Direct answer
An "AI catcher" is a detection tool that analyses a piece of writing and estimates how much of it was likely produced by a generative AI model such as ChatGPT, Claude, or Gemini. The most widely used one in higher education is Turnitin's AI writing detector, which reports a percentage of "qualifying text" plus sentence-level highlights rather than a simple guilty/innocent verdict [1]. Understanding what a catcher actually measures — and what it cannot measure — is the difference between panicking over a number and knowing whether your draft is genuinely at risk.
What Is an AI Catcher and How Does It Actually Detect AI-Written Text?
An AI catcher does not "recognise" AI writing the way a human recognises a friend's handwriting. It scores text statistically. Detectors work on prose segments above a minimum length, comparing patterns of word choice, sentence rhythm, and predictability against models of both human and machine-generated writing, then surface an overall percentage alongside highlighted passages so a reader can see which sentences drove the number [2]. That design matters: the output is a probability-shaped indicator attached to specific text, not a finding about authorship.
Turnitin is explicit that its AI writing indicator is a signal to be weighed alongside context, drafting history, and conversation with the student — not a determination that misconduct occurred [1]. The report separates the AI-writing percentage from the similarity score, so a document can show 0% similarity (nothing matched against published sources) and still carry a high AI indicator, or the reverse [2]. Students who conflate the two numbers routinely misread their own reports.
Two practical consequences follow. First, length matters: short submissions and heavily quoted or reference-heavy documents give the detector less qualifying prose to analyse, so the resulting signal is weaker and should be read more cautiously [1]. Second, the percentage is an aggregate — a 40% indicator does not mean "40% of your essay was written by AI" in any literal sense; it means a proportion of the analysed qualifying text was classified as AI-generated [2].
Why Do AI Catchers Sometimes Flag Human Writing as AI?
False positives are a real, documented limitation rather than a conspiracy theory. Because detection is probabilistic, errors can run in both directions: text written entirely by a person can be classified as AI-generated, and text produced by a model can pass as human [3]. Turnitin's own guidance acknowledges both false positives and false negatives as inherent to the approach, which is why the indicator is framed as one piece of evidence among several rather than a standalone judgement.
Certain writing styles are structurally more likely to be flagged. Highly formulaic academic prose — the "this essay will argue that… furthermore… in conclusion" register that tutors themselves teach — is predictable, and predictability is exactly what a detector is measuring. Non-native English writers who have been drilled into templated academic phrasing are therefore over-represented among false positives, a concern that has been raised repeatedly in the literature and in institutional guidance [3]. Editing tools that smooth sentence variety can push a human draft in the same direction.
The practical takeaway is not that detectors are useless, but that a score without the highlighted passages is close to meaningless. A flagged paragraph you can point to, explain, and evidence with your drafts and notes is a very different situation from a number in isolation — and institutions are advised not to act on the score alone for precisely this reason [3]. If you are challenged, the strongest defence is version history: dated drafts, notes, and reference material that show how the text evolved.
How Can I See What an AI Catcher Will Flag in My Draft Before I Submit It?
The most effective way to reduce risk is to look at the same kind of report your institution will look at, before the deadline rather than after it. Instructor-facing guidance on AI writing reports emphasises reviewing the specific flagged passages and discussing them, rather than escalating on the headline figure — which means the useful question for a student is always "which sentences are flagged, and why?" [4]. You cannot answer that question from a gradebook percentage; you need the highlighted report.
Once you can see the flagged segments, revision becomes targeted instead of blind. The guidance for handling AI writing reports centres on students revising the passages that triggered the indicator and being able to explain their own drafting process [4]. In practice that means rewriting predictable, generic stretches in your own voice, adding specific evidence and course-specific examples, and keeping your notes so the revision trail is visible. It also means separating AI flags from similarity matches, since the two are produced by different mechanisms and are fixed in different ways [4].
A pre-submission check is the low-risk version of this process. Turnitin0 is an independent service, not affiliated with Turnitin, LLC, that runs your draft through a Turnitin check and returns two PDFs in one order — an AI detection report and a similarity report, matching what instructors see in their LMS. Documents must be English, over 300 and under 30,000 words, and under 20 MB; turnaround is under 15 minutes in 98% of cases, and files are checked without being added to Turnitin's student paper database. One thing to expect: Turnitin displays *% instead of an exact figure when its AI confidence is below the 20% threshold, so a *% reading is a low-confidence signal, not a hidden high score.
Reading about how a catcher scores text is abstract until you see the report layout your tutor will open — the cover, the percentage, the flagged sentences, and the similarity summary side by side. That is the exact view turnitin0 delivers before you submit, so you can revise the flagged passages while there is still time to act on them.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI catcher the same thing as a plagiarism checker?
No. A plagiarism or similarity checker looks for matches against published sources and previously submitted papers, while an AI catcher estimates whether text was machine-generated. Turnitin reports them as two separate indicators, and a draft can score low on one and high on the other [2].
Can an AI catcher prove I used ChatGPT?
No. Turnitin describes its AI writing indicator as a signal requiring human interpretation, not a determination of misconduct [1]. Because false positives and false negatives both occur, institutions are advised to weigh the score alongside drafting evidence and conversation [3].
Why does my report show *% instead of a number?
Turnitin displays *% when its AI confidence falls below the 20% threshold. It is a low-confidence signal rather than a suppressed high score, and it typically appears on short or mixed documents where there is little qualifying text to analyse [1].
Does rewriting flagged sentences actually help?
Yes, and it is more effective when targeted. Reports are designed to show which specific passages triggered the indicator, so revising those segments — rather than rewriting the whole document — addresses the actual cause [4].
Can I check a draft without it counting as a real submission?
Yes, with an independent pre-submission service. Turnitin0 checks your file without adding it to Turnitin's student paper database and does not share reports with third-party databases, so a practice check does not follow you into your institutional submission.