Direct answer
Checking a document for AI generation means running it through a detector that scores how statistically machine-like its prose is, then reading that score as a probability rather than a verdict. Turnitin's AI writing detection reports the share of a document's qualifying prose it classifies as AI-generated, and it presents that figure alongside flagged segments so a reader can inspect the evidence rather than trust a single number [1]. The practical difficulty is not running the check — it is interpreting what comes back, because the same text can score differently depending on length, editing history, and how much of the document qualifies for analysis [1]. This guide explains what an AI generation check measures, why scores sometimes appear as an asterisk, and how to check a draft without exposing it to a repository.
What Does An AI Generation Check Actually Measure?
An AI generation check measures predictability, not authorship. Detectors estimate how likely each word in a passage is given the words before it, and machine-written prose tends to sit in the low-perplexity, low-variation range that language models naturally produce [2]. Turnitin's capability reports the percentage of a document's qualifying prose that its model classifies as AI-generated, and it breaks that figure down into individual flagged segments so the reader can see exactly which passages triggered the signal [2]. Nothing in the output identifies which tool produced the text — ChatGPT, Claude, and Gemini all leave a similar statistical signature, and a high score does not distinguish between them [2].
The qualifying-prose rule matters more than most students expect. Only prose of sufficient length is scored, so reference lists, tables, short-answer responses, and brief submissions may be excluded from the calculation entirely [2]. That means a check run on an excerpt can misrepresent the full document: a 400-word introduction lifted out of a 4,000-word essay carries a different denominator and can produce a materially different percentage [2]. The same principle explains why detectors are less reliable on heavily edited or mixed human/AI text, where the statistical signature is genuinely ambiguous — which is why the report shows segments rather than issuing a single pass/fail verdict [1].
A further consequence is that the score is document-relative, not passage-absolute. Adding or removing text changes the denominator, so identical sentences can carry different percentages in different drafts [2]. This is not a flaw in the check; it is what a proportion-based metric does. Readers who treat the number as a fixed property of their writing will misread it, and readers who treat it as a snapshot of one document at one moment will read it correctly [1].
Why Does A Turnitin AI Score Show An Asterisk Instead Of A Percentage?
An asterisk appears when the AI score falls below Turnitin's confidence threshold, meaning the detection signal is too weak for the system to report a specific figure [3]. In practice, *% is a low-confidence result rather than a numeric score, and it should be read as "no strong machine signal detected here" — not as a certified clean result and not as an accusation [3]. Students often misread the asterisk in both directions: some assume it means zero percent AI, others assume the system is hiding a high score. Neither reading is supported by how the threshold works [3].
The report itself separates two categories that are easy to conflate. Text classified as AI-generated is treated differently from text classified as AI-paraphrased, and each flagged segment can be inspected individually to see which category it falls into [3]. That distinction matters when a draft was written by hand and then polished with an AI assistant: the paraphrased category is where that pattern usually surfaces, and it is also where human revision can genuinely change the outcome [3]. Because the score is document-relative, the same passage may move above or below the threshold once surrounding text is rewritten [3].
Institutional context completes the picture. Turnitin's own guidance frames the AI indicator as one signal among several, to be read alongside draft history, citation practice, and the student's own account of how the work was produced [1]. A flag is a prompt for a conversation, not a finding of misconduct, and a low score is not a guarantee of authenticity [1]. Understanding that framing is what lets a reader use the check constructively instead of panicking at a number or dismissing an asterisk [3].
Can You Check A Draft For AI Generation Without Adding It To Turnitin's Student Paper Database?
Students cannot generate an official Turnitin AI report themselves, because report generation sits with instructors and administrators inside an institutional account [4]. That structural fact creates the real risk: a submission made to a repository-enabled assignment is stored and matched against future submissions, so test-submitting a final draft to "see the score" can create the very similarity problem the student was trying to avoid [4]. Turnitin's student-facing guidance is explicit that a draft should not be burned on a live assignment just to obtain a preview [4].
The workable alternative is a non-repository check — a submission that is scored but not added to the student paper database and not shared onward [4]. This is where turnitin0.com fits: it returns a Turnitin AI detection report and a similarity report as two downloadable PDFs from the same file, checked without repository retention, with most orders completing in under fifteen minutes. A check costs $3.80 for a single scan, or $2.75 per check in a ten-check pack, and the service is independent of Turnitin, LLC.
Whatever route a student takes, the interpretive discipline is the same. Keep drafts and revision history, because a flagged score is contextualised by evidence of the writing process, and a documented trail of human revision is the strongest answer to a false positive [4]. Read the segment breakdown rather than the headline percentage, since the flagged passages are what an instructor will actually look at [3]. And run the check early enough that a genuinely high score can still be fixed by rewriting rather than explained after the fact [4].
A check is only useful if you can act on what it returns — and the difference between a calm revision and a panicked email at 2am is usually whether you saw the report before your instructor did. turnitin0.com exists for that gap: the same two PDFs your professor sees, in your hands while the draft is still yours to change.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a low AI score mean my writing is definitely human?
No. A low score means the detector found no strong machine signal in the qualifying prose, which is evidence rather than proof [1]. Heavily edited AI text and lightly polished human text can both land in ambiguous territory [2].
Why did my score change between two checks of the same essay?
Because the AI percentage is a proportion of the document's qualifying prose, so adding, cutting, or reformatting text changes the denominator [3]. Small edits near the threshold can move a result from an asterisk to a number or back [3].
Can I check a draft without it being stored in Turnitin's database?
Yes, provided you use a non-repository check rather than test-submitting to a live assignment, which stores the file and matches it against future submissions [4]. Confirm the retention terms before uploading anything you plan to submit [4].
What should I do if the report flags a section I wrote myself?
Read the flagged segments rather than the headline figure, then gather your drafts and revision history to show the writing process [4]. A segment-level flag is a prompt for review, not a finding of misconduct [1].
Does the check tell me whether ChatGPT, Claude, or Gemini wrote the text?
No. Detection identifies a statistical signature consistent with machine generation; it does not attribute text to a specific model [2].