How teachers check work for AI: Turnitin, other detectors, drafts and conversations
Short answer
Teachers rarely rely on a detector score alone. They look at the work itself, compare it with your earlier writing, check sources, ask about your drafts and version history, and above all talk to you about the work. The best protection is a visible process and work you can explain.
Detector reports
Where a university uses an AI detector, the report highlights passages and gives a score. Turnitin's own guidance tells instructors there is no target score, to read it alongside the institution's policy and the rubric, and to use it to initiate a conversation rather than draw a conclusion. Not every university uses one: UCL says it is not investing in detection software. See AI essay checkers explained.
Draft and version history
Documents written over days, with notes, outlines and revisions, look very different from text that appears complete in one paste. Markers may ask to see earlier drafts or the version history in your word processor or cloud document. For how version history works as evidence, see the guides at 21learn.org.
Comparing with earlier work
Teachers who know your writing notice sudden changes in style, vocabulary or quality. They also check the details: references that do not exist, quotations not found in the source, facts that are subtly wrong. Walters and Wilder suggest that instructors examine the works cited in papers, since fabricated references are a distinctive feature of generated text.
The conversation
The most common next step is simply a conversation. Turnitin advises instructors with doubts to ask the student to explain what the essay is about, the process they followed, their ideas and key takeaways. Some universities formalise this: UCL's process can include an investigatory viva, which assesses on the balance of probabilities whether there is evidence that the work, or parts of it, were not authored by the student.
If you wrote the work, this is where it shows. You can explain why you made each argument and chose each source, because you made those decisions.
If you are asked to discuss your work
Being asked to talk about your work is not an accusation in itself. Prepare the way you would for any meeting about an assignment:
- Re-read your paper and your main sources.
- Be ready to explain your thesis, why you structured the argument as you did, and why you chose each source.
- Bring your notes, drafts and version history.
- If you used AI in an allowed way, bring your record and acknowledgement.
- Ask what the process is, what happens next, and whether you can bring a supporter.
Your students' union or university advice service can explain the procedure and your rights. Answer honestly. If you made a mistake with a rule, saying so early is usually better than defending something you cannot explain.
What you can show
- Notes, outlines and drafts with dates.
- Version history from the document you wrote in.
- Your reading log or annotated bibliography.
- A record of any AI use your course allowed, and your acknowledgement.
For accuracy research on detectors, see afip.org and ai-detector.co. For the rules that apply to you, read how to read your university's AI policy and is using AI plagiarism?, and before you hand in, run the pre-submission checklist.
More on this in Integrity.
FAQ
Can teachers tell if I used ChatGPT?
Not with certainty from a score. They look at the work, your sources, your earlier writing and your drafts, and they ask you about it.
What is an investigatory viva?
A meeting, used by some universities such as UCL, to assess whether a piece of work was authored by the student, usually by discussing it with them.
Will my version history help me?
Yes. A dated history of notes and drafts is strong evidence that the work developed over time.
What if I used AI in an allowed way?
Show your acknowledgement and your record of use. Allowed, acknowledged use is not misconduct.
Is being asked about my work an accusation?
Not necessarily. Many processes start with a conversation to understand how the work was produced.
Should I write in a document that saves version history?
It helps. Cloud documents and most word processors keep a history that shows how the work developed over time.
Can a teacher fail me on a detector score alone?
Policies differ, but vendors and many universities say a score should prompt a closer look and a conversation, not decide the outcome.
This guide explains common rules; your institution's policy decides what is allowed on your work.
Sources
- Understanding the false positive rate for sentences of our AI writing detection capabilityTurnitin
- Three categories of GenAI use in assessmentUCL Teaching and Learning
- Fabrication and errors in the bibliographic citations generated by ChatGPT (Walters and Wilder, Scientific Reports 13, 14045)Scientific Reports
