Direct answer
Checking for AI-generated text means running a draft through a detector that segments the prose and estimates what share of it looks machine-written, then reading that estimate as a signal rather than a verdict. Turnitin's AI writing detector does exactly this: it highlights suspected passages, reports an overall percentage of qualifying text, and surfaces the result to instructors inside the LMS [1]. Because the score is probabilistic, the practical skill is not just generating a number but interpreting it correctly — especially when the report shows *% instead of a precise figure [3].
How Do Detectors Actually Identify AI-Generated Text?
Detection is a statistical comparison, not a plagiarism match. Turnitin's indicator works at the level of prose segments, evaluating how predictable and uniform the word choices are relative to what its models expect from human academic writing [2]. Text produced by large language models tends to be low-perplexity and low-variance, which is why heavily templated human writing can occasionally look similar. The detector reports what it found, not why the writer wrote it that way.
Coverage matters as much as the percentage. Only qualifying prose is analysed, so reference lists, short quotations, headings, and very brief passages are typically excluded from the calculation [2]. That is why a 4,000-word essay with a large block-quoted appendix can return a lower percentage than a 1,500-word reflective piece with no quotations at all. Two students with identical habits can therefore receive different scores purely because of document structure.
Mixed authorship produces mixed scores, and this is the most common source of confusion. A draft that was outlined by a model, filled in by hand, then polished with an assistant often lands in the middle of the range rather than at either extreme [2]. A partial percentage is describing a partial pattern in the text — it is not accusing the writer of a partial offence. Treating the number as a proportion of "how much cheating happened" misreads what the tool measures.
Finally, the detector is deliberately positioned as a review prompt. Turnitin frames the indicator as one input for an educator's judgement, not as standalone proof of misconduct [2]. That framing matters for students too: a flagged paragraph is a reason to look closely at how that paragraph was drafted, not a reason to panic about the whole submission.
What Does a Turnitin AI Score Percentage Actually Mean?
The headline number is the share of qualifying prose that the detector flagged, expressed as a percentage of the analysed text — not a percentage of your grade, your effort, or your integrity [3]. If the report shows 34%, it means roughly a third of the analysed prose carried the pattern the model associates with machine generation. The remaining two-thirds did not. Neither figure tells an instructor who typed which sentence.
Turnitin also suppresses precision at the low end. When the AI signal falls below its confidence threshold, the report displays *% instead of an exact percentage, signalling a low-confidence result rather than a precise measurement [3]. Students frequently misread this as a glitch or a hidden score. It is neither: it is the system declining to overstate what it can support. A *% result is generally the outcome you want on a draft, because it means the detector did not find a strong enough pattern to commit to a number.
Context changes the reading of any score. Institutional policies differ on whether the indicator may be used in misconduct proceedings at all, and many universities require human review of the flagged passages before any decision [3]. A high percentage on a heavily structured methods section, or on a paragraph you genuinely wrote from your own notes, is a known limitation of statistical detection rather than evidence of wrongdoing. Knowing that in advance is what lets you respond calmly if a report surprises you.
The score is also time-sensitive. Detectors are retrained as models change, so a passage that scored low last semester may score differently today, and vice versa [3]. This is one reason a single score should never be treated as a permanent property of your writing. It describes one document, checked at one moment, against one version of a model.
Can You Check Your Own Draft for AI-Generated Text Before Submitting?
Yes — and doing so is the single most useful habit in this whole process. Running a draft through a detector before the deadline turns an opaque institutional judgement into something you can act on while you still have time to revise [4]. The value is diagnostic: you see which passages carry the pattern, then decide whether the wording genuinely reflects your thinking or drifted into generic phrasing during editing.
The mechanism matters for anyone worried about self-checking. Non-repository checks evaluate a file without adding it to Turnitin's student paper database, so a draft check does not create a future similarity match against your own final submission [4]. Checking early also avoids the worst-case scenario, which is discovering a flagged section after the submission window has closed and no revision is possible.
Policy is the caveat worth reading carefully. Institutions vary on whether students may use checking tools independently, and some treat repeated pre-submission checks as a misuse of the system [4]. Check your module handbook or ask your tutor before relying on any third-party service, and keep the purpose clearly diagnostic — revising your own prose, not reverse-engineering a threshold.
Where a draft does come back flagged, the fix is editorial rather than cosmetic. Rewrite the flagged passages in your own voice, add specific evidence and reasoning that a model would not have produced, and re-check [4]. That sequence — check, revise, re-check — is what converts a score from a source of anxiety into a working tool.
If you want that same check, revise, re-check loop available before your deadline rather than after it, turnitin0 produces the AI detection report and the similarity report together in one order — the same two PDFs your tutor sees in the LMS — so you can read the flagged passages side by side and revise with the actual report in front of you.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a high AI score proof that text was generated by AI?
No. Turnitin presents the indicator as a review prompt for educators rather than standalone proof of misconduct [2]. Highly formulaic human writing can trigger the same patterns, which is why flagged passages are meant to be read in context.
Why does my report show *% instead of a number?
*% appears when the AI signal falls below Turnitin's confidence threshold, so the system declines to report a precise percentage [3]. It indicates a low-confidence result, not a missing or hidden score.
Does checking a draft add my work to the student paper database?
Not with a non-repository check, which evaluates the file without adding it to Turnitin's student paper database [4]. Always confirm the repository setting on whichever service you use before uploading.
Can I lower a score by paraphrasing flagged sentences?
Paraphrasing alone often preserves the same predictable pattern. Rewriting flagged passages in your own voice and adding specific evidence and reasoning is the more reliable revision route [4].
Do all universities treat pre-submission self-checks the same way?
No. Policies vary, and some institutions restrict independent checking [4]. Check your module handbook or ask your tutor before relying on any third-party service.