Direct answer
An AI scanner reads text as a statistical pattern rather than as meaning: it measures how predictable each word is (perplexity) and how much sentence rhythm varies (burstiness), then flags the stretches that look too smooth and too uniform. Turnitin's AI writing detector does this only on qualifying prose, produces an overall percentage of scored text, and shows *% instead of a number when the signal falls below its 20% confidence threshold [1]. The result is a signal for review, not a verdict on authorship.
How Does an AI Scanner Analyze Text and Decide What to Flag?
An AI scanner does not search for a database of AI sentences. It scores your text probabilistically, one segment at a time, and compares the result against the distribution it learned from large volumes of human and machine writing. Turnitin's detector leans on two signals: perplexity, which measures how surprising or predictable each word choice is, and burstiness, which measures how much sentence length and rhythm vary across a passage. Human academic prose is uneven — a long clause followed by a blunt short sentence — while model output tends toward a steady, mid-length cadence that scores low on both measures [2].
The scanner also applies strict eligibility rules before it scores anything. Only running prose is evaluated; reference lists, block quotations, headings, tables, and short strings are excluded from the analysed segment, and a submission needs a substantial body of continuous prose before a report is generated at all. This is why a 400-word reflection and a 4,000-word dissertation do not produce comparable reports, and why a document padded with citations can show a very different percentage than one dense with argument [2].
Flagging is segment-level, not sentence-level. The detector marks contiguous passages that cross its internal threshold and then aggregates them into a single percentage of the text it actually scored. A document can therefore be 30% flagged while containing no single sentence that reads as obviously machine-written — the flagged unit is the paragraph, and the paragraph is judged as a whole [2].
Finally, the scanner is transparent about its own error rate. Turnitin states that its detector can produce false positives, particularly with highly formulaic academic registers — literature reviews, methodology boilerplate, and standard essay scaffolding — and with non-native-English writing that has been heavily polished by grammar tools. That admission is deliberate: the score is designed to open a conversation about a draft, not to close one about a student [2].
What Do the AI Score and the Asterisk (*%) on a Scan Report Actually Mean?
The headline percentage on an AI writing report is the share of scored prose that the detector flagged — not the share of your whole document. If your report says 24%, it means roughly a quarter of the qualifying prose segments were flagged, while the excluded material (references, quotes, headings) contributed nothing to that figure either way. Reading the number as "24% of my essay was written by AI" is the single most common misreading of these reports [3].
The asterisk is where most confusion starts. When Turnitin's confidence in the AI signal falls below its 20% threshold, the report displays *% instead of an exact figure. That is not zero and it is not a clean bill of health — it is a low-confidence band meaning the detector found some machine-like characteristics but not enough to commit to a number. Students who see *% and assume they are safe are reading the symbol backwards; students who see *% and panic are reading it backwards too [3].
Context matters as much as the figure. Turnitin positions the score as one indicator among several, to be weighed against drafting history, version control, and a conversation with the author, and it explicitly cautions against using the percentage as the sole basis for a misconduct finding. Institutions set their own thresholds and their own escalation rules, which is why the same report can be a non-event at one university and a required meeting at another [3].
Practically, the score is most useful as a diagnostic on your own draft. A cluster of flags in your literature review usually points to summarising that mirrored your sources' phrasing too closely; flags spread evenly across an argument often trace back to a section you drafted with a model and lightly edited. Either way, the report tells you where to look, and the exact percentage tells you far less than the location of the highlighted passages [3].
Can You Scan Your Own Text Before Submitting It to Your University?
You can, and Turnitin's own guidance supports students seeing and discussing their AI score rather than encountering it for the first time in a misconduct letter. The practical caveat is policy, not technology: institutions vary widely on whether draft checks are permitted, whether they count against a submission allowance, and whether a self-checked draft must be disclosed. Confirm your department's position before you build a workflow around pre-submission scanning [4].
Where draft checks are allowed, the useful move is to scan early enough to act on the result. A scan run the night before a deadline tells you that you have a problem but gives you no room to fix it; a scan run a week out turns the report into a revision plan. Turnitin's guidance frames the score as an input to revision — identifying passages that read as machine-generated so the author can rewrite them in their own voice, with their own examples and their own reasoning [4].
There is also a legitimate integrity argument for self-scanning. If a section of your draft was drafted with ChatGPT, Claude, or Gemini and then edited, you may genuinely not know how much of the machine cadence survived your edits. A scan answers that factual question before your institution answers it for you, and it lets you correct an honest mistake — over-reliance on a model for scaffolding — instead of defending one [4].
What self-scanning cannot do is make a flagged draft unflagged. The report is descriptive; it does not rewrite anything. If the scan shows heavy flagging, the work that follows is either substantive revision in your own words or a rewrite pass that restores natural variation to the prose. Treat the scan as the diagnosis and the revision as the treatment, and confirm with your tutor whenever the two diverge [4].
If you want that diagnosis before your deadline rather than after it, turnitin0 runs your draft through a real Turnitin AI detection check and returns the same AI and similarity reports your institution would see — typically in under 15 minutes, with no repository submission and no subscription.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does an AI scanner store my text?
Not necessarily. Turnitin0 checks your file without adding it to Turnitin's student paper database and without sharing reports with third-party databases, and you can delete files from your account. Institutional submissions follow your university's own repository settings, which you should check separately [1].
Why does my scan show *% instead of a number?
Turnitin displays *% when the AI signal falls below its 20% confidence threshold. It indicates a low-confidence detection band, not a zero score, so read it alongside the highlighted segments rather than as a clean result [3].
Can a human-written essay be flagged by an AI scanner?
Yes. Formulaic academic registers and heavily proofread non-native-English writing are the two most commonly cited sources of false positives, which is why Turnitin advises against treating the percentage as proof of misconduct [2].
Does the AI score include my references and quotes?
No. Only qualifying running prose is scored; reference lists, block quotations, and headings are excluded from the analysed segment, so the percentage reflects your body text alone [2].
Is scanning my draft before submission allowed?
It depends on your institution. Turnitin's guidance supports students seeing their AI score, but departments differ on whether draft checks are permitted, so confirm the policy with your tutor before relying on a pre-submission scan [4].