Direct answer
Turnitin's AI text detector is a sentence-level classifier built into the Similarity Report that estimates how much of a document was probably generated by AI, then reports that estimate as a percentage of qualifying text [1]. It does not read minds and it does not prove authorship; it flags patterns in wording and structure that resemble large language model output, and Turnitin itself frames the result as a signal for human judgement rather than a verdict [1]. For students, the practical consequence is that the score you receive is a probabilistic estimate of your draft, not a certificate of innocence.
How does the Turnitin AI text detector decide which parts of a document are AI-generated?
The detector does not evaluate a document as one undifferentiated block. It works in sentence-level units, scoring individual sentences and then grouping consecutive sentences into predicted AI segments, which is why a report can show a clean opening paragraph and a flagged middle section inside the same essay [2]. That segmentation matters more than most students realise: a single rewritten paragraph can move the reported figure substantially because the percentage is calculated from the qualifying text that was flagged, not from the document as a whole [2].
There is also a hard length gate. Turnitin does not produce an AI score for very short submissions — documents below roughly 300 words fall under the minimum threshold and return no AI percentage at all [2]. This is a frequent source of confusion when a student checks a discussion post or a short reflection and sees nothing where they expected a number.
Turnitin describes the flagged total as a proportion of qualifying text, meaning the detector distinguishes the body of the submission from material it does not treat as the author's own prose [2]. Quoted passages, reference lists, and similar non-qualifying elements are handled separately from the scored text, which is why the report presents both a percentage and a count of flagged segments rather than one bare number [2]. Reading both figures together tells you far more than the percentage alone.
The model is trained to recognise statistical regularities associated with generated text, so it responds to uniformity — evenly paced sentence lengths, low lexical variance, predictable transitions — rather than to any single tell-tale phrase [1]. Turnitin is explicit that detection is probabilistic and that false positives are a known limitation, which is why institutional guidance treats a flag as a prompt for a conversation rather than an automatic finding of misconduct [1].
What does the AI score percentage actually mean, and why does Turnitin sometimes show an asterisk instead of a number?
When Turnitin displays *% instead of a figure, it means the detected amount of AI-generated text fell below the detector's confidence threshold — currently 20% — so the numeric score is suppressed rather than shown [3]. This is the single most misread element of the whole report. *% is not a zero, and it is not an accusation; it is Turnitin saying the signal was too weak to state a percentage with confidence [3].
The percentage itself is a document-level aggregate of sentence-level predictions [3]. That has two consequences students routinely miss. First, a low overall score can still contain a genuinely flagged passage, because the flagged sentences are diluted by the unflagged remainder. Second, a high score can be driven by a handful of long, formulaic paragraphs while the rest of the essay is entirely your own writing.
Turnitin positions the score as one input among several and warns that it should not be used as the sole basis for an academic integrity decision [3]. Reports are designed to be read alongside the draft, the assignment brief, and the student's own account of how the work was produced. Institutions that treat the number as self-executing tend to be the ones generating appeals.
Practically, the number is best read as a risk indicator for your specific draft. A score in the low single digits on a long essay is a very different signal from the same figure on a 400-word submission, and the segment count is what tells you which situation you are looking at [3]. That is why the report shows both.
Can you check your own draft with Turnitin before your instructor sees it?
Whether you can self-check through Turnitin depends entirely on how your institution has configured its account, because Turnitin does not sell a public consumer checker [4]. Where a university enables a draft or practice assignment, students can submit there and receive a report without that submission counting toward the graded version [4]. Where the setting is switched off, there is no official route for a student to see their own AI score in advance.
There is a second consideration even when self-checking is available: depending on the institutional settings, a draft submitted through the official system may be retained in the paper repository, and that stored copy can later match against your own final submission [4]. Students who run a draft through an institutional assignment without checking the settings sometimes create the very similarity problem they were trying to avoid.
This institutional dependency is the gap that independent pre-submission checkers exist to fill. A non-repository check on your own draft — one that returns both an AI detection report and a similarity report without adding your file to a student paper database — gives you the same two documents your instructor will look at, before the deadline rather than after it [4]. For a student whose university has not enabled draft submissions, that is often the only way to see the verdict early enough to act on it.
The useful habit is to check early, read the segment breakdown rather than only the headline percentage, and revise the specific passages that were flagged instead of rewriting the entire document [4]. That keeps your own argument and citations intact while addressing the part of the text the detector actually responded to.
Reading the mechanics is one thing; seeing your own draft's score, flagged segments, and similarity matches side by side is what actually removes the guesswork. turnitin0 gives you that preview — real Turnitin AI and similarity reports on your own file, delivered in under 15 minutes in 98% of cases, with no repository storage and no subscription.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin's AI detector work on documents under 300 words?
No. Turnitin does not return an AI score for very short submissions, so a short paragraph or discussion post will show no percentage at all [2]. If you need a meaningful reading of your prose, the draft has to be long enough to clear the minimum threshold.
Is a high Turnitin AI score proof that a student cheated?
No. Turnitin describes detection as probabilistic and acknowledges false positives as a known limitation, so the score is a signal for review rather than a finding [1]. Institutional policy, not the number itself, determines what happens next.
What is the difference between *% and 0% on an AI report?
*% means the detected amount fell below Turnitin's 20% confidence threshold and the numeric score was suppressed, while 0% would indicate no qualifying text was flagged [3]. They are not interchangeable, and *% should not be read as a clean zero.
Can I run my own draft through Turnitin without my university?
Turnitin does not offer a public consumer checker, so access normally runs through your institution's configuration [4]. Independent pre-submission services provide the same report types on your own file without a repository submission.
Does checking my draft first affect my final submission?
Not with a non-repository check, which does not add your file to Turnitin's student paper database. Institutional draft assignments are different — depending on settings, those submissions may be stored and can later match against your final version [4].