply the principle to a new, appropriately challenging question without AI. In science, this might mean predicting what changes if one variable is altered. In English, it might mean justifying a reading of an unfamiliar paragraph. In geography, it might mean analysing a second map.
The teacher listens for reasoning, not mere recall of AI phrasing. If the learner cannot explain the key step, the next lesson should provide targeted teaching and guided practice rather than punishment.
This last stage is particularly important because the educational objective is transferable understanding, not the production of a flawless first draft.
What this looks like in three Australian classrooms
English: from polished opinion to defensible argument
A Year 9 English class examines whether a persuasive text uses evidence fairly. Students first annotate the text and write a claim supported by a quotation. The teacher models how an argument can be strong even when it acknowledges a counterargument.
Only then, where permitted, students ask an approved AI tool to suggest a competing interpretation. Each student must decide whether that counterargument is reasonable and locate textual evidence for the decision. They hand in an annotated draft showing one claim they kept, one they revised and why.
The next day, students respond to a new short passage without AI. The teacher learns more from their explanations than from a single polished paragraph.
Mathematics: from instant solutions to method selection
A Year 8 mathematics class works with linear relationships. After explicit teaching and worked examples, students attempt a problem on paper. Some students explain why a graph and an equation represent the same relationship; others confuse slope with intercept.
If school rules allow, an approved tool can generate a parallel example or offer a hint. Students must then identify the error in a worked solution and solve a changed problem unassisted.
The danger is allowing the tool to calculate every result before the learner has practised the method. The opportunity is using feedback to expose a misconception while the student still chooses and applies the mathematics.
Humanities: from plausible summaries to source judgement
A Year 10 class explores competing perspectives on a question of Australian environmental management. Students begin with teacher-selected reliable materials and acknowledge the cultural and local context of the topic.
An AI-generated summary may be used as a text to critique rather than as a source of truth. Students identify whose evidence is represented, whose is missing and what cannot be concluded without further investigation. For First Nations knowledge and perspectives, teachers use authoritative community-informed curriculum resources and relevant permissions, rather than asking an AI chatbot to speak on behalf of a community.
The final task asks students to justify a conclusion with identified sources. That practice develops judgment far more directly than a contest over who can obtain the longest answer.
The classroom decision table
| Learning objective | AI could appropriately assist with | Thinking the student must retain | Evidence a teacher can collect |
|---|---|---|---|
| Interpret an argument | Offer a contrasting interpretation | Choose and justify textual evidence | Independent annotation and oral explanation |
| Understand a maths procedure | Provide a limited hint or parallel item | Select and execute the method | New problem solved unaided |
| Evaluate information | Produce a claim for scrutiny | Verify credibility, bias and omissions | Source comparison and reasoned correction |
| Improve a draft | Highlight unclear transitions | Decide which revision strengthens meaning | Annotated revision with reasons |
| Plan an investigation | Raise possible variables | Select the testable question and method | Student-designed plan and rationale |
This table should not be read as universal permission to use generative AI in these subjects. The same activity may be inappropriate for a different age group, assessment condition or tool. It is a planning aid: begin with the intended learning and protect the thinking that makes it possible.
How teachers can distinguish fluent AI use from real learner agency
Learner agency is more than giving students options. It involves acting with understanding, making choices, monitoring progress and seeking the right assistance.
Watch what happens before students prompt: do they define the question, recall relevant concepts and decide what help is needed? Watch what happens during the exchange: do they challenge the response, request clarification and compare it with trusted sources? Watch what happens after: can they explain their choices and work on a nearby task without AI?
These observations are formative, not an AI-detector score. Students may speak, write, draw or demonstrate understanding through an accessible alternative. A short verbal explanation should never be treated as a single definitive indicator of capability for students with language differences, disability or communication needs.
For teachers seeking broader observation strategies, the Learning Coach tools and learning resources present approaches concerned with learner profiles, learning processes and autonomy. They do not replace subject assessment or professional judgement.
What about homework and formal assessments?
Australian schools need a clear distinction between practice, feedback, collaboration and evidence used for grades or credentials. A student may reasonably use an approved tool to clarify a concept during revision while being prohibited from using it to generate assessed work.
In New South Wales, the NSW Education Standards Authority's guidance on AI in RoSA and HSC assessments requires students to follow academic integrity rules and acknowledge material appropriately. On 1 September 2026, the NSW Government announced new limits on take-home HSC assessment tasks as one measure intended to protect authentic evidence of learning. Those NSW decisions should not be described as identical rules in every Australian state or territory.
Schools should state clearly when AI is allowed, what disclosure is required, which sources students must verify and how an assessment will establish individual attainment. Teachers can combine in-class drafting, supervised short tasks, process evidence, conferences and unaided transfer questions according to curriculum and accessibility requirements.
Treating every elegant sentence as evidence of misconduct is unfair. Treating every AI-assisted product as evidence of mastery is equally unreliable. Transparent expectations and varied assessment evidence serve students better than either extreme.
Safety, access and equity come before the prompt
A lesson design is incomplete unless it considers student age, approved technology, privacy, safeguarding, school governance and the accessibility of alternatives. A general-purpose public chatbot may not be suitable or approved for children. Never require students to enter names, learning records, assessment details, health information or other personal data into an unapproved service.
The NSW Department of Education's generative AI guidelines stress appropriate tools and safe, ethical use. Schools should consult current state and sector guidance before any student-facing activity.
Equity is not achieved simply by making an AI link available. Some students have more access to devices and adult support than others. Some multilingual students may be poorly represented by language models, and some students need accessibility adjustments that make an unaided task genuinely comparable. Provide non-AI pathways to the same learning objectives.
