The arrival of capable generative AI has forced a reckoning that assessment design needed anyway. For years many university assessments quietly measured the wrong things — the ability to summarise a textbook, reproduce a standard essay, or recall facts under exam conditions — because those tasks were easy to set and mark. When a free tool can now produce a passable essay in seconds, that kind of assessment no longer measures learning; it measures access to software.
The good news is that the response to AI is the same as the response to good pedagogy: design assessments that measure genuine understanding, application and judgement. This guide covers how to build assessment that actually tests learning — aligning tasks to outcomes, designing authentic and AI-resilient work, writing rubrics that make marking fair and fast, giving feedback that improves future work, and handling integrity in a way that goes beyond a futile detection arms race. For related material, see our Teaching & Assessment hub.
Constructive alignment: start from outcomes
Good assessment begins not with the task but with the learning outcomes. The principle of constructive alignment — associated with the educationalist John Biggs — is that intended outcomes, teaching activities and assessment must all point at the same thing. If your outcome says students will be able to critically evaluate competing theories, then a multiple-choice test cannot assess it, no matter how convenient. The assessment must require the student to actually perform the verb in the outcome.
Work backwards: for each learning outcome, ask what evidence would genuinely demonstrate a student has achieved it, and design the task to elicit exactly that evidence. This discipline eliminates the busy-work assessments that measure effort rather than learning, and it naturally produces tasks that are harder for AI to complete convincingly, because real application to a specific context is exactly what generic tools do poorly. The Higher Education Academy's principles, now stewarded by Advance HE, remain a useful reference — see advance-he.ac.uk.
Designing authentic, AI-resilient tasks
Authentic assessment asks students to do something that resembles real professional or scholarly practice, applied to a specific, current or local context. Instead of a generic essay on a topic, ask students to analyse a particular case, dataset, community problem or recent event; instead of describing a method, ask them to apply it and justify their choices. Authentic tasks are more motivating, more valid, and far harder to outsource to a generic tool because the specificity resists off-the-shelf answers.
You can also make the process of learning part of the assessment — drafts, reflective logs, oral defences, or presentations where students explain and defend their reasoning in real time. These make it clear whether a student genuinely owns their work. The aim is not to trap students but to design tasks where doing the real thinking is the easiest path to a good grade. Our guide on integrating teaching and research explores using your own research context to create authentic tasks.
Writing rubrics that mark themselves
A good rubric is the single most powerful tool for fair, fast and consistent marking. It breaks the assessment into explicit criteria — for example argument, use of evidence, analysis, structure and referencing — and describes what each performance level looks like for each criterion. This makes grading transparent to students, defensible when challenged, and dramatically quicker because you are matching work to descriptions rather than agonising over each mark from scratch.
Share the rubric with students before they begin, so it doubles as a teaching tool that tells them exactly what good work looks like. Where a course is taught by several markers, a shared rubric plus a short calibration exercise on a few sample scripts keeps grading consistent across the team. Write rubrics in specific, observable language — avoid vague terms like good or excellent without saying what distinguishes them — so two markers reach the same judgement independently.
Feedback that changes future work
Feedback only has value if it can influence future performance, yet much feedback arrives too late, is too vague, or comes attached to a final grade the student never revisits. Prioritise feedback that is specific and actionable — pointing to what to do differently next time rather than only what went wrong — and, where possible, deliver it at a point where students can still use it, such as on a draft or an early assessment in a sequence.
You do not have to write more to give better feedback. Structured rubric comments, brief audio notes, and common-error summaries shared with the whole class can be more useful and far more efficient than lengthy individual annotations. The AcademicStaff assessment toolkit includes rubric templates and a feedback-bank feature that lets you reuse and personalise recurring comments, cutting marking time while improving consistency. For the strategy behind sustainable teaching effort, see our teaching and assessment resources.
Academic integrity beyond detection
Trying to win an arms race against AI through detection software is a losing strategy; detectors are unreliable, produce false positives that unfairly harm students, and are quickly outpaced. A more durable approach rests on three pillars: design (assessments that require genuine, contextual thinking, as above), transparency (clear, discipline-appropriate policies on what use of AI is and is not permitted), and education (teaching students why integrity matters and how to use tools ethically).
Be explicit in each assessment about whether and how AI may be used, since a blanket ban is often unrealistic and inconsistent across a programme. Where you suspect misconduct, follow your institution's formal academic-integrity procedures and rely on evidence and process rather than a detector's probability score. Framing integrity as a professional value students are learning, not merely a rule they might break, produces far better long-term behaviour.
Managing your own marking workload
Well-designed assessment protects your time as well as your students' learning. Rubrics, feedback banks and calibrated marking all reduce the hours each grading cycle consumes. Beyond that, be strategic about how much you assess: over-assessment burdens both students and staff without improving learning, so a smaller number of well-designed, well-spaced assessments usually beats a crowded schedule of minor ones.
Batch your marking into dedicated sessions rather than fragmenting it, and reuse strong assessment designs across cohorts with refinements rather than rebuilding each year. Treat assessment design as an investment: the effort you put into a clear task and rubric once pays back every time you run the course. For protecting time more broadly across teaching, research and service, our teaching and assessment hub and the AcademicStaff workload tools work together.
The AI-Era Assessment Design Kit
A downloadable kit with an authentic-task design worksheet, three ready-to-adapt analytic rubric templates, a feedback comment bank starter, and a sample AI-use policy statement for your assessments.
Want the full article?
Enter your email for free access to the rest of this guide and our TEQSA resource library.
Frequently asked questions
How do I stop students using AI to cheat on assessments?
Rely on design rather than detection. Set authentic, context-specific tasks that require genuine thinking, be explicit about permitted AI use, and incorporate process elements such as drafts or oral defences. Detection software is unreliable and produces false positives, so it should never be the primary safeguard.
What makes a good assessment rubric?
A good rubric breaks the task into explicit criteria and describes what each performance level looks like for each criterion in specific, observable language. Share it with students beforehand so it also teaches them what good work looks like, and use it to keep marking consistent across multiple markers.
How can I give better feedback without spending more time?
Use structured rubric comments, brief audio notes, and whole-class common-error summaries instead of lengthy individual annotations. Deliver feedback while students can still act on it, such as on drafts, and reuse a bank of recurring comments personalised to each student.
The AcademicStaff editorial team writes practical, evidence-based guidance for university staff — drawing on sector reporting, funder guidelines and the lived administrative reality of academic work. Every guide is reviewed for accuracy against current Australian higher-education practice.