TerbitTerbit Zero AIPractical AI writing tools

Responsible AI guide

How Teachers Can Review Suspected AI Use

Use evidence, student discussion and policy—not a detector score alone—to reach fair conclusions.

By Terbit Zero AI Editorial TeamReviewed 2026-10-16Practical guidance—not a determination

Separate concern from conclusion

A sudden change in style, an invented citation or a detector alert may justify closer review. None proves cheating by itself. Treat the initial concern as a question: does the submitted work reflect the student’s own learning under the course rules?

This neutral framing reduces confirmation bias and allows evidence that supports or contradicts the concern to be considered fairly.

Review the work before confronting the student

Read the assignment closely. Identify specific passages that appear inconsistent, unsupported or unrelated to class material. Check references and quotations. Compare with earlier work cautiously: students improve, receive editing help and write differently across formats.

If a detector is used, preserve the complete output, limitations, version and date. Do not present a percentage as a finding of guilt.

  • Check whether the assignment instructions permitted any AI assistance.
  • Look for drafts, outlines, notes and document history.
  • Verify sources rather than relying on writing style alone.
  • Consider language-learning and accessibility factors.
  • Apply the same process consistently across students.

Use a learning-focused conversation

Ask the student to explain the thesis, evidence and revision process. Questions tied to the actual submission are more informative than accusations. A student who understands the work may be able to explain why a phrase was chosen, show research notes or recreate a key step.

The conversation may reveal misconduct, permitted assistance, poor citation practice or a misunderstanding of the rules. Each requires a different educational response.

Document a proportionate decision

Base the decision on the published policy and combined evidence. Explain which facts were accepted, what remains uncertain and how the outcome follows the institution’s process. Provide the student with the relevant review or appeal route.

Longer term, design assessments that make learning visible: staged drafts, oral explanation, source annotations, personal application and in-class checkpoints. Good assessment design reduces dependence on detection technology.

Terbit editorial analysis

Start with the learning objective. If the assignment measures reasoning, ask the student to explain choices and sources. If it measures language production, review drafts and revision history. Evidence should match what the task was designed to assess.

Avoid public confrontation. Preserve the submission, apply the written policy and give the student a private opportunity to respond before any conclusion.

How this guide was reviewed

  • Checked against current primary or official guidance listed below.
  • Separated provider claims from Terbit Zero AI analysis and practical recommendations.
  • Used a transparent scenario to show the reasoning process; examples are illustrative, not benchmark results.

Read our full Editorial Standards and Review Methodology.

Evidence ladder example

A detector flags an essay, but the student provides an outline and version history.

  1. Check whether the drafts explain the development of the final argument.
  2. Discuss any unexplained passages with the student.
  3. Apply the published course policy and document evidence for and against a breach.

Practical checklist

  • Identify specific concerns in the submitted work.
  • Collect process and source evidence.
  • Meet the student without presuming guilt.
  • Document the outcome and appeal process.

Sources and further reading