Schools Need an AI Workload Receipt Before They Call it Time Saved

Australian schools are moving past the question of whether artificial intelligence will enter classrooms. The harder question is whether the promised savings survive contact with the school day.

Education Today has described 2026 as a pivotal year for AI in schools, with leaders building policies, staff capability and tool choices intended to improve learning and productivity. That is the right direction. Yet a tool that produces a lesson outline in seconds may still create extra work when a teacher must check every claim, adjust the reading level, remove unsuitable examples, protect student information and explain why the output cannot simply be accepted.

The missing measure is not adoption. It is net workload.

Schools should require an AI workload receipt for every recurring use they plan to scale. The receipt would record the task, the minutes apparently saved, the checking and correction time added, the errors found, the information that could not safely be entered, and the professional judgement still required. Only the net result should count as time saved.

This matters because polished output can hide weak learning and invisible labour. Education Today has warned that general-purpose GenAI can improve the quality of student work without producing durable learning gains, especially when students offload the thinking they need to practise. The same illusion can affect staff. A fast first draft looks efficient, but the teacher may spend longer verifying it than they would have spent creating a reliable resource from scratch.

That does not make the technology useless. It means schools need evidence about where it genuinely helps.

A useful receipt begins with a narrow workflow. Consider a teacher using AI to create differentiated practice questions. The school should compare the full process before and after adoption: preparing the prompt, reviewing the questions, checking curriculum alignment, correcting ambiguity, adapting for individual students and recording any privacy concerns. If the tool saves 20 minutes but adds 12 minutes of review and four minutes of correction, the gain is four minutes, not 20.

The same discipline should apply to parent communications, meeting summaries, administrative reports and student feedback. Each use carries a different risk. A draft newsletter may need a light check. Feedback that affects a student’s progress requires much closer professional review. The receipt should make that difference visible rather than treating all uses as equivalent.

This approach fits the Australian Framework for Generative AI in Schools, which centres teaching and learning, human and social wellbeing, transparency, fairness, accountability, privacy, security and safety. It also reflects NESA’s June AI transparency statement, which requires a clear purpose, risk assessment, appropriate approval and human judgement for consequential decisions.

Schools can turn those principles into five practical questions.

What specific task is the tool supporting? Who remains accountable for the final result? What must a qualified person verify? What new risks or correction work appear? What evidence would justify expanding, changing or stopping the use?

The answers should be reviewed with teachers, not imposed on them. Staff closest to the workflow know where the hidden labour sits. They also know when a system saves time only because someone else has inherited the checking burden. Inviting teachers to report near misses and disappointing results without embarrassment is essential. Otherwise, leaders will see impressive demonstrations while the correction work stays buried in evenings and weekends.

Schools should also distinguish between tool training and judgement practice. Education Today’s account of a secondary-school Copilot trial stressed deliberate, ethical and transparent engagement. That standard requires more than showing staff how to produce an output. Teachers need practice spotting confident errors, protecting sensitive information, rejecting inappropriate suggestions and explaining their decision to students and families.

A workload receipt would improve purchasing decisions as well. Vendors often demonstrate the fastest part of a workflow. Schools should ask them to support a full trial that measures review time, correction rates, accessibility, privacy controls and the effect on student thinking. A product that reduces clerical work without weakening learning deserves consideration.

One that merely moves effort into verification does not.

The receipt should remain simple enough to use. A one-page record completed during a short pilot is better than a governance system so elaborate that nobody maintains it. The goal is not to slow responsible experimentation. It is to prevent enthusiasm from being mistaken for evidence.

Australian educators are right to explore tools that could return time to teaching. But the most important number is not how quickly AI creates an answer. It is how much trustworthy work remains after a teacher has made that answer safe, accurate and educationally sound.

Before schools call AI a workload solution, they should keep the receipt.

Gleb Tsipursky, PhD, a behavioural scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook