What AI detectors do teachers use

Mostly whichever one was already installed. That single fact explains most of what people find confusing about this.

Most teachers do not pick a detector. Somebody chose one once, at institution level, and it arrived bolted to the system where work gets submitted. Nobody in the room that day was thinking about your cohort.

Turnitin, because it was already there

Turnitin is the most widely deployed by a distance, and the reason is boring: institutions had it for similarity checking years before AI detection existed. When the AI writing indicator was added to the same report in April 2023, nobody had to procure anything, run a pilot or approve a budget. It just appeared.

It sits inside the learning management system, so submissions get scored whether or not any teacher went looking. The report shows a similarity score and an AI writing indicator side by side, and those two measure completely different things, which is the single most common confusion we see. One is matching your text against a corpus. The other is a prediction with nothing behind it to click through to. See how Turnitin's AI detection works.

Copyleaks and GPTZero turn up where somebody chose AI detection deliberately rather than inheriting it, GPTZero especially where individual departments went their own way.

Everything else gets used ad hoc, by a teacher checking one submission that already looked odd.

Your own choice matters less than it looks

If your institution runs Turnitin in the LMS, that score exists regardless. Running a second tool afterwards does not replace it, it adds a number next to it.

Still worth doing. Detectors disagree for structural reasons, and a second measurement tells you whether the first one reflects the text or reflects that tool's calibration. But the institutional score is the one sitting in the record.

Four things to check before leaning on any number

Its false positive rate. Turnitin publishes under 1% of documents and around 4% of sentences, and says documents scoring below 20% have a higher incidence of false positives. Plenty of tools publish nothing at all.

Whether it shows you where. A document percentage hides the useful part. Flagged text bunched in one section means something different from the same percentage spread evenly, and only per-passage output tells them apart.

What it was measured on. A detector benchmarked on casual writing will exceed its published rate on formal academic prose, because regular writing sits nearer the line whoever produced it.

How it behaves on your students specifically. This is the one that gets skipped. Published research keeps finding higher false positive rates for non-native English speakers, for reasons in why non-native English writing gets flagged. If a good share of your cohort writes in a second language, the vendor's average understates what you will see.

What the number actually is

A probability from a classifier with a measurable, non-zero error rate. Not a match against a source document. And it carries no record of how the text was made.

Turnitin's own documentation says the indicator should not be the sole basis for action. Their analysis also found that 54% of falsely flagged human sentences sit directly next to a sentence scored as AI written, and another 26% two sentences away. Wrong flags cluster around real ones rather than scattering, which makes mixed documents the hardest case of all, and mixed documents are increasingly the normal case.

What follows from a score is a matter for your institution's own procedures. Those vary, and we are not the right people to describe them.

Getting a second opinion

Ours is free and unlimited, no account. It scores each paragraph separately rather than handing you one number for the document, and we publish the false positive rate: 0.8% at the public setting, 0.4% stricter, measured on pre-LLM academic writing.