Direct answer
An AI detection tool is software that analyses a piece of writing and estimates how much of it was produced by a generative AI model such as ChatGPT, Claude, or Gemini, usually by scoring patterns in word choice, sentence rhythm, and predictability rather than by proving authorship [1]. Turnitin's detector is built into its existing report workflow, so an instructor sees an AI writing percentage next to the similarity score for the same submission [1]. The output is a probability signal, not a verdict, and Turnitin itself advises that it should never be the sole basis for an academic misconduct decision [3].
How does an AI detection tool actually work?
An AI detection tool does not search for a hidden watermark or a database of AI-written sentences. It reads the text you submit and asks a statistical question: how predictable is each next word, given the words that came before it? Language models are trained to pick highly probable continuations, so AI-generated prose tends to sit in a narrower band of predictability than human prose. Turnitin's detector evaluates passages for these machine-like patterns and reports the share of qualifying prose it considers AI-generated [2].
The report is deliberately granular rather than binary. Turnitin's guidance explains that the AI writing indicator appears alongside the similarity score in the same report view, so a reviewer can see both signals at once instead of juggling two separate tools [2]. Sentence-level highlighting lets the reader inspect exactly which passages triggered the flag and which did not. That matters because a single overall percentage hides a lot: a paper with a genuine 40% flag usually has identifiable clusters, not a uniform haze.
Two technical details shape how you should read any score. First, Turnitin excludes certain material from the calculation — very short documents and quoted text do not count toward the percentage, which is why a heavily quoted literature review may score lower than its visible AI-looking prose would suggest [2]. Second, Turnitin shows an asterisk (*%) instead of an exact figure when the AI signal falls below its confidence threshold, because a low-confidence reading is not precise enough to display as a number [3]. An asterisk is not a clean bill of health; it is a statement that the tool could not say much either way.
Finally, the tool is measuring text, not process. It has no access to your browser history, your drafts, your notes, or your conversations with a tutor. Everything it knows is in the file you uploaded, which is precisely why the guidance around it emphasises human interpretation over automated judgement [2].
How accurate are AI detection tools, and can they flag human writing by mistake?
No AI detection tool is perfectly accurate, and Turnitin says so directly: its own documentation acknowledges that the detector can produce both false positives and false negatives, and states that the report should not be used as the sole basis for an adverse decision about a student [3]. A false positive means human writing was flagged as machine-generated. A false negative means AI writing slipped through clean. Both happen, and both are documented rather than hypothetical.
Certain writing profiles are recognised as higher-risk for misattribution. Non-native English writers are a frequently cited concern, because a smaller working vocabulary and more formulaic sentence structures can resemble the statistical signature of generated text [3]. The same applies to heavily edited prose: when a writer repeatedly smooths and standardises their own sentences, the result can drift toward the predictable patterns a detector associates with AI. A student who has legitimately polished a hand-written essay with a grammar tool may therefore see a flag they cannot easily explain.
This is why the report design matters as much as the headline number. Because Turnitin highlights specific sentences and presents the score in a confidence band, a reviewer can distinguish between a paper where a few transition sentences tripped the model and a paper where entire sections read as machine-generated [3]. The percentage alone is the least informative part of the report; the highlighted passages are where the actual evidence sits. Turnitin's framing throughout is that the indicator is one signal among several, to be weighed against drafting history, supervision conversations, and the student's own account [3].
The practical takeaway for a student is that a flag is not an accusation and a clean score is not a guarantee. Treat any single detector run as a noisy measurement. If the stakes are high — a final dissertation, a resubmission, an integrity hearing — a second reading of the same draft, ideally before submission rather than after, is far more useful than arguing about a number afterwards.
What should you do if an AI detection tool flags your paper?
Start by reading the report properly instead of reacting to the percentage. Turnitin's interpretation guide directs reviewers to look at which sentences are highlighted and to compare those passages against the student's drafting history [4]. Do the same for yourself: open the highlighted sections and ask honestly whether they were generated, paraphrased from a model, or written by you in a flat register. Being able to answer that question specifically is far more persuasive than a general denial.
Next, assemble your process evidence. The guide recommends treating the report as one piece of evidence within a conversation rather than as a verdict, and suggests that students be invited to explain their process, produce earlier drafts, and discuss their sources [4]. Save your version history, your notes, your reading annotations, and any dated files. If your institution has a policy conversation with you, arriving with a timeline of how the document was built changes the discussion from assertion against assertion into a review of evidence.
Then act on the text itself. Where highlighted passages genuinely came from a model, rewrite them in your own reasoning and voice rather than swapping synonyms — detectors respond to structure and predictability, not just vocabulary. Where the passages are yours but read formulaically, vary sentence length and add the specific, idiosyncratic detail that human writing carries naturally. After revision, re-run the check to see whether the flagged passages actually changed, since the guide notes the report can be regenerated on a revised version [4]. If the flag persists on text you wrote yourself, that itself is useful information to bring to your tutor.
Finally, keep the sequencing right. The most defensible position is knowing your own report before anyone else does. A pre-submission check turns an unpredictable external judgement into a known quantity you can still do something about, and it gives you a documented baseline if a later institutional report looks different.
Reading a detector's output is one thing; seeing the actual report your institution's system will generate is another. turnitin0.com exists for that gap: students upload a draft and receive the Turnitin AI detection report and similarity report together, laid out the way a marker sees them, so the flagged passages and the percentage are in front of you while you can still revise. Reports are generated without adding your file to Turnitin's student paper database, most orders complete in under 15 minutes, and more than 100,000 reports have been delivered to over 20,000 students. If you would rather know the score before the deadline decides it for you, that is the point of checking early.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does an AI detection tool prove that a student used AI?
No. Turnitin's own documentation states the detector can produce both false positives and false negatives and should not be used as the sole basis for an adverse decision [3]. The report is a signal for human judgement, and the highlighted sentences are the evidence a reviewer is meant to examine.
Why does my report show *% instead of a number?
An asterisk appears when the AI signal falls below Turnitin's confidence threshold, meaning the tool cannot state a precise percentage [3]. It indicates a low-confidence reading rather than a confirmed zero.
Does quoted material or a short document affect the score?
Yes. Turnitin excludes quoted text and very short documents from the AI percentage calculation, so a heavily quoted paper may score differently than its visible prose suggests [2]. This is one reason the highlighted passages matter more than the headline figure.
Can I check my own draft before submitting it?
Students commonly do, and it is the most reliable way to know what a report will say while revision is still possible. A pre-submission check shows the same AI indicator and similarity view described in Turnitin's guides [2], so you can inspect flagged sentences and decide what to rewrite.
What is the single most useful thing to look at in an AI report?
The sentence-level highlighting, not the percentage. Turnitin's guidance tells reviewers to examine which passages were flagged and weigh them against the student's drafting history and account [4], and that same reading is the best way to prepare your own response.