The 2026 NAPLAN results came out on 5 August, and two findings led the coverage. Year 3 numeracy had its lowest average score since the current scale began in 2023. Average reading scores in Years 5, 7 and 9 were lower than in 2025. Explanations followed quickly. Federal Education Minister Jason Clare wondered whether students spend so much time writing short messages that they get little practice at long paragraphs, and NSW Education Minister Prue Car said she was keen to see whether pen and paper testing would help.

Generative AI is the newest item on that list of suspects. Whether it belongs there depends on what the results show and when the problems began.

What the NAPLAN 2026 results actually show

NAPLAN moved to a new scale and new proficiency levels in 2023, and results from before then can't be compared directly with results since. A score that is the lowest since 2023 is the lowest of four years.

The same release had good news that got less airtime. Numeracy averages in Years 5, 7 and 9 were the highest since 2023, and about 29,000 more students reached the Exceeding level than in 2023. Writing in The Conversation, Sally Larsen of the University of New England found the differences between cohorts since 2023 were negligible in size.

ACARA's chief executive, Stephen Gniel, still called the Year 3 numeracy drop "clearly concerning", and the wider numbers explain why. About one in three students has fallen short of expectations every year on the new scale, and one in ten needs additional support. A small change over four years sits on top of a large group of students who have been behind the whole time.

Year 3 numeracy is a teaching question

Year 3 students are eight and nine years old, and chatbots are an unlikely explanation for their maths results. The debate has been about curriculum and instruction. Trisha Jha of the Centre for Independent Studies argues the Foundation to Year 2 maths curriculum spends too little time building fluency. Education ministers are rolling out a national Year 1 numeracy check, and ACARA is reviewing the Foundation to Year 2 maths curriculum.

Reading shows what a change in teaching can do. Clare declared "the reading wars are over" and pointed to schools using explicit teaching, phonics and small group tutoring, where he said results are going up year after year.

Writing was falling before ChatGPT

Writing is where AI could do the most harm, and the dates show when its decline began. In October 2022 the Australian Education Research Organisation (AERO) published an analysis of more than 10 million NAPLAN writing results from 2011 to 2021. Persuasive writing scores had declined since 2011, most clearly in Years 7 and 9. By Year 9, 85% of students were constructing sentences at or below the level expected of Year 7, and most could only punctuate at the level expected of Year 3.

ChatGPT was released in November 2022, a month after that report. The decline it describes was measured before students had access to a chatbot.

Researchers have argued about how big the decline is. Associate Professor James Ladwig criticised the report for leaving out its error estimates, calculated that the headline drop came to about one mark out of 48, and asked whether Year 9 students try hard on a test with no stakes for them. Those are fair questions about size and certainty. A decade in which writing went nowhere in Years 3 and 5 and slipped in Years 7 and 9 still points at how writing has been taught, well before AI was in any classroom.

By Year 9, 85% of students were writing sentences at or below the level expected of Year 7. That was measured before any of them had used ChatGPT.

Where AI fits in the NAPLAN results

Students sit NAPLAN under supervision with no chatbot available. Any effect AI has on the results will come from the practice students do, or avoid, in the months before the test.

A 2025 study published in PNAS tested what happens to learning when students practise with AI. Researchers gave nearly 1,000 students in grades 9 to 11 at a high school in Turkey access to GPT-4 for maths practice. One group used a standard chat interface. Another used a tutor version that gave teacher-designed hints and did not hand over answers. During practice, the chat group scored 48% higher than students without AI, and the tutor group 127% higher. On the exam that followed, taken without AI, the chat group scored 17% lower than students who never had access. The tutor group scored about the same as those students.

That study was in maths. Writing is even easier to hand to a chatbot, and the practice a student skips when a chatbot drafts the paragraph is sentence construction and punctuation, the same skills AERO found missing in Year 9.

There is no NAPLAN evidence yet that AI caused any part of the 2026 results. The concern is about the next few years: the skill that was already weakest is now the easiest one to outsource.

So is it AI or teaching methods?

It's both, at different stages. Teaching explains how the writing weakness built up over a decade, and it is where the Year 3 numeracy debate sits. AI makes the weakness easier to leave unfixed, because the practice students most need is the practice a chatbot will happily do for them.

In both sets of evidence, the way the teaching or the tool was set up changed the outcome. Reading results rose in schools that changed how they taught it. In the PNAS study, students whose AI was set up to give hints did no worse on the unassisted exam than students who never used it.

What schools can do this term

edit_note A check worth running

Take a recent Year 7 or Year 9 writing task and mark which parts a chatbot could do for the student. If the list includes building the sentences and punctuating them, the task lets students skip the practice NAPLAN says they need most.

The Year 1 numeracy check and the early years curriculum review will take several years to reach Year 3 results. Writing practice in Years 7 to 9 is something a school can change this term. For the senior years, NSW has already moved: from Term 4 2026 an HSC course can carry only one take-home task, which we cover in NSW has capped take-home assessment.

Sources

NAPLAN Literacy AI Policy