Teachers check for AI by combining several signals: an AI detector score, the document's version history, how the writing compares with the student's earlier work, and a short conversation about the content. No single signal is proof. Detectors in particular make mistakes, so a score should start a conversation, not end one.
Most schools end up using a mix, because each method has a weakness that another covers.
AI detectors are statistical guesses. They can flag human writing, and research on this has raised particular concern about students who write in a second language, whose clear and simple style may be misread. Detectors can also miss text that has been edited after generation.
For that reason many schools say in policy that a detector result alone cannot support a penalty. If your school has no written position, ask for one before relying on any tool.
Start by looking at the work itself: checkable claims, invented citations and details that do not match the class. Then look at version history. Then talk to the student, privately and without accusation, and ask them to walk you through how they wrote it.
Keep notes of what you saw and said. If a formal process follows, those notes matter more than a percentage from a tool.
The most durable fix is in the assignment. Ask for drafts and reflections along the way. Connect prompts to class discussion or local events. Add a short oral check on major pieces. State clearly in the syllabus when AI is allowed, for what, and how it should be cited, because students are more honest when the rule is clear.
Teachers should not have to invent policy alone. We help schools think through workflows where AI is used under clear rules, such as drafting feedback that the teacher reviews. We do not sell a detector, and we would not advise basing decisions on one. If your team wants a plain-language approach to AI use in classrooms, the k-12 guide is a good starting point.