Direct answer
Artificial intelligence detection tools are software systems that estimate how likely a piece of text was generated by a large language model, and in academic settings the most widely deployed example is Turnitin's AI writing detection, which is built directly into the similarity report rather than sold as a standalone product [1]. These tools do not "recognise" AI authorship the way a human reader might; they analyse statistical patterns across prose segments and return a confidence-weighted signal [1]. For students and instructors in the US, UK, Canada, Australia, New Zealand, and Ireland, understanding what these tools measure — and what their scores do and do not mean — is now part of normal academic writing practice.
How do artificial intelligence detection tools actually work?
Most detectors, including Turnitin's, work by measuring how predictable and how variable a piece of writing is. The model looks at perplexity, which reflects how surprising each word choice is given the preceding context, and burstiness, which reflects how much sentence length and structure vary across a passage [2]. Human academic writing typically shows higher variability — a short punchy sentence followed by a longer analytical one — while AI-generated prose tends to be more uniform and statistically smoother, and that uniformity is what the detector keys on [2].
The process is segment-based rather than all-or-nothing. Turnitin divides submitted prose into overlapping passages and scores each one for the likelihood that it was machine-generated, then aggregates the flagged segments into a single percentage [2]. This is why a report can show that part of a document is flagged while the rest reads as human-written, and why a mixed draft — AI framework with hand-written analysis — often produces a partial score rather than a clean pass or fail [1].
Two structural details matter for interpreting results. First, the detector only flags prose segments that cross its confidence threshold, so quotations, references, and bibliography sections are excluded from the analysis entirely [1]. Second, availability depends on the institution: detection is only active when the instructor's organisation has enabled it, which means two students at different universities can receive completely different levels of feedback on the same draft [1].
How accurate are AI detection tools, and why do scores sometimes show as *%?
No detector, including Turnitin's, claims certainty, and the reporting format is designed around that limitation. Turnitin only displays a numeric AI percentage when its model reaches at least 20% confidence that text is AI-generated; below that threshold the report shows *% instead of a number [3]. A *% result therefore does not mean zero AI use — it means the detector found some indication of AI writing but not enough confidence to assign a precise score [3].
False positives are a real and documented possibility, particularly with highly formulaic writing, heavily edited drafts, or text that has been run through paraphrasing tools [3]. This is precisely why Turnitin frames the AI writing indicator as a signal that supports instructor judgement rather than a verdict that replaces it [3]. A flagged score is a prompt to look more closely at specific passages, not proof of misconduct.
The practical implication for students is that a single number is a weak basis for panic or relief. Because the detector scores overlapping segments and aggregates them, the same draft can shift meaningfully after even modest revision of the flagged passages [2]. Reading the report at segment level — which paragraphs triggered the signal and why — is far more useful than reacting to the headline percentage [3].
How can students check their AI score with a real Turnitin report before submitting?
The core problem is visibility: in most institutions, students cannot see the AI writing indicator on their own submissions at all — only instructors see it inside the similarity report [4]. That gap means a draft can look finished to a student and still surface a flag the moment it enters the institutional system [4]. Pre-submission checking exists specifically to close that gap by returning the same report format instructors see, including the AI score, flagged segments, and similarity summary [4].
Working from an actual report changes how students revise. Instead of guessing which paragraphs read as machine-generated, they can see the specific flagged passages and rewrite those sections in their own voice while leaving the rest of the document intact [4]. Because the detector evaluates statistical uniformity, targeted edits to flagged segments — varying sentence length, adding specific evidence, breaking predictable phrasing — are the changes most likely to move the result [2].
Used this way, a detection tool becomes a diagnostic rather than a threat. Students who understand that scores are confidence-weighted and segment-based can treat a flag as editorial feedback, revise deliberately, and submit with evidence of what the report actually said rather than uncertainty about it [3][4].
If you want to see exactly what your instructor's report will show — the AI score, the flagged segments, and the similarity summary — turnitin0 lets you run that check on your own draft before you submit.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
Get [Real Turnitin](https://www.turnitin0.com/) AI & Similarity Report
FAQ
Do artificial intelligence detection tools prove that a student used AI?
No. Turnitin's own documentation treats the AI writing indicator as a signal that supports instructor judgement rather than a verdict, and false positives are possible with formulaic or heavily edited text [3]. A flag is a reason to examine specific passages, not proof of misconduct.
What does it mean when a Turnitin report shows *% instead of a number?
It means the detector found some indication of AI writing but did not reach its 20% confidence threshold, so no precise percentage is displayed [3]. It is a low-confidence signal, not a statement that the text is 0% AI.
Why can't I see my own AI score after submitting?
In most institutions the AI writing indicator is visible only to instructors inside the similarity report; students generally cannot see it on their own submissions [4]. That visibility gap is the main reason students look for a pre-submission check.
Which parts of my document are excluded from AI detection?
The detector only flags prose segments that cross its confidence threshold, and references, bibliography, and quotation sections are excluded from the analysis [1]. That is why a fully referenced essay can still show a partial score on its body text.
Can revising flagged passages change my score?
Yes. Because detection is segment-based and measures statistical uniformity, targeted rewrites of the specific flagged passages — varying sentence structure and adding concrete evidence — are the edits most likely to change the aggregated result [2].