Direct answer
An AI check is a pre-submission scan that estimates how much of your document reads as machine-generated, and Turnitin reports that estimate as a percentage of qualifying prose rather than a verdict on your intent. The number is produced by segmenting your text and classifying each segment, so a single score can sit alongside paragraph-level highlighting that shows exactly where the signal came from [1]. Understanding that mechanism matters more than chasing a single figure, because a high score on a formulaic methodology section and a high score on a hand-written reflection mean very different things.
What Does an AI Check Actually Measure, and What Do the Score Bands Mean?
Turnitin's AI writing detection does not read your document as one block of text. It splits the submission into segments, evaluates them against patterns associated with large language model output, and then aggregates the results into an overall percentage that represents the share of qualifying prose flagged as AI-generated [2]. That word "qualifying" carries a lot of weight: certain content is excluded from the calculation entirely, so the denominator is not simply your total word count. The practical consequence is that a 40% score on a 2,000-word essay does not mean 800 words were written by a chatbot — it means 40% of the scorable text carried the signal.
The report pairs the headline number with per-paragraph highlighting, and this is where the diagnostic value lives [2]. A student who sees three highlighted paragraphs inside a literature review can go and read those paragraphs, recognise the polished-but-generic phrasing that triggered the classifier, and rewrite the actual problem area. A student who only looks at the aggregate figure learns nothing actionable and typically responds by rewriting sections that were never flagged in the first place. Read the highlights first, then the number.
It is also worth being precise about what the score is not. Turnitin positions its AI writing indicator as an indicator — a signal that warrants human review — rather than proof of misconduct [2]. Formulaic academic conventions, heavily templated lab reports, and prose that has been through multiple rounds of grammar polishing can all produce elevated scores from genuinely human writing. Institutions are expected to apply academic judgement to the report, not to treat the percentage as an automatic finding. If you are ever asked to discuss a report with a tutor, the strongest thing you can bring is your drafting history and your sources.
Finally, the score is model- and version-sensitive. Detection systems are updated, and the same text can score differently across releases or across different tools. That instability is a feature of probabilistic classification, not evidence that one tool is "lying." It is the reason a check is best treated as a snapshot taken close to your submission date rather than a permanent property of your document.
Why Does Turnitin Show an Asterisk (*%) Instead of an Exact AI Percentage?
When Turnitin displays an asterisk in place of a number, it is telling you that the detected AI signal fell below its confidence threshold for reporting a precise figure [3]. The threshold sits at 20% confidence: below that point the system will not commit to a numeric percentage, because the underlying evidence is weak enough that a specific number would imply more certainty than actually exists. Crucially, the asterisk is not the same as a clean zero. It represents a low-confidence, non-zero signal — the system saw something, but not enough to quantify responsibly [3].
This distinction trips up a lot of students. Seeing *% and concluding "so I'm fine" is a reasonable reading but an incomplete one, because the asterisk is a statement about measurement confidence rather than about your writing. The more useful interpretation is: the signal is weak, and weak signals are exactly what you would expect from short passages, technical vocabulary, or text that has been heavily edited toward uniformity. The same logic explains why some content is not scored at all — very short submissions and non-English text fall outside the scope of the detector, so their absence from the report is a coverage gap rather than a clean result [3].
There is a second-order effect worth knowing about. Because the asterisk hides the number, a reader comparing two drafts cannot tell whether one is at 3% and the other at 19% — both render identically. If you are tracking whether your revisions are working, that compression destroys the very signal you need. The workaround is to look at the highlighted regions rather than the headline figure: if the flagged paragraphs shrink between drafts, your edits are landing, regardless of what the summary row says [3].
Can I Check My Own Draft With Turnitin Before I Submit It?
Turnitin does not provide a public self-service check for students. Access to the AI writing report is provisioned through an institution's learning management system, which means a student account only exists because a course or assignment has been set up by an instructor [4]. There is no open upload page where any student can run a draft and download the resulting report — that gating is deliberate, and it is why "free Turnitin checker" pages you encounter are almost always either a different product wearing the name or a similarity-only tool that says nothing about AI writing.
What Turnitin itself recommends is process rather than preview: keep your drafts, keep your revision history, and be transparent with your instructor about how you used AI tools in your workflow [4]. That advice is sound and it is also the best defence available if a report is ever questioned, because a document's edit trail is far more persuasive than any single detection score. The limitation is that this advice does not tell you, before you press submit, whether your current draft is going to come back highlighted.
That gap is what independent pre-submission checking services exist to fill. The important thing is to be clear-eyed about what you are buying: a preview of the same report format your instructor sees, generated without adding your file to a student paper repository, so that a check does not itself create a similarity match against your own later submission. Students who check early enough to act on the result get the diagnostic value described above; students who check the night before a deadline get a number and no time to use it.
If you want that preview before your deadline rather than after it, turnitin0 runs the check on your .docx, .pdf, or .txt draft and returns both the AI detection report and the similarity report as downloadable PDFs — the same two documents your tutor would see in the LMS. Turnaround is under 15 minutes in 98% of cases, the file is never added to Turnitin's student paper database, and there is no subscription to cancel afterwards. Students who upload a draft with a week to spare can use the paragraph highlights to fix the specific passages that triggered the classifier, then re-check and confirm the flags have cleared.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI check the same as a plagiarism check?
No. A similarity check compares your text against source databases and previously submitted papers, while an AI check estimates whether the prose itself reads as machine-generated. They run as two separate reports against the same file, which is why a draft can show 0% similarity and still carry an AI writing score [1].
Does a high AI score prove I used ChatGPT?
It does not. Turnitin describes its AI indicator as a signal requiring human review rather than a finding of misconduct, and formulaic or heavily polished human writing can trigger it [2]. The report's paragraph highlights are the evidence you would discuss with an instructor, not the aggregate percentage alone.
Why did my score change when I resubmitted the same text?
Detection models are updated over time, and classification is probabilistic, so identical text can score differently across releases or tools. Treat a check as a snapshot taken near your submission date rather than a fixed property of the document.
What does *% mean on an AI report?
It means the AI score fell below Turnitin's 20% confidence threshold, so the system withheld an exact number [3]. It is a low-confidence, non-zero signal — not a clean zero — and very short or non-English text is excluded from scoring altogether.
Can I check my draft without my university knowing?
Turnitin's own reports are provisioned through an institutional LMS, and there is no public self-service upload for students [4]. Independent checking services exist precisely to give students a pre-submission preview without routing the file through their course.