Direct answer
An AI detector is software that estimates how much of a piece of writing was produced by a generative AI model rather than by a human, and Turnitin's version reports that estimate as a percentage of the submission. It does not read minds or trace keystrokes; it analyses statistical patterns in the text itself and flags the passages that look machine-generated [1]. For students, the practical consequence is simple: the detector produces a number attached to your draft, and that number can travel to your instructor without your ever seeing it first.
Introduction
An AI detector is software that estimates how much of a piece of writing was produced by a generative AI model rather than by a human, and Turnitin's version reports that estimate as a percentage of the submission. It does not read minds or trace keystrokes; it analyses statistical patterns in the text itself and flags the passages that look machine-generated [1]. For students, the practical consequence is simple: the detector produces a number attached to your draft, and that number can travel to your instructor without your ever seeing it first.
Understanding what that number means — and what it does not mean — is the difference between panicking over a stray flag and fixing a genuine problem before it costs you marks. This guide walks through how detection works, why the score sometimes appears as an asterisk, and how to preview your own result before submission.
What Does an AI Detector Actually Measure, and How Accurate Is It?
An AI detector measures pattern regularity. Turnitin's model divides a submission into segments and evaluates each one for the statistical signatures that tend to accompany machine-generated prose — unusually uniform sentence rhythm, predictable word choices, low burstiness, and tightly consistent phrasing [2]. The output is not a forensic finding but a probability-style indicator: a judgement about how likely the text is to have been produced by a large language model. That distinction matters, because a probability is something you can argue with and something an instructor is expected to interpret in context.
Accuracy is therefore conditional rather than absolute. Turnitin's own guidance acknowledges that false positives are possible, particularly in heavily formulaic academic writing, in text written by non-native English speakers, and in passages that have been heavily edited or paraphrased by tools [2]. A high score is a prompt for a human to look closer, not proof of misconduct, and Turnitin explicitly positions its report as a signal for review rather than a verdict [2]. Institutions that treat the percentage as automatic guilt are misusing the tool, and most academic integrity policies now say so.
What the detector is genuinely good at is consistency. The same passage run through the same model twice will generally land in a similar band, which is why a score that jumps from single digits to eighty percent between drafts is worth investigating on your side before your tutor investigates it for you [2]. The detector is a mirror of textual style, and style is something you can change.
Why Does a Turnitin AI Score Show an Asterisk (*%) Instead of a Percentage?
The asterisk is a confidence marker, not a missing value. When the amount of text Turnitin's model identifies as AI-generated falls below its confidence threshold — currently around twenty percent — the report replaces the exact figure with an asterisk rather than publishing a number the system cannot stand behind [3]. You are seeing a low-confidence signal, which is materially different from a confirmed detection. Turnitin designed the display this way precisely so that a weak signal is not read as a strong accusation.
In practice, *% usually means the detector found some AI-like patterns but not enough evidence to commit to a figure [3]. It is the outcome most commonly produced by human writing that has been lightly polished with a grammar tool, by technical writing with repetitive structure, and by short submissions where there is not enough text to build a stable estimate. Students often misread the asterisk as "score hidden because it is high"; the opposite is closer to the truth.
The asterisk also carries an important limitation for anyone trying to self-assess. Because the threshold sits near twenty percent, a draft that genuinely contains a modest amount of AI-assisted phrasing may register as *% and give you no numeric feedback at all [3]. If your goal is to know whether your revisions actually worked, an asterisk is a weak instrument — it tells you that you are near the floor, not where exactly you are standing.
Can You Check Your Own Draft With a Turnitin AI Detector Before You Submit?
Normally, no — your institution's Turnitin licence does not include a student self-check button. Turnitin is deployed through an LMS assignment, which means the first time most students see an AI or similarity report is after the deadline has passed and the report has already reached their instructor [4]. That asymmetry is the single biggest source of avoidable stress in the whole process: the person with the least information is the one being judged.
Non-repository checking closes that gap. A non-repository check runs your document through the same detection and similarity engines without adding it to Turnitin's student paper database, so your draft is not permanently indexed and cannot later be matched against itself [4]. That matters for two reasons. First, submitting a draft to a repository-based checker can inflate your own final similarity score when you submit the real version. Second, a non-repository check lets you iterate — revise, re-check, revise again — without leaving a trail of indexed copies behind you.
The workflow that experienced students actually use is unglamorous: finish the draft, run a non-repository check, read the flagged passages rather than just the headline number, rewrite the sentences that genuinely sound machine-generated, and re-check once. Turnitin's own guidance for students emphasises exactly this — that the report is most useful when it drives revision of specific flagged text rather than a scramble to lower a number [4]. If a passage is flagged and you wrote it yourself, you now know which sentence to restructure; if it is flagged and you did lean on a model, you now know before your tutor does.
Reading a report is one thing; producing one on demand is another. turnitin0 exists for that second step — an independent pre-submission check that returns the actual Turnitin AI detection report and similarity report, the same documents your professor sees in the LMS, so you can review every flagged segment while there is still time to act on it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI detector the same thing as a plagiarism checker?
No. A similarity checker compares your text against source databases to find uncredited matches, while an AI detector estimates whether the wording itself was machine-generated [1]. Turnitin reports the two separately, which is why a submission can have zero similarity and still carry an AI flag.
Does a high AI score prove I cheated?
No. Turnitin frames its AI indicator as a signal for human review, not a finding of misconduct, and false positives are documented [2]. Formulaic academic phrasing, heavy editing, and non-native English writing are all recognised contributors to inflated scores.
Can I get an exact AI percentage if my report shows an asterisk?
Not from that report. The asterisk appears when detected AI text falls below Turnitin's confidence threshold of roughly twenty percent, so the system withholds a precise figure [3]. Re-checking a revised draft is the only way to see whether the signal has moved.
Will checking my draft add it to Turnitin's student paper database?
It depends on the service. A non-repository check does not index your file, so it will not be matched against your own final submission later [4]. Always confirm the repository setting before uploading a draft anywhere.
What is the fastest way to lower a flagged score?
Rewrite the specific flagged segments in your own voice rather than paraphrasing them mechanically — vary sentence length, replace generic transitions, and add concrete detail [4]. Then re-check to confirm the change registered.