Turnitin flagged my essay as AI, but I wrote it

Three things, all on its own blog:

• –That it does not decide misconduct: “Turnitin does not make a determination of misconduct”; it provides “data for educators to make an informed decision.”
• –That it can be wrong: its false positive rate is “not zero”, so the instructor is to apply their “professional judgment” and knowledge of the student and the assignment.
• –That it aims for a false positive rate below 1%, while acknowledging “there is still a small risk of false positives.”

Turnitin, Understanding false positives within our AI writing detection capabilities, 16 March 2023.

Below 1% sounds small. Vanderbilt University, turning the detector off later that year, pointed out that at 1% around 750 of the 75,000 papers its students submitted in a year could have been wrongly labelled (Vanderbilt’s guidance). Somebody is in that 1%, and it may be you.

A detector estimates how much a text resembles what a model would write. Clear, even, carefully edited prose resembles it more than a messy draft does, and so does writing in a second language: a Stanford study found several widely used detectors wrongly flagged 61% of essays written by non-native English speakers for a language exam, against about 5% of essays by native speakers, as MIT Sloan’s summary of it puts it (MIT Sloan). Running your essay through a grammar tool can push it the same way. Why detectors flag human writing goes into it.

• –The document’s version history. If you wrote in Google Docs, or in Word on OneDrive, it shows the essay growing over days, with dates. Share the original document, not a copy, which starts with no history.
• –Your notes, outline, earlier drafts and sources, each with the date you made or found it.
• –A short, plain account of when and where you wrote it, and over how many sittings.
• –An offer to talk the essay through: why it is built the way it is, what you cut, what you would change. Somebody who wrote an essay can usually do that without preparing.

If you wrote in Google Docs, your teacher may already use Draftback, which replays a document’s history like a film; see Google Docs version history as evidence.

Ask, in writing, to see the report and to meet. Say plainly that you wrote the essay and that you can show how. Quote Turnitin’s own line about professional judgment if it helps: it is not an accusation of your teacher, it is the tool’s maker agreeing with you that the score is not the end of it.

If it goes further, your school will have a written academic integrity process. Read it, ask what evidence it relies on, and ask whether an adviser can come with you. Accused of using AI walks through the rest.

Do not rewrite the essay to lower the score, and do not run it through a tool that promises to make it read as human. That changes the number, not what happened, and an essay that changed after the accusation is harder to defend.

Write it somewhere that keeps one. Google Docs keeps a version history, and Grammarly Authorship records what was typed and what was pasted. iwrote records how the essay is written as you write it in the browser, then gives you a link to hand in with it, which opens a certificate your teacher can check in their own browser. It covers writing done on iwrote from the first word, so it cannot vouch for an essay already handed in.

Handing in an essay with a certificate says which boxes to untick and what to keep.