Direct answer
Turnitin checks a submitted document against three comparison pools — its repository of previously submitted student papers, licensed publications and subscription databases, and the public internet — and separately runs an AI writing detection pass over qualifying text. It reports a similarity index with matched sources and, where enabled, an AI writing indicator with sentence-level highlighting. Turnitin does not itself decide whether plagiarism occurred; it surfaces evidence that an instructor interprets [1].
What Does Turnitin Compare Your Paper Against?
Turnitin's similarity engine works by matching strings of text in your submission against an index it has already built. That index is not a single database but a composite of three source families, and the match overview in the report groups results accordingly: internet sources, publications, and student papers [2]. Knowing which family a match belongs to changes what the match actually means, because a hit against a journal article and a hit against an essay submitted at another university in 2019 are very different events.
The student paper pool is the part students most often overlook. Every institution that has the repository switched on contributes submissions to it, which is why a passage you reused from your own earlier coursework can surface as a student-paper match rather than an internet match [2]. Turnitin's own documentation is careful to note that a match proves the text exists in the index — it does not prove who wrote it first, or that any copying was deliberate [1].
Institutional settings also change the surface area of the check. Instructors and administrators can exclude direct quotes, exclude the bibliography, set a word-count threshold below which matches are ignored, and decide whether the current submission is stored in the repository for future comparisons [2]. Two students can submit near-identical work and receive materially different reports purely because of how their departments configured the assignment.
What Turnitin does not check is equally important. It does not verify that your citations are accurate, that your sources exist, or that your argument is sound — a fabricated reference with correct formatting may pass through a similarity check untouched. It also does not read images, scanned handwritten pages, or text inside figures, so a screenshot of a paragraph is invisible to the similarity engine [1].
Does Turnitin Detect AI-Generated Writing, Not Just Plagiarism?
Yes — AI writing detection is a separate report from the similarity report, and it runs on a different mechanism. Rather than matching text against an index, the AI detector assesses patterns in the prose itself and returns an AI writing indicator: the percentage of qualifying text in the submission that it classifies as likely AI-generated [3]. The two reports sit side by side, which is why a paper can show 0% similarity and a high AI indicator, or the reverse.
The indicator has a display quirk worth understanding before you read a report. When the detected proportion falls below Turnitin's 20% confidence threshold, the interface shows an asterisk-and-percentage form rather than an exact figure, signalling a low-confidence signal rather than a precise measurement [3]. Above that threshold the report also highlights flagged sentences individually, so an instructor can inspect the specific passages rather than acting on the headline number alone.
Turnitin is explicit that this output is probabilistic and imperfect. Its documentation warns that the AI detector can produce both false positives and false negatives, and that it should not be used as the sole basis for any academic integrity decision [3]. Human academic writing that is highly formulaic — literature review scaffolding, methods boilerplate, heavily templated lab reports — is the classic false-positive profile, and non-native English writers have been the subject of documented concern in exactly this area.
The practical consequence is that "does Turnitin check for AI" is the wrong framing for your own risk assessment. The better question is which passages of your draft a detector would classify as machine-like, and whether you can defend them as your own drafting. That is a question you can answer in advance rather than after a misconduct referral [3].
How Can You See What Turnitin Would Flag Before You Submit?
The report is designed to be read formatively, not just punitively. Turnitin's own guidance describes the similarity report as a learning tool: the intended workflow is to open the match overview, look at each highlighted passage, and revise the paraphrasing or citation that produced the overlap before the final submission is graded [4]. Many institutions enable resubmissions precisely so that students can act on an early report rather than discover the problem at marking time.
Reading a report well requires knowing what to ignore. A match to a quoted passage that is correctly attributed is not misconduct, and a match to a bibliography is a formatting artefact rather than a borrowing problem — both are commonly excluded from the index by default settings [2]. The matches that deserve your attention are unquoted, uncited runs of prose that align closely with a single source, and matches against student papers, which are the hardest to explain after the fact [4].
The gap in the student workflow is timing. Turnitin's instructor-facing tools return a report only after a submission has been made to an assignment, and the repository setting on that assignment may store your draft permanently. Reviewing the same two outputs — similarity matches and AI indicator — on a private copy beforehand lets you revise with the report in hand while the submission you eventually make is the clean one [4].
That is the workflow turnitin0 was built around: upload your draft, and you receive the same two PDFs a professor sees — the similarity report with matched sources and the AI detection report with its indicator and sentence highlighting — without your file entering Turnitin's student paper repository. Turnitin0 is an independent service and is not affiliated with Turnitin, LLC; it exists so you can read the evidence about your own draft before anyone else does, in most cases within 5–15 minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin check every file type?
Turnitin accepts common text formats such as.docx,.pdf, and.txt, but its similarity engine reads text, not pixels — scanned pages and text inside images are not compared [1]. Turnitin0 applies the same practical limit, accepting English.docx,.pdf, and.txt files between 300 and 30,000 words.
Is a high similarity score automatically plagiarism?
No. The similarity index is the proportion of text that matched a source, and it includes correctly quoted and cited material unless those categories are excluded by the instructor's settings [2]. Turnitin states plainly that the report is evidence for an instructor to interpret, not a verdict.
Does Turnitin check AI and plagiarism in the same report?
They are two separate reports generated from one submission: the similarity report covers matching against the repository, publications, and the internet, while the AI writing report covers machine-like prose patterns and returns its own indicator [3]. Both are normally visible to the instructor.
Why does Turnitin show an asterisk instead of an AI percentage?
An asterisk form appears when the detected AI proportion is below Turnitin's 20% confidence threshold, meaning the signal is weak rather than zero [3]. It is a low-confidence display, not a clean bill of health.
Can I see what Turnitin would flag before I submit?
Yes, by running the same two reports on a private copy of your draft first. Reviewing matches and AI-flagged sentences before submission is the revision workflow Turnitin's own guidance recommends, and it is the workflow turnitin0 provides without adding your file to the student paper repository [4].