Purpose
This guide supports faculty in adapting rubrics and marking practice for assessments where students may use AI tools, including grammar checkers and generative AI, where permitted by the relevant course or School policy. It assumes familiarity with the fundamentals of rubric design, covered in the companion guide, Rubrics.
Start with the assessment policy
Before designing the rubric, establish the intended approach to generative AI for the specific assessment. Course guidance should specify permitted, prohibited and required uses rather than relying on a vague statement such as "AI may be used responsibly." Clear expectations, transparent disclosure and alignment with learning outcomes are central principles of AI-aware assessment design.
| Approach | Meaning | Rubric implication |
|---|---|---|
| AI not permitted | Students must complete the work without generative AI, except explicitly approved accessibility tools | State the restriction clearly; assess breaches through academic-integrity procedures, not through speculative writing-style judgements |
| AI permitted for limited support | Examples: grammar support, brainstorming, translation, formatting or feedback | State permitted uses and disclosure expectations; assess final academic quality against the learning outcomes |
| AI permitted or required as part of the task | Students may use AI to generate, compare, critique or improve material | Include a criterion that assesses critical AI literacy, verification, attribution and judgement in using the tool |
Avoid duplicating AI-related criteria
Where a criterion assessing AI use is included, keep it to one place in the rubric rather than folding it into every criterion.
| Avoid duplication | Better allocation |
|---|---|
| Assessing "AI use" under every criterion | Assess disclosure and academic integrity under scholarly practices; assess only the quality of the final communication under writing and organisation |
Do not use AI detection as a marking criterion
A marker should not award or reduce marks because prose "sounds like AI," because it is unusually polished, or because an AI-detection tool produces a score.
These are not reliable indicators of authorship.
Independent research has found that widely used AI-detection tools misclassify human-written text at meaningful rates, and that this bias falls disproportionately on multilingual and non-native English writers: a 2023 Stanford study found seven commonly used GPT detectors flagged non-native English writing samples as AI-generated in roughly 61% of cases, compared to near-perfect accuracy on native-speaker writing (Liang et al., 2023, Patterns). Similar concerns have been raised for neurodivergent students and students who write in a highly formal or formulaic style.
Address concerns about potential misconduct through the institution's academic-integrity procedures and evidence requirements, not by embedding suspicion into ordinary academic-quality descriptors. The rubric should assess what is demonstrated in the work: accuracy, reasoning, evidence use, disciplinary understanding, methodological judgement and effective communication.
Distinguish product quality from integrity
For a final essay where faculty see only the submitted version, markers can assess:
- Whether the argument is coherent and sustained
- Whether evidence is used accurately and critically
- Whether the prose is clear, precise and appropriate for its audience
- Whether the work is generic or specifically responsive to the topic
- Whether sources, quotations and AI use are appropriately disclosed
Markers cannot fairly assess, without additional evidence:
- Whether a student drafted multiple versions
- How much time was spent writing
- Whether each sentence originated with the student or a tool
- Whether the student accepted or rejected AI suggestions during drafting
Where the institution requires evidence of process or authorship, it should be an explicit part of the assessment design, for example an AI-use statement, annotated bibliography, oral discussion, source-verification task, staged submission, or reflective commentary.
Build disclosure into scholarly practice
For permitted AI use, students should be asked to disclose material use of generative AI. A concise declaration might include:
- Tool used: for example, ChatGPT, Claude, Grammarly or Microsoft Copilot.
- Purpose: for example, brainstorming, language editing, outlining, translation or feedback.
- Scope: which part(s) of the work were affected.
- Student oversight: how outputs were checked, adapted, verified or rejected.
A practical model is: content affected + action taken + tool used + purpose + human oversight.
Sample student declaration
I used Grammarly to identify grammar, punctuation and clarity issues in the final draft. I also used [tool name] to generate possible ways of grouping themes in my outline. I did not use AI-generated references or evidence. I selected, read and cited all sources myself, and I developed, checked and revised the final argument and wording.
The Department/School should adapt this example to its own AI policy. A declaration is not a substitute for academic integrity; it is a tool for transparency.
Feedback on AI disclosures
When an AI declaration is incomplete or unclear, give the same specific, actionable feedback you would give on any other part of the submission:
| Less helpful feedback | More helpful feedback |
|---|---|
| “Too much AI.” | The AI declaration does not explain how generated material was checked or adapted. Revise future disclosures to identify the tool, purpose, scope of use and your own verification or editorial decisions. |
Marker guide: writing and AI
For a final written submission, use the following distinction.
| Markers may assess | Markers should not infer |
|---|---|
| Coherence, structure and progression of argument | Whether the student made multiple drafts |
| Clarity, precision and appropriateness of academic language | Whether wording was generated by AI solely because it is polished or formulaic |
| Whether the response is specific to the task, evidence and disciplinary context | The student's effort, time spent or private drafting process |
| Whether claims, citations and AI disclosures meet academic standards | AI use or misconduct from writing style alone |
| Independent scholarly contribution, where supported by the evidence required in the task | The precise proportion of text written with or without a tool |
Applying this in marking
The general marking workflow (before, during and after marking) is covered in the companion guide, Rubrics. For AI-aware assessments specifically, before marking begins:
- Confirm the permitted and prohibited uses of AI for that assessment.
- Confirm whether an AI-use declaration was required, and if so, review it before or alongside the submission.
- Decide, with the programme team, how to handle work that raises academic-integrity concerns; this sits outside the rubric and outside ordinary marking.
- Where AI critical literacy is itself an assessed criterion, review the agreed descriptors for that criterion before marking begins.
References
- Liang, W. et al. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7).
- Frontiers in Education: rubrics and assessment in the age of generative AI
- TEQSA (Australia): Principles, Criteria and Standards for Assessment in the Age of Generative AI
- Taylor & Francis: assessment design and generative AI in higher education
- Taylor & Francis: academic integrity and generative AI