Direct answer
An AI detector is a tool that estimates how much of a piece of writing was generated by a large language model rather than written by a person. It does this by analysing statistical patterns across sentences — word choice, sentence rhythm, and predictability — and then reports the share of qualifying text it considers AI-generated [1]. Turnitin's AI writing detection is the version most university students meet, and it reports a percentage of flagged prose rather than a verdict on the whole paper [1]. Because these tools work on patterns rather than proof, understanding what the number means matters as much as the number itself.
How Does an AI Detector Actually Decide Whether Text Was Written by AI?
An AI detector does not look for a hidden watermark or a tag left behind by ChatGPT. It builds a statistical profile of your sentences and compares that profile against models trained on large collections of both human-written and machine-written academic prose [2]. The core signals are predictability and uniformity: language models tend to choose the most probable next word, which produces prose with unusually even sentence lengths, low lexical surprise, and repetitive connective phrasing [2].
Turnitin's implementation works at the sentence level first. Each sentence receives an internal probability, and the system then aggregates those probabilities into a single percentage describing how much of the qualifying text was flagged [2]. That aggregation step is why two papers with similar writing quality can receive quite different scores — the score reflects the proportion of flagged sentences, not the strength of the evidence in any one of them.
It is also why the detector's scope is narrower than most students assume. Turnitin's detection focuses on long-form, paragraph-structured prose; reference lists, direct quotations, and short-form answers are excluded from the calculation [1]. Text that has been substantially edited by hand may still carry the statistical signature of its AI draft, which is one reason manually "tweaking" a ChatGPT paragraph often fails to move the number [1].
How Accurate Are AI Detectors, and Why Do They Sometimes Flag Human Writing?
Accuracy is genuinely good but not absolute, and the honest answer is that no detector publishes a guarantee that applies to your specific discipline. False positives are a documented risk, and they cluster in predictable places: highly formulaic genres, heavily templated lab reports, and writing by non-native English speakers whose phrasing is more regular than the model expects [2]. Formulaic writing looks statistical because it is statistical — that is a property of the genre, not evidence of misconduct.
The single most common misreading is treating the headline percentage as a probability that the student cheated. It is not. The report expresses the proportion of qualifying text flagged, and Turnitin's own guidance tells reviewers to examine the highlighted sentence ranges and apply academic judgement rather than act on the number alone [3]. A paper can show a substantial percentage driven by two or three flagged paragraphs while the rest is clearly original.
For students, the practical implication is that the score is a signal to investigate, not a verdict. If a report flags a section, the useful question is whether that section genuinely reads like machine output — uniform sentence rhythm, generic transitions, no specific evidence — and whether you can point to the drafting history behind it [3]. Keeping drafts, notes, and version history is the most reliable defence available, because it addresses the underlying question the detector is only approximating.
Can You Check Your Own Draft With a Turnitin AI Detector Before You Submit It?
Institutionally licensed Turnitin does not give students a self-check option — report generation sits with the instructor's assignment, so a student cannot run their own draft through the campus system [4]. That gap is exactly why independent pre-check services exist: they return the same two reports an instructor would see, letting a student read the flagged sentence ranges while there is still time to revise [4].
The value of a pre-check is diagnostic rather than cosmetic. Seeing which sentences were flagged tells you whether the problem is a whole AI-drafted section that needs rewriting from your own evidence, or a handful of over-polished transitions that need to sound like you [4]. Students who check early tend to revise structurally; students who check the night before tend to make superficial edits that leave the pattern intact.
One detail worth knowing before you read any report: Turnitin displays *% instead of an exact percentage when AI detection falls below its confidence threshold. That asterisk is a low-confidence signal, not a zero, and it should be read as "no strong evidence" rather than "cleared" [1]. Understanding that convention prevents both unnecessary panic and false reassurance.
Reading a report is one thing; getting one on your own draft before the deadline is another. That is the specific problem turnitin0 was built for — a pre-submission check that returns the Turnitin AI detection report and the similarity report together, so you see the same flagged sentence ranges your instructor will see, typically in under fifteen minutes and without your file entering Turnitin's student paper database.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does an AI detector know which tool wrote my text?
No. Detectors identify statistical patterns associated with machine-generated prose in general; they do not attribute text to ChatGPT, Claude, or Gemini specifically [2].
What does an asterisk instead of a percentage mean?
Turnitin shows *% when AI detection falls below its confidence threshold, which signals a low-confidence result rather than a confirmed zero [1].
Can my university punish me based on the AI score alone?
Turnitin's guidance directs reviewers to examine flagged sentence ranges and apply judgement; the score is intended as one input, not a standalone finding [3].
Will editing a few words lower my AI score?
Usually not meaningfully. Sentence-level rewriting changes the statistical profile; swapping synonyms leaves the underlying rhythm intact [2].
Can I run my own draft through Turnitin before submitting?
Not through an institutional licence, which does not offer student self-checks. Independent services return the same report format for pre-submission review [4].