A pupil uses AI on a task and produces a better answer than they would have alone. Did they learn more? Often, the evidence suggests, they learned less.
The Australian Education Research Organisation (AERO), the country's national education evidence body, calls this the “performance paradox”: pupils “perform better on a task when using AI but learn less in the process”. It sits at the centre of guidance AERO published on 14 October for school leaders and teachers, alongside a companion explainer for policymakers.
AERO is open about the limits. Research on AI and pupil learning in schools is limited, most of it comes from secondary and higher education, and the guidance is described as provisional. That honesty is part of what makes it useful: it builds from what is well established about how pupils learn, rather than from claims about AI.
The core test: support or replace
AERO's starting point is that learning is a change in long-term memory, built through effortful thinking and practice. AI helps when it supports that work and harms when it replaces it.
Its example is clear enough to put in front of any staff meeting. AI may support learning if it is used to edit an essay a pupil has written, or to generate a worked example in maths. It bypasses learning if a pupil uses it to write the whole essay or to produce the answer to a maths problem.
The guidance names further risks: pupils becoming over-reliant on AI to plan and check their work, developing “illusions of knowledge” because they can get answers quickly, and lacking the prior knowledge to spot inaccurate, biased or fabricated outputs.
Five considerations for schools
AERO groups its guidance into five considerations:
- Prioritise evidence-based teaching and learning. If AI is used, it should scaffold learning, for example by asking probing questions rather than giving answers.
- Monitor for risks to learning. Watch for the performance paradox and cognitive offloading that undermines learning.
- Set clear expectations for AI use. Make explicit when pupils work independently and when AI may help, and teach how AI works, its limits and how to question its outputs.
- Consider approaches to assessment. Ask whether take-home and unsupervised tasks still give valid evidence, how pupils should acknowledge AI use, how to keep assessment fair for pupils who choose not to use AI, and whether AI detectors disadvantage groups such as pupils learning English as an additional language.
- Apply a culturally responsive lens. AI systems trained mainly on Western data can misrepresent cultures and knowledge, so teachers should give pupils trusted local sources to test AI responses against.
Extra care with younger pupils
AERO suggests schools may need to take greater care when introducing AI in primary years, including considering age restrictions for AI tools and placing more emphasis on teacher guidance and oversight. Younger pupils are still building the background knowledge needed to recognise when an AI output is wrong.
That does not mean primary pupils cannot learn about AI. It means the emphasis falls on understanding, discussion and teacher-led examples rather than pupils using general-purpose tools independently.
A test for every task
The most practical thing a school can take from AERO is a single question for teachers to ask of any task where AI might appear: is the AI doing work that supports the thinking, or the thinking itself?
Teachers can apply it in a minute. Pupils can learn to apply it too, and when they can, they are practising exactly the judgement AI literacy is meant to build.
Sources: Australian Education Research Organisation, AI in schools: considerations for school leaders and teachers, 14 October 2026; AERO, AI in schools: considerations for policymakers, 14 October 2026. AERO describes the guidance as provisional and notes very limited research from primary settings.