Direct answer
An artificial intelligence tester is a detection tool that analyses a piece of writing and estimates how much of it was likely produced by a generative AI model such as ChatGPT, Claude, or Gemini. It does not read the author's mind or recover a hidden "AI watermark"; it measures statistical properties of the prose — chiefly how predictable each word is given the words before it — and reports the share of qualifying text that matches the pattern of machine generation [1]. Turnitin's detector, the version most universities actually run, scores documents in segments of roughly 300 words and only assesses long-form English prose, so short answers, reference lists, and code are outside its scope [1]. The output is a probability estimate for human review, not a verdict of misconduct [1].
How does an artificial intelligence tester decide whether text was written by AI?
Detection rests on the fact that language models generate text by repeatedly choosing the most probable next token. Human academic writing is comparatively erratic: it swings between dense technical phrasing, personal hedging, awkward transitional sentences, and the occasional odd word choice. An AI-generated passage tends to stay inside a narrower band of predictability across hundreds of words, and that sustained uniformity is the signal a tester is looking for [2]. This is why detectors evaluate passages rather than isolated sentences — a single polished sentence proves nothing, but a page of uniformly smooth, evenly weighted prose is statistically unusual [2].
Turnitin's model produces a document-level percentage that represents the proportion of qualifying text flagged as AI-generated, not a confidence score about authorship [2]. Because the calculation is bounded by which text qualifies, a paper with heavy quoting, tables, or short-answer structure can show a lower percentage simply because less of it was assessed [2]. Understanding that denominator matters more than fixating on the headline number.
The practical consequence students miss is that light editing rarely changes the result. Swapping a few synonyms, reordering clauses, or adding a transition sentence leaves the underlying predictability pattern largely intact, so the flagged segment keeps scoring as AI [2]. What does move the score is substantive rewriting: introducing genuine specificity, first-person reasoning, discipline-specific evidence, and the kind of uneven emphasis that comes from actually thinking through a problem [2].
Finally, every credible tester carries a false-positive caveat, and Turnitin states its own openly: the score is an indicator that supports academic judgement, not a substitute for it [2]. Highly formulaic human writing — lab reports, standard methodology sections, non-native-speaker prose drilled in template patterns — can read as machine-like. That is why institutional policy normally pairs the report with draft history, supervision, and a conversation with the student rather than acting on the number alone [2].
What does an AI detection report actually show, and what does the score mean?
The report is a document, not a verdict, and it is structured to be read alongside the submission itself. Turnitin's AI Writing Report displays an overall percentage, colour-codes it into bands, and overlays highlights on the exact segments the model flagged, so a reader can see where the signal came from rather than only how large it was [3]. A single flagged paragraph in an otherwise unflagged essay tells a very different story from uniform highlighting across every section.
One display detail causes more confusion than any other: when the detector's confidence falls below its threshold, the report shows an asterisk in place of an exact figure — rendered as *% — rather than a precise number [3]. Students often read this as "score hidden" or "score zero." It is neither. It is a low-confidence signal meaning the model found some machine-like patterning but not enough to commit to a percentage, and it should be interpreted with corresponding caution [3].
Reading the report well means reading three things together: the percentage, the distribution of highlights, and the nature of the assignment [3]. A 100% score on a 400-word reflective paragraph that was drafted in a chat window is a very different situation from a 22% score concentrated in a standard definitions section that any student would write the same way. Instructors are explicitly guided to weigh the score against context, earlier drafts, and the student's own account of how the work was produced [3].
The similarity report is a separate document and answers a separate question. Similarity measures overlap with existing published and submitted sources; the AI score measures machine-generation patterning. A paper can have 0% similarity and a high AI score, or 30% similarity from properly cited quotations and no AI signal at all [3]. Treating the two as one number is one of the most common misreadings in pre-submission panic.
Can you test your own draft with Turnitin before you submit it?
Not through your university account. Institutional Turnitin access is provisioned for instructors and administered through the LMS; students submit into an assignment that the instructor created, which means there is no general-purpose "check my draft" button available to you before the deadline [4]. Any site promising a login to "your university's Turnitin" is describing something that does not exist in the institutional product.
What does exist is independent, non-repository pre-submission checking. A non-repository check runs the same class of analysis without adding your file to Turnitin's student paper database and without sharing the report with third-party databases, so the draft you test does not later collide with your own final submission as a self-match [4]. This is the mechanism that makes previewing safe: the point of a pre-submission test is to see the report, revise, and then submit cleanly.
The workflow that actually helps is sequencing, not scraping for a magic number. Run the check early enough that you still have time to rewrite flagged segments properly — not paraphrase them — and re-check after revision to confirm the signal dropped [4]. Because flagged passages need substantive rewriting rather than synonym swaps, the earlier you see the highlights, the more realistic that revision is.
Turnitin0 exists for exactly this step: an independent service, not affiliated with Turnitin, LLC, that returns a Turnitin AI