Generative AI has made one uncomfortable assessment question impossible to avoid: when a polished response arrives on a teacher’s desk, what evidence remains that the student did the thinking?
The weakest response is to turn every task into a surveillance exercise. The other weak response is to assume that because AI is now widely available, the finished product no longer needs to demonstrate learning. Neither position gives teachers what they actually need: credible evidence of what a student understands, alongside realistic preparation for a world in which AI tools will be common.
The more useful starting point is to redesign the evidence, not simply police the tool.
Begin with the learning claim
Before changing an assessment, write one sentence that completes this prompt: “A successful student must be able to…”
If the sentence says “produce a 1,500-word report”, it describes a product. If it says “compare competing explanations, select relevant evidence and defend a conclusion”, it describes learning. Generative AI can produce the first. The assessment must make the second visible.
This distinction matters because the Australian Framework for Generative AI in Schools supports responsible and ethical use that benefits learning. The objective is not to pretend the technology does not exist. It is to ensure that its use does not obscure whether learning has occurred.
Use three layers of evidence
A strong AI-era assessment usually gathers evidence at three points rather than relying on one final submission.
1. Evidence of preparation
Ask students to submit a small piece of thinking before substantial drafting begins. This might be an annotated source, a concept map, a proposed method, a worked calculation or a short explanation of the approach they intend to take.
The checkpoint does not need to create a large marking burden. Its purpose is to establish what the student notices, knows and plans before the final product takes shape.
2. Evidence in the finished work
Keep the final task, but sharpen the parts that require judgement. A generic essay prompt is easy to outsource. A task that asks students to apply a concept to a local case, compare two plausible responses, explain the limitations of their evidence or justify a decision is more revealing.
AI may still assist with the work. That is not automatically a problem. The important question is whether the student must make consequential choices that can be examined.
3. Evidence of ownership
Add a brief follow-up in which the student explains one decision, responds to a new example or revises a section after feedback. This can be a two-minute conversation, a short in-class paragraph, an audio reflection or an annotated change log.
The aim is not to catch students out. It is to confirm that they can retrieve, explain and use the knowledge represented in the submission.
State the AI conditions clearly
“Use AI appropriately” is too vague to guide students or teachers. Each task should state one of three conditions:
- AI-free: the task is completed without generative AI because independent performance is the evidence being assessed.
- AI-assisted: specified uses are permitted, such as brainstorming questions, receiving feedback or checking clarity, but the student remains responsible for decisions and accuracy.
- AI-integrated: using and evaluating an AI tool is part of the learning task.
Where AI is permitted, ask for a short disclosure: the tool used, the purpose for which it was used, and what the student changed or rejected. A disclosure should be proportionate. It is there to make the process visible, not to turn a Year 8 task into a compliance report.
Do not make every task “AI-proof”
Some knowledge and skills still need to be demonstrated under controlled conditions. Students need opportunities to write, calculate, recall, perform and reason without assistance. Those tasks can sit alongside richer assignments in which AI use is taught and evaluated.
A balanced assessment program therefore includes both secure demonstrations of individual capability and authentic tasks that reflect contemporary practice. This is consistent with the broader direction of assessment reform work published by TEQSA, which focuses on assurance of learning while preparing students to use AI responsibly.
A 15-minute assessment audit
Take one existing task and ask:
- What learning must the student personally demonstrate?
- Which part of the task could AI complete without that learning?
- What small checkpoint would reveal the student’s thinking?
- What follow-up would confirm ownership?
- What AI use is prohibited, permitted or required?
- How will those conditions be explained to students?
There is no need to rebuild an entire assessment calendar at once. One task redesigned well is more useful than a sweeping policy that leaves classroom practice unchanged.
The practical shift
Generative AI has not removed the need for assessment. It has exposed how often assessment relied on a finished product as a proxy for learning.
The practical response is to collect better evidence: a trace of preparation, a task that requires judgement and a short demonstration of ownership. That approach protects the integrity of assessment without abandoning the equally important work of teaching students how to use new tools well.
Latest posts by John Bigelow (see all)
- When Students Turn to AI Companions: A School Conversation Guide - October 5, 2026
- A 30-Minute Cyber Incident Exercise Every School Can Run - September 18, 2026
- A Practical Digital Accessibility Check for Learning Materials - September 17, 2026
You must be logged in to post a comment.
There are no comments
Add yours