Direct answer
A test for AI writing is a check that scans a document and reports what share of its text a detector classifies as machine-generated. Turnitin's AI writing detection runs inside the same report workflow as its similarity check, highlights the flagged segments, and returns a percentage rather than a verdict [1]. The result is a probability signal for a human to interpret, not proof that a student used ChatGPT, Claude or Gemini [1]. Because the score depends on the exact text submitted, the only reliable test is one run on the final draft in the file format your institution accepts [1].
How Do You Test a Piece of Writing for AI Detection?
Testing a piece of writing for AI detection means separating two questions that students routinely blur: what did the detector flag and what does that flag imply. Turnitin's reports are built to highlight matched or flagged segments in context rather than to issue a judgement about intent [2]. That design choice matters when you are testing your own draft, because a highlighted paragraph is a prompt to look closer, not a conclusion about authorship [2].
A practical test therefore has three parts. First, run the document through a detector that produces a segment-level view, so you can see where the signal sits instead of only a headline number. Second, read the flagged passages against your own drafting history — did that section come from an outline you expanded, a paragraph you pasted and lightly edited, or prose you wrote from scratch? Third, decide what to change based on the passages, not the total. A single flagged sentence in a 3,000-word essay and a fully flagged methods section call for very different responses.
The format of the test matters as much as the tool. Detectors parse text extracted from a file, so a .docx with tracked changes, embedded tables, or unusual styling can be read differently from a clean export. Turnitin supports a defined set of file types and languages for AI detection, and text outside that envelope is simply not assessed [1]. If your draft contains heavy quoting, appendices, or reference lists, expect the reported percentage to reflect whatever the detector counted as qualifying text — which is one reason a raw number can look alarming next to a document that is mostly bibliography.
Finally, treat any single test as one data point. Detectors are probabilistic classifiers: the same passage can score differently across tools, versions and runs, and Turnitin itself frames its output as an indicator that supports academic judgement rather than replacing it [2]. Running the test early, then again after revision, tells you far more than a single reading taken minutes before a deadline.
What Does the Turnitin AI Score Actually Mean, and What Counts as Flagged?
The AI score in a Turnitin AI Writing Report is a percentage of the qualifying text in the submission that the detector classifies as AI-generated — not a percentage of the whole document, and not a confidence rating about the author [3]. Turnitin reports that figure alongside segment-level highlighting, so the headline number and the highlighted passages should always be read together [3]. A high score with a handful of scattered highlights and a moderate score with one dense block are different problems.
Low-confidence output is displayed differently. When the detector's signal falls below its confidence threshold, Turnitin shows an asterisk in place of an exact percentage, signalling that the result is not strong enough to report as a number [3]. Students often misread that asterisk as an error or a hidden score; in practice it is the detector declining to overstate what it found. On the other side of the ledger, false positives are a documented possibility, particularly with highly formulaic prose, heavily templated academic phrasing, or text that has been edited so many times that its original rhythm is gone [3].
"Flagged" is therefore a narrower idea than "AI-written." A segment is flagged when the detector's model assigns it a high enough probability of machine generation to surface it in the report; whether that flag becomes an integrity conversation depends on your institution's policy, your module handbook, and the judgement of the marker [3]. Two students with identical scores can face completely different outcomes because the surrounding evidence — drafts, version history, supervision notes — differs.
The most useful way to read a score is comparatively. Test a draft, revise the passages that carry the signal, test again, and watch which segments move. A number that falls as you rewrite in your own voice is telling you something real about your text; a number that stays flat no matter what you change is telling you something about the limits of the detector [3].
Can You Test Your Own Draft Before Submitting It to Turnitin?
Yes — and Turnitin's own guidance points toward transparency rather than secrecy around AI indicators. The company recommends that institutions approach AI writing detection formatively, discussing with students what the report shows and how it should be interpreted, instead of treating the score as a purely disciplinary instrument [4]. That framing supports a pre-submission self-check: understanding how detection works is treated as part of learning to write with these tools responsibly, not as an attempt to game them [4].
There is, however, an important distinction between checking your own work and accessing the institutional system. The AI Writing Report is generated inside an assignment submission in an institution's Turnitin account, so a student cannot simply log in and generate one for a draft that has not been submitted. Any external check you use is a preview of the same underlying signals, not the institution's own record — which is exactly why the output should be read as an early warning rather than a final verdict [4].
Policy is the other half of the answer. Turnitin is explicit that acceptable use of AI, and the consequences of a flagged report, are set by each institution rather than by the detector [4]. Before you build a testing habit, check your module handbook for rules on AI assistance, on draft checking services, and on what you are expected to declare. A test that helps you revise is very different from a test used to reverse-engineer a threshold — and only one of those is defensible if a marker asks how the final text was produced.
The practical habit that survives all of this: test early enough that a flag still changes what you write. A report you receive five minutes before a deadline can only cause panic; a report you receive two days before it can send you back to the paragraphs that need your own thinking in them [4].
If you would rather see the actual report than guess at the number, turnitin0 runs the same Turnitin AI and similarity check on your own draft and returns both PDFs together — typically in under 15 minutes, without adding your file to Turnitin's student paper database. You upload the document you are about to submit, and you read the same cover page, score, flag highlights and similarity summary your marker would see.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a test for AI writing the same as a plagiarism check?
No. A similarity check compares your text against a database of sources and highlights matches, while an AI writing test estimates whether text was machine-generated [1]. Turnitin reports both, but they answer different questions and should be read separately.
Why does my AI score show an asterisk instead of a number?
An asterisk means the detector's signal fell below its confidence threshold, so it declines to report an exact percentage [3]. Treat it as a low-confidence indicator rather than a zero or an error.
Can a detector flag text I wrote myself?
Yes, false positives occur, particularly with formulaic or heavily templated academic phrasing [3]. That is why Turnitin describes its output as an indicator supporting human judgement rather than proof of misconduct [2].
Does testing my draft guarantee the same result when my university runs it?
No. Scores depend on the exact file, the text counted as qualifying, and the detector version, so a preview is an early warning rather than a guarantee [1]. Use it to revise, not to predict a final number.
Will a draft check get my file added to Turnitin's student paper database?
Only if the check is run in repository mode. turnitin0 checks without adding your file to Turnitin's student paper database, and you can delete files from your account afterwards.