AERO's October 2026 advice also calls for cultural responsiveness. In the Australian context, that includes careful treatment of Aboriginal and Torres Strait Islander knowledge, languages and perspectives, with authentic sources and community protocols rather than model-generated stereotypes.
A two-week teaching sequence schools can actually trial
A school does not need to change every assessment at once. A small, carefully governed trial can focus on one year group, one learning objective and one approved tool.
During the first week, a teaching team chooses a curriculum concept, teaches the necessary knowledge explicitly and records how students handle a short task without AI. Teachers then introduce a limited prompt–critique activity and model the difference between feedback and answer substitution. Students keep a brief record of what they asked, what they verified and which decisions they made themselves.
During the second week, students work on a comparable task, practise challenging AI content with authoritative materials and complete a new task without AI. Teachers compare the quality of students' reasoning, not just the attractiveness of their first submissions. They also ask students what the tool helped them understand and when it made them less attentive.
The team then reviews examples together, looking for misunderstandings, accessibility barriers, over-reliance and genuine improvement. The trial is a local teaching exercise, not a controlled study. Its observations cannot support claims of general effectiveness without stronger design and data.
Teachers who already use metacognitive instruction will recognise this emphasis on planning, monitoring and evaluation. The Education Endowment Foundation's guidance on metacognition and self-regulated learning provides a broader research-informed foundation for that teaching work.
The teacher's role becomes more important, not less
When AI can instantly produce explanations, the teacher's distinctive contribution is not merely supplying another explanation. It is diagnosing where learners are in their understanding, choosing the right level of support, noticing what they can do independently and gradually transferring responsibility back to them.
This is closely aligned with a coaching approach to teaching: clarify the learning goal, invite students to make their thinking visible, help them identify usable resources, support reflection and avoid creating unnecessary dependence. Coaching complements explicit instruction and subject expertise; it does not replace them. Beginners still need teachers to explain, demonstrate and correct.
The educational value lies in asking better questions at better moments. “What did the AI say?” can become “What did you know before you opened it?” “Why is this source convincing?” “Which step would you choose without it?” and “Where do you need a hint rather than an answer?”
For educators interested in strengthening these practices systematically, Learning Coach training introduces the coaching perspective on learning, autonomy and learner-centred support. Programme details and eligibility should be checked on that page; professional training should not be mistaken for proof that any AI teaching routine has been independently validated.
Five questions Australian parents can ask tonight
At home, the conversation need not begin with an accusation about ChatGPT. Ask your child to show you the task and explain its purpose. What part did they try first? What did the AI help with? Did they compare the response with their school resources? What would they change in the answer? Could they solve one similar question alone?
A child who cannot yet do the last part may simply need clearer instruction, more practice or an appropriately sized hint. The useful parental response is not to become a second answer engine. It is to help the child choose a next learning step and understand school rules.
The Learning Coach blog explores the wider question of building learning autonomy, beyond any individual digital tool.
Frequently asked questions
Is prompt engineering the same as critical thinking?
No. Prompt engineering concerns how a person structures instructions to an AI system. Critical thinking also involves knowledge, reasoning, evidence appraisal, judgement and willingness to reconsider a conclusion. A student can be good at one and still need instruction in the other.
Should Australian schools ban generative AI?
There is no single answer across all ages, tools or tasks. The national framework provides principles, while specific school, sector, jurisdiction and assessment rules determine what is permitted. Teachers should protect foundational learning and use AI only when its purpose and safeguards are defensible.
How can a teacher know whether AI helped a student learn?
Compare the learner's initial reasoning, choices during assisted work and performance on a relevant new task without AI. Use multiple accessible forms of evidence and interpret them formatively. A polished submission or an AI-detection score cannot establish understanding on its own.
What is the best AI prompt for learning?
Usually one that asks for a limited kind of help the student can act on: a clarifying question, targeted hint, alternative example or challenge to their existing explanation. Even a good prompt is insufficient without prior knowledge, teacher guidance and independent follow-up.
The question that matters after the screen goes dark
Australia's next generation should be capable of doing two things at once: working intelligently with powerful AI tools and exercising judgement that remains their own. These goals need not conflict, but they will not become compatible by accident.
A student who can command an AI system has gained a digital skill. A student who can explain, challenge, decide and learn without it has gained something harder to outsource.
The strongest classroom question is not “How clever was your prompt?” It is “What can you now understand and do that you could not do before?”
For teachers and educational leaders who want to develop that learner-centred practice, explore the Become a Learning Coach pathway and consider how coaching questions, explicit teaching and evidence of independence can work together in your setting.
References and further reading
Australian Government Department of Education — Australian Framework for Generative AI in Schools. National policy principles; framework review endorsed in June 2025.
Australian Curriculum, Assessment and Reporting Authority — Artificial intelligence curriculum connection. Curriculum guidance on AI, digital literacy and critical thinking.
Australian Education Research Organisation — AI in schools: Considerations for school leaders and teachers. Evidence explainer, October 2026.
NSW Department of Education — Guidelines regarding generative AI. Jurisdictional guidance.
NSW Education Standards Authority — Generative AI in RoSA and HSC. Academic integrity and acknowledgement rules.
NSW Government — New rules on AI in take-home HSC assessments. Dated policy announcement, 1 September 2026.
OECD — Digital Education Outlook 2026. International evidence synthesis.
Bastani et al. (2025), PNAS — Generative AI without guardrails can harm learning. Primary experiment in high-school mathematics, not an Australian national estimate.
Education Endowment Foundation — Metacognition and Self-Regulated Learning. Research-informed teaching guidance.
