Direct answer
Direct Answer - Turnitin detects two broad categories of issues: text that matches existing sources (plagiarism/similarity, including web pages, publications, and other student papers) and text that its model believes was generated by AI writing tools [1]. It reports both as signals — a similarity percentage and an AI-writing percentage — rather than as verdicts, and it does not detect contract cheating, undisclosed translation, or idea theft that leaves no textual trace [1][3]. Understanding what the system does and does not look at is the fastest way to judge your own submission risk.
What Exactly Does Turnitin's Similarity Report Detect, and How Does It Flag Matches?
The Similarity Report works by comparing your submission against Turnitin's repository, which contains archived student papers, web pages, and academic publications [1]. When text overlaps with any of those sources, Turnitin highlights the passage and attaches the matching source, so the report is a map of where your words already exist elsewhere rather than a single pass/fail number [1]. Each match is colour-coded and clickable, letting a marker jump straight to the exact sentence and the exact source that overlapped [1].
Crucially, the similarity percentage is not a plagiarism score. Turnitin's own guidance states plainly that the score is not an assessment of plagiarism and must be read alongside the matched sources themselves [2]. A paper can show a high percentage simply because it quotes heavily and cites correctly — quoted material and bibliography entries still appear in the report unless a filter is manually applied [2].
Matches are also grouped by source type — internet, publications, and student papers — and sorted by match size, so an instructor can prioritise the largest overlaps first [2]. The report shows where in the document each match sits, which is what allows a marker to distinguish legitimate citation from unacknowledged copying [2]. In practice, this means the similarity report detects textual overlap, not intent, and it cannot tell you whether that overlap was accidental or deliberate [1][2].
Can Turnitin Detect AI-Generated Writing, and How Reliable Is the AI Score?
Yes — Turnitin runs a separate AI writing detection indicator that is distinct from the similarity score [3]. When the model has enough text to analyse, it reports the percentage of the submission flagged as AI-generated along with the number of flagged segments, so you can see whether a small passage or most of the document triggered the signal [3]. The indicator only appears under certain conditions: the submission must be English-language long-form text in a supported file type, and very short pieces may not return an AI result at all [3].
Turnitin is explicit that its AI detection is not 100% accurate. The company acknowledges the model can produce both false positives (human writing flagged as AI) and false negatives (AI writing that slips through), which is why it positions the indicator as a signal for educators rather than proof of misconduct [3]. This is also why the AI score is displayed differently from a normal percentage in some cases: when detection confidence falls below Turnitin's threshold, the report shows an asterisk-style value instead of an exact figure, signalling a low-confidence result [3].
What the AI detector therefore does detect is statistical patterns typical of large language models such as ChatGPT, Claude, and Gemini — predictable phrasing, uniform sentence rhythm, and low lexical surprise [3]. What it does not detect reliably is AI-assisted editing of human prose, lightly paraphrased AI text, or AI output that has been substantially rewritten by hand [3]. Treating the AI percentage as one input among several, rather than a final judgement, matches how Turnitin itself describes the tool [3].
How Can Students See What Turnitin Detects on Their Own Draft Before Submitting?
The practical problem is access. Institutional Turnitin licences are designed for markers, and in most university setups students cannot run their own draft through the same system before the deadline [4]. That means the first time many students see a similarity or AI percentage is after the work has been graded — which is exactly the wrong moment to discover a problem [4].
Turnitin's own guidance frames detection as a conversation starter between students and educators rather than a punitive instrument, and it stresses transparency about what the system does and does not flag [4]. Students who understand how matches and AI segments are presented can revise with evidence — fixing genuine overlap, adding citations, or rewriting passages that read as machine-generated — instead of guessing [4]. The blog guidance is clear that knowing the boundaries of detection is the first step to using AI tools responsibly rather than avoiding them entirely [4].
That leaves a gap: you need a way to preview the same two reports your professor will see, on your own draft, before you submit [1][4]. A pre-submission check that returns a Turnitin AI detection report and a similarity report gives you the same matched sources, flagged segments, and score bands that an institutional run would produce, so you can act on the findings while there is still time to revise [1][2].
If you want to know exactly what Turnitin will say about your draft, the fastest route is to see the real report first. turnitin0 lets students preview the same Turnitin AI and similarity reports their professors see — matched sources, flagged AI segments, and score bands included — so you can fix what the detector actually catches before the deadline, not after.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin detect ChatGPT or only copied text?
Both. The similarity report detects textual overlap with its repository, while a separate AI writing indicator flags text its model believes was machine-generated by tools such as ChatGPT, Claude, or Gemini [1][3]. The two results are reported independently, so a draft can score low on similarity and still trigger an AI flag.
Can Turnitin detect paraphrased or rewritten content?
It detects close paraphrase well because the underlying phrasing still overlaps with a source, but heavy rewriting that changes structure and vocabulary can reduce the match [2]. Turnitin's own guidance warns that a low similarity score does not prove originality, only that no large textual matches were found [2].
Is the Turnitin AI detection score always accurate?
No. Turnitin states its AI detection is not 100% accurate and can produce both false positives and false negatives, and it shows a non-numeric value when confidence falls below its threshold [3]. That is why the indicator is designed as a signal for educators to interpret, not a definitive verdict [3].
Can I check my own draft in Turnitin before submitting?
Usually not through your university, because institutional access is set up for markers rather than student self-checks [4]. Many students use a pre-submission check to preview the same AI and similarity reports before the deadline [4].
What can Turnitin NOT detect?
It cannot reliably detect contract cheating (someone else writing your essay), undisclosed translation from another language, idea theft with no textual overlap, or AI text that has been substantially rewritten by hand [3][4]. Detection is a signal about text patterns, not about authorship intent.