Teachers now face a new question when grading: did the student write this, or did a chatbot? An AI detector can help you decide which submissions deserve a closer look, but only if you treat it as one piece of information among several. This page describes a fair workflow for checking assignments, reading the result and talking to students without turning a probability into an accusation.
Start with a clear policy
Before you check anything, students need to know what is allowed. State in the assignment whether AI tools may be used for brainstorming, for language correction or not at all, and how permitted use should be acknowledged. A detector result means little if the rules were never explained.
How to check an assignment
- Remove the student's name and other personal details before you upload the text, and work with anonymised copies where possible.
- Paste the text (at least 250 characters) or upload the .docx, .pdf, .odt or .txt file the student submitted.
- Leave out quotations, the reference list and any template text you provided, so that only the student's own writing is analysed.
- Run the check and read the verdict together with the highlighted fragments.
- Download the PDF report if you need to keep a record.
How to read the result
The verdict ranges from AI generation detected, through an uncertain result and some suspicious fragments, to no significant AI signs. Each fragment is marked as likely AI, suspicious or human, with a confidence value. Look at where the highlighted fragments are: a few flagged sentences in a definition or in a summary of a source are common in honest work, while an entire argument marked as likely AI is a reason to look further.
Be aware of false positives
Some students are more likely to be flagged even when they wrote every word themselves:
- students writing in a second language, whose sentences can be simpler and more regular;
- students following a rigid structure, as in lab reports, summaries or five-paragraph essays;
- students who used grammar or translation tools that smooth the wording;
- authors of very short texts, which give the detector little to analyse.
The detector estimates probabilities and can be wrong in both directions. It can also miss generated text that a student has rewritten. Never grade or sanction on the basis of the result alone.
What to do when a result worries you
- Compare with earlier work. A sudden change in vocabulary, structure or quality compared with writing done in class says more than any single check.
- Talk to the student. Ask them to explain their argument, the sources they used and how they wrote the text. Students who did the work can usually talk about it in detail.
- Ask about the process. Notes, outlines, drafts and the version history of a document show how a text developed.
- Follow your institution's procedure if serious doubts remain, and share the report as supporting information, not as proof.
Design assignments that are harder to outsource
Tasks connected to class discussions, personal experience, local data or a specific reading are harder to complete with a chatbot. Short drafts written in class, staged submissions and brief oral explanations also give you a sense of each student's voice to compare against. The detector is free without registration within a daily limit, and signing in or a PRO plan gives larger limits for checking a whole class.