Generative AI has not created a new problem so much as exposed an old one: a great deal of what we have historically assessed was memorisation and information retrieval, both of which a language model now does in seconds. For educators from Nairobi to Newcastle, the honest response is not an arms race of detection tools but a redesign of what and how we assess.
This article sets out a discipline-agnostic framework for authentic assessment — tasks that measure the reasoning, judgement, and applied skill we actually want graduates to have. The approach works whether you teach 15 postgraduates or lecture to 800 first-years, and it does not depend on any particular technology being banned or permitted.
Why Traditional Assessment Is Breaking
The classic unseen essay and the take-home report were always imperfect proxies for learning, but they held together because producing them required a student to engage with the material. When a model can generate a competent draft on almost any topic, assessments that reward polished prose about general knowledge stop discriminating between students who learned and those who did not. Detection software is unreliable, produces false positives that fall hardest on non-native English speakers, and traps staff in adversarial relationships with their classes.
The productive move is to shift from assessing the artefact to assessing the process and the person's reasoning within it. This does not mean abandoning writing; it means designing writing tasks that require the student's specific context, data, choices, and defence. You will find related teaching strategies across our Teaching & Assessment collection, and it is worth reviewing your institution's academic-integrity policy alongside sector guidance such as that published by TEQSA on academic integrity, whose principles translate well across national systems.
The Principles of Authentic Assessment
Authentic assessment asks students to do something a competent professional in the field would actually do, under conditions that resemble real practice. Four principles make a task authentic: it is situated in a genuine context; it requires judgement under uncertainty rather than a single correct answer; it produces something with a real audience or use; and it makes the student's reasoning visible, not just their conclusion.
A useful test is to ask whether a stranger could complete the task with no knowledge of your specific class. If they could, the task is measuring general capability that AI now supplies cheaply. If completing it well requires the student to draw on a dataset you provided, a placement they undertook, a debate held in your seminar, or a personal design decision they must defend, then you are assessing learning that belongs to that student. This is also how you connect coursework to the workplace skills employers value — a theme explored in our guide to embedding employability.
Designing Tasks AI Cannot Complete for the Student
The most robust tasks bind the assessment to something only that student possesses. Ask for analysis of primary data the student collected or that you distributed uniquely; require reflection on a specific placement, fieldwork, or lab session; or build in an oral component where students defend their written submission in a five-minute viva. A short authentic conversation reveals depth of understanding faster than any plagiarism scan.
Staged assessment also raises the cost of shortcutting. Rather than one final submission, collect a proposal, an annotated bibliography or data plan, a draft, and a reflective account of what changed and why. Each stage carries modest weight, receives light feedback, and creates a paper trail of genuine engagement. You can permit AI use openly at some stages and require students to document their prompts and their edits, turning the tool into an object of critical study rather than a forbidden temptation.
Writing Rubrics That Reward Reasoning
A rubric built for authentic assessment weights the things machines do poorly and students must own: the quality of judgement, the justification of choices, the handling of ambiguity, and the connection to the specific context. De-emphasise criteria that reward surface fluency, since fluency is now cheap. Make at least one criterion about the defensibility of decisions — why this method, this source, this interpretation.
Share the rubric before students begin and, where possible, run a calibration exercise with a sample answer so expectations are transparent. Clear criteria reduce disputes, speed up marking, and make feedback more actionable. Consistent rubric design across a programme also helps teaching teams mark reliably at scale, a persistent challenge covered in more depth in our Teaching & Assessment resources.
Building Integrity Into the Task, Not Just Policing It
Academic integrity is far more effectively designed in than enforced after the fact. Explain to students why a task is structured as it is and what skill it builds toward their career; students who see the purpose cheat less. Provide a clear, non-punitive channel for questions about acceptable AI use, and state your rules for each assessment explicitly rather than assuming a blanket policy is understood.
Where integrity breaches do occur, follow due process and lean on your institution's stated procedures; guidance from national quality bodies and the broader scholarly-integrity infrastructure increasingly frames integrity as a shared responsibility rather than a policing exercise. The workload of monitoring and documenting cases is real — AcademicStaff's workload calculator helps you account for assessment design, marking, and integrity casework when negotiating teaching allocations, so the hidden labour of good assessment is visible rather than absorbed silently.
Managing the Feedback Workload
Authentic, staged assessment can generate more marking if you are not deliberate about it. Protect yourself by making early stages formative and lightweight — a checklist, a single comment, or peer feedback against your rubric rather than full written critique. Reserve your detailed effort for the stage where it changes the final outcome most.
Reuse is your ally. Maintain a bank of high-quality feedback comments keyed to your rubric criteria, and personalise them rather than writing every comment from scratch. Group common issues into a single whole-class debrief instead of repeating the same note across dozens of scripts. These practices keep authentic assessment sustainable across large cohorts, so that the redesign improves learning without quietly consuming every evening you have.
The Authentic Assessment Design Kit (Free Checklist + Rubric Template)
A one-page task-design checklist plus an editable reasoning-focused rubric template you can adapt to any discipline, built to reward genuine learning and reduce AI shortcuts.
Want the full article?
Enter your email for free access to the rest of this guide and our TEQSA resource library.
Frequently asked questions
Should I ban generative AI in my assessments?
Blanket bans are hard to enforce and often unfair to students who use assistive tools legitimately. A more durable approach is to design tasks that require the student's own context and reasoning, and to state clearly for each assessment where AI use is permitted, required to be documented, or excluded.
Is AI-detection software reliable enough to base decisions on?
No. Current detectors produce significant false positives and negatives, and they disadvantage non-native English writers in particular. Use them, at most, as one weak signal among many, never as sole evidence in an integrity case.
Does authentic assessment always mean more marking?
Not if it is designed well. Make early stages formative and lightweight, use a comment bank keyed to your rubric, and run whole-class debriefs for common issues. Concentrate detailed feedback where it most changes the outcome.
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.