Direct answer
AI detection software analyzes a piece of writing and estimates how likely it is to have been generated by a large language model rather than written by a human. In higher education the dominant tool is Turnitin's AI writing detector, which is embedded in the similarity report and scores text passage by passage before aggregating to a document-level percentage [1]. That percentage is a probabilistic signal, not proof of misconduct, and most universities require a human to review it before any academic integrity decision is made [1].
How Does AI Detection Software Like Turnitin Actually Decide Whether Writing Is AI-Generated?
Turnitin's detector is a transformer-based language model trained to recognise the statistical fingerprint of machine-generated prose. Rather than "knowing" who wrote something, it estimates, for each passage, the probability that the text was produced by an AI model, and then aggregates those passage-level estimates into a single document percentage [2]. This is why the output is expressed as a share of the text — for example, "34% of this document appears to be AI-generated" — and not as a verdict about authorship [2].
Two statistical properties drive most of that estimate: perplexity (how predictable each next word is) and burstiness (how much sentence length and structure vary). Human academic writing tends to be uneven, with abrupt shifts in rhythm, hedging, and idiosyncratic phrasing; model output tends to be smoother and more uniformly predictable, which is exactly the pattern the detector is tuned to notice [2].
Turnitin also limits where detection runs. It analyses the long-form body text of a submission, not reference lists, quotes, or very short answers, because those sections are too short or too formulaic for a reliable signal [2]. The tool does not identify which model was used, and it does not attempt to name an author — it only returns a likelihood that a stretch of prose resembles AI output [1][2].
How Accurate Is AI Detection Software, and Why Do False Positives Happen?
Turnitin has published a false-positive rate of under 1% in its own testing on documents where at least 20% of the text was AI-generated, but the company is explicit that the error rate is not zero [3]. That distinction matters: a low error rate across a large population still means some individual students receive a flag on writing they produced themselves.
False positives cluster in predictable places. Highly formulaic, heavily structured academic prose — literature review templates, methods boilerplate, standard essay scaffolding — can look statistically "smooth" enough to resemble model output. The same risk is elevated for non-native English writers whose drafting style is more regular than the detector's training baseline expects [3].
Turnitin's own reporting design acknowledges this uncertainty. When its confidence falls below roughly 20%, the report displays an asterisk (*%) instead of an exact number, signalling a low-confidence estimate rather than a firm finding [3]. Because the score is probabilistic, institutional policy — not the tool — determines consequences, and many universities require a human reviewer to weigh context before any misconduct finding is recorded [3].
Can I See My Own Turnitin AI Detection Score Before I Submit My Assignment?
Under normal institutional setup, no. Turnitin checks are run through a course or assignment that an instructor controls, and students generally cannot trigger a repository check against their own institution's account or see the resulting report themselves [4]. The report belongs to the instructor's workflow, which is why students often discover their AI percentage only after the fact.
That gap is exactly what pre-submission checking services exist to close. A student can upload a draft — typically a.docx,.pdf, or.txt file — and receive a Turnitin-format AI detection report alongside a similarity report before the real deadline [4]. This turns an opaque post-hoc score into something actionable while there is still time to revise.
Non-repository checking matters here. When a draft is checked without being added to Turnitin's student paper database, a later official submission is not matched against the student's own earlier upload, so the pre-check does not create a self-plagiarism or similarity artefact [4]. The practical use of these reports is diagnostic: identify the flagged passages, rewrite them, and re-check — not to contest an instructor's finding after the fact [4].
Reading about how detection works is useful, but it does not tell you what your own draft scores. turnitin0 gives students a way to see the same AI and similarity reports their instructors would see — before the deadline, not after.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does AI detection software prove a student used AI?
No. Turnitin's AI indicator is a probabilistic estimate of how much of a document resembles machine-generated text, and the company states it should not be treated as proof of misconduct [1]. Institutions typically require human review and context before any integrity finding is made [3].
Why does my Turnitin report show an asterisk instead of a percentage?
An asterisk (*%) appears when the detector's confidence is below its threshold — roughly 20% — so it cannot report a reliable exact figure [3]. It signals a low-confidence estimate, not a confirmed AI score.
Can AI detection software flag human-written essays?
It can, though Turnitin reports a false-positive rate under 1% in its own testing [3]. Risk rises with highly formulaic academic prose and with writing by non-native English speakers, whose drafting patterns can resemble model output statistically [3].
Does detection work on short answers or reference lists?
Turnitin runs AI detection on long-form body text rather than reference lists, quotations, or very short responses, because those sections are too brief or too formulaic to score reliably [2].
Can I check my own AI score before submitting?
Not through your university's Turnitin account, since instructors control submissions and report visibility [4]. Pre-submission services let you upload a draft and receive a Turnitin-format AI and similarity report, with non-repository checks that avoid adding your draft to the student paper database [4].