Teacher investigation guide
How teachers can check suspected AI work fairly
A detector flag is a prompt for proportionate review, not proof of misconduct. Start with the rules, preserve the work and gather evidence the student can understand and answer.
Start with the applicable policy
Establish what assistance was allowed for this assignment, course and date. Brainstorming, translation, grammar tools, tutoring and generative drafting may be treated differently. A fair finding needs a clear rule that the student could reasonably understand.
Preserve the original evidence
Keep the submitted file, timestamp, brief, declared sources and the exact version of any report. Record the detector name, version, settings and limitations. Do not repeatedly upload student work to third-party services without checking institutional privacy, retention and consent requirements.
Check sources and subject knowledge
Verify citations, quotations and factual claims. Look for references that do not exist, do not support the sentence, or concern a different topic. These are assessable problems whether the cause is AI, misunderstanding or poor research.
Look for process evidence
Review drafts, notes, version history and earlier comparable work where policy permits. Ask the student to explain the argument, define key terms, locate a source and describe how a paragraph changed. The aim is to understand the process, not to stage a surprise test.
Keep contrary evidence and an appeal route
Record evidence that supports the student as well as evidence that raises concern. Consider disability, assistive technology, language background and authorised editing. Follow the institution’s normal decision-maker, standard of proof and appeal procedure.
Related reading
Try an evidence-led review
Check claims and public wording separately
Paste a short, anonymised passage. The result does not establish authorship, intent or misconduct.