Responsible-use guide
How to review suspected AI-assisted coursework fairly
A proportionate review combines a clear policy, a suitable detector report, writing-process evidence, subject knowledge and a genuine opportunity for the student to respond.
Before a concern arises: make the rule reviewable
A fair investigation depends on expectations that existed when the work was completed. “Do not use AI” can be too vague when students encounter predictive text, grammar tools, translation, search summaries, coding assistants and generative chat in the same workflow.
- Describe permitted, disclosure-required and prohibited uses for the specific assessment
- Explain whether brainstorming, outlining, translation, editing and generated drafting are treated differently
- Give a simple disclosure format and examples of an adequate statement
- Tell students what process evidence, if any, they are expected to retain
- Publish how concerns will be raised, reviewed and appealed
- Ensure any submitted work is processed only through services approved by the institution
Triage the detector result before contacting the student
- 01
Preserve the exact submission
Keep the original file, submission time, assessment instructions and policy. Do not repeatedly edit and rescan different extracts until a concerning score appears. - 02
Check the input quality
Confirm the sample is English continuous prose, meets the minimum length and has not been distorted by PDF extraction, references, page furniture or a template. - 03
Read the complete report
Inspect sentence variation and source candidates where available. Record trimming or missing sentence details rather than quoting only the overall number. - 04
Identify a policy-relevant concern
Translate the signal into a question that can be investigated, such as undisclosed generated drafting where original composition was required. - 05
List alternative explanations
Consider translation, accessibility tools, taught structure, editing, feedback, technical convention and extraction errors before seeking only confirming evidence.
Review process and subject-matter evidence
Process evidence can connect the finished document with how the student developed it. Subject-matter evidence can test whether the student understands the reasoning and sources. Neither should become a rigid checklist that assumes every genuine writer works in the same way.
| Evidence | Question it may help answer | Caution |
|---|---|---|
| Outline and notes | Did the argument and evidence develop coherently? | Some students plan mentally or across several tools |
| Draft and version history | Are there traceable stages of composition and revision? | Offline work, copied files and privacy settings can leave gaps |
| Sources and annotations | Can the student locate and explain the evidence used? | A source list alone does not prove how prose was produced |
| Earlier comparable work | Is the style or subject understanding materially different? | Writers improve, receive feedback and vary by task |
| Factual and citation problems | Are there invented, mismatched or misunderstood claims? | Human writers also make factual and citation errors |
| Detector report | Which passages triggered the current model? | The report is probabilistic and can be wrong |
Do not treat missing drafts as a confession. Ask what records the assessment expected and what alternative evidence can answer the same question.
Hold a fair conversation about the work
Give the student enough information to understand the concern. Begin with the work and process rather than asking them to disprove an unexplained percentage.
Useful neutral questions
- How did you move from the assignment question to this thesis or approach?
- Which source most changed your argument, and why?
- Can you explain this passage or calculation in another way?
- What feedback or tools did you use at each stage?
- How did this section change between drafts?
- Is there anything about your writing process that would help explain the report?
Questions to avoid
Avoid asking “Why did you cheat?” or “How did you beat the detector?” before a finding exists. Do not use speed, nervousness, accent or an unfamiliar communication style as a proxy for authorship.
Reach and record a policy-based human decision
- Apply the institution's stated standard of evidence and normal procedure
- Distinguish a detector classification from a finding about the student's conduct
- Explain which evidence was persuasive and why
- Address credible alternative explanations rather than omitting them
- Use a proportionate outcome tied to the established rule
- Provide the student with reasons and the available review or appeal route
- Retain only the records needed under the institution's data policy
If a score is the only concerning item, the next action should usually be further review, not an adverse finding. Read AI detection as evidence and the false-positive guide before assigning it more weight.