Direct answer
Yes — Turnitin detects both. Alongside its traditional similarity (plagiarism) report, it runs a separate AI writing detection indicator that estimates how much of a submitted document was likely generated by AI models, and it presents that estimate as its own percentage with highlighted passages [1]. The two signals are independent: a paper can show 0% similarity and still carry a high AI score, or show heavy matching against sources with no AI flag at all. Neither number is a verdict — the AI indicator is a probabilistic signal designed to support instructor judgement, not replace it [1].
Does Turnitin detect AI-generated writing, or only plagiarism?
Turnitin runs two distinct checks on the same submission, and conflating them is the single most common mistake students make. The similarity report compares your text against its databases of web pages, journals, and previously submitted papers and returns a percentage of matched text [1]. The AI writing indicator is a separate analysis that examines the linguistic patterns of your prose and estimates the share of the document likely produced by a generative model [1]. They live in the same submission view, but they answer completely different questions.
The practical consequence is that passing one check tells you nothing about the other. A student who rewrites an AI draft into their own words can drop the AI score while accidentally raising similarity through close paraphrase of a source; a student who writes entirely from scratch can show 0% similarity and still be flagged if their style is unusually uniform [2]. Turnitin's own guidance frames the AI indicator as a signal that requires human interpretation, precisely because it is measuring style rather than proving authorship [2].
There is also a scope limit worth knowing. The AI indicator only evaluates qualifying prose — passages long enough to analyse — so short quotes, headings, reference lists, and fragmented bullet points typically fall outside the calculation [2]. That means the percentage you see is a share of analysable text, not a share of the whole document. A 40% AI score on a 3,000-word essay does not mean 1,200 words were copied from ChatGPT; it means 40% of the text that qualified for analysis looked AI-like to the model [2].
Finally, the AI indicator is explicitly not an accusation. Turnitin describes its output as probabilistic and cautions that formulaic academic phrasing, heavy templating, and non-native English writing patterns can all push a score upward without any AI involvement [2]. Understanding that limitation is what separates a student who panics at a number from one who reads the report correctly.
How does Turnitin's AI writing detection actually work, and how accurate is it?
The detector works at the sentence-and-passage level rather than on the document as a whole. It segments your submission into qualifying text, scores each segment, and then aggregates those segment scores into the overall percentage shown at the top of the report [3]. The report also highlights the specific passages that drove the score, so you can see whether the flag is concentrated in one section or spread evenly across the essay [3]. That passage-level view is far more useful than the headline number.
Accuracy is the part most students misread. Turnitin positions the AI indicator as a confidence signal, not a certainty, and the report is built to reflect that: when the system's confidence falls below its threshold, the score is displayed as an asterisk-style low-confidence marker rather than a firm percentage [3]. A low-confidence result means the detector is not asserting that the text is AI-written — it means the evidence is weak in either direction [3]. Treating a starred result as a clean bill of health is as wrong as treating a high score as proof of misconduct.
Context changes the reading of any score. Turnitin's own interpretation guidance stresses that the percentage must be weighed against the assignment type, the drafting history, and the writer's known style, and that instructors should discuss flagged passages with the student rather than act on the number alone [3]. For students, the useful takeaway is directional: a high AI score with a tight cluster of flagged paragraphs points to a specific section worth rewriting, while a diffuse mid-range score across a formulaic methodology chapter is often just the genre talking.
The detector is also model-agnostic in practice. It is not checking for a watermark or a signature from a specific tool; it is measuring statistical regularity in word choice and sentence construction [3]. That is why light editing of an AI draft rarely clears a flag completely, and why genuine restructuring — new argument order, concrete examples, your own analysis — moves the number far more reliably than swapping synonyms [3].
What does a real Turnitin AI and similarity report look like before you submit?
A standard submission view contains two panels that students frequently mix up. The similarity panel lists matched sources with colour-coded overlapping highlights and a percentage of matched text, and Turnitin is explicit that this percentage is not a plagiarism score — it counts matches, including properly cited quotations and the bibliography [4]. The AI panel sits alongside it with its own percentage and its own highlighted passages [4]. Reading them together is the only way to know what an instructor will actually see.
The similarity panel rewards a specific reading habit: open the source list before reacting to the number. A 22% similarity figure driven entirely by your reference list and three block quotations is a completely different situation from 22% concentrated in an unquoted paragraph of your literature review [4]. Colour coding makes that distinction visible in seconds, and the match-by-match breakdown tells you whether a citation is missing or merely formatted unusually [4].
The AI panel rewards the same discipline at passage level. Because the report highlights the segments that contributed to the score, you can locate the exact paragraphs that read as machine-generated and decide what to do with them — add primary evidence, break the rhythm, introduce your own hedging and examples, or rewrite the section outright [4]. Students who only look at the headline percentage end up guessing; students who read the highlights end up editing with a target.
Seeing this layout before you submit is genuinely valuable, because both panels are what your instructor opens in the LMS. Turnitin0 produces the same two PDFs — an AI detection report and a similarity report — from a single pre-submission check, so you can inspect the flagged passages and the source matches while you still have time to revise rather than after a mark has been decided.
If the report shows a passage you did not expect to be flagged, the next step is not guesswork — it is seeing the same two panels your instructor will open, with the exact highlights and source matches attached. Turnitin0 runs your draft through a pre-submission check that returns a Turnitin AI detection report and a similarity report together, without adding your file to Turnitin's student paper database.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a 0% similarity score mean my paper is safe from the AI detector?
No. The similarity report and the AI writing indicator are separate analyses [1]. A paper can show no matched sources and still carry an AI score, because the AI indicator measures writing style rather than source overlap [2].
Can Turnitin prove that I used ChatGPT?
No. Turnitin describes the AI indicator as probabilistic and expects instructors to interpret it alongside drafting history and assignment context rather than treat it as proof [3]. It estimates likelihood, it does not establish authorship.
Why does my report show an asterisk instead of an AI percentage?
Turnitin displays a low-confidence marker when its confidence falls below the reporting threshold, meaning the evidence is weak in either direction rather than absent [3]. It should not be read as a clean result.
Does the AI detector analyse my whole document?
It evaluates qualifying prose only, so short quotes, headings, and reference lists typically fall outside the calculation [2]. The percentage reflects the share of analysable text, not the full word count [3].
If my similarity score is high, does that mean I plagiarised?
Not necessarily. The similarity percentage counts matched text, including correctly cited quotations and bibliography entries [4]. Open the source list and check where the matches actually sit before drawing conclusions.