A teacher usually does not need a faster accusation. They need a review process that is consistent, explainable, and fair to the student. An AI detector can help identify passages that deserve attention, but the score should never be the whole case.
The process below is designed for essays and other substantial writing. Short answers, quotations, reference lists, and highly formulaic tasks often provide too little useful evidence for a confident detector result.
1. Start with the assignment policy
Define what help was allowed before evaluating the work. A policy might permit spelling correction, translation, brainstorming, or feedback while prohibiting generated final prose. If the boundary was not clear to students, do not invent a stricter rule after submission.
Separate authorship from tool use. A student may have used an allowed tool without surrendering authorship, or may have submitted generated text that contains no copied source material. The review must address the actual rule for that assignment.
2. Preserve the process evidence
Collect only the records your institution permits you to review. Useful evidence can include an outline, notes, source annotations, version history, teacher feedback, and earlier drafts. These materials often answer the authorship question more directly than a detector score.
Do not require access to a private personal account or device unless an established policy and appropriate authority allow it. Student privacy still matters during an integrity review.
3. Read the essay before running a detector
Look for issues that require ordinary academic review:
- claims that do not match the cited source;
- references that do not exist or cannot be located;
- abrupt changes in vocabulary, tone, or command of the subject;
- confident statements with no evidence;
- an argument the conclusion does not follow.
None of these proves AI use. Human writers make mistakes, and generated text can be carefully edited. The point is to identify questions for the next stage.
4. Prepare the text carefully
Use only the student's prose. Remove the assignment prompt, teacher-provided templates, long quotations, bibliography, and copied reference material when those sections are not being evaluated for authorship. Keep a copy of exactly what you submitted to the detector.
Before uploading student work, confirm that the tool and your institution's policy permit the submission. Consider data retention, access, and whether personal information should be removed.
5. Treat the result as a lead
Review document-level and sentence-level signals together. A high overall score may be driven by a small repeated section. A mixed result may show that only part of the document deserves closer reading. Do not convert the score into a claim that the student cheated.
If you use more than one detector, use the same prepared text and record what each score means. Different systems can disagree because they use different data, features, and thresholds.
6. Check sources and drafts
Verify a sample of quotations, citations, and factual claims. Compare the final essay with available drafts and feedback. Look for a plausible development of the argument, including revisions that respond to the assignment.
A clean process record can outweigh a detector flag. A missing process record is not proof by itself, especially when drafts were not required.
7. Speak with the student
Use neutral, specific questions. Ask the student to explain the thesis, describe how a source supports a paragraph, or walk through a revision choice. Give them a real opportunity to provide context and evidence.
A meeting should not be a surprise test of memory. Stress, disability, language background, and time since submission can affect how someone responds. Follow the institution's normal procedure and document both the evidence and the student's explanation.
8. Make the decision under an established process
A detector result should not carry a case on its own. Use the same standard of evidence for every student, provide an appeal path, and record why the available evidence supports the outcome.
This human-centered approach matches current education guidance. UNESCO emphasizes human agency and equitable policy. The U.S. Department of Education's education-leader toolkit emphasizes transparency, privacy, safety, and non-discrimination when schools adopt AI systems.
You can use the Detecting AI workbench to inspect sentence-level signals. Use those signals to guide a review, not to replace one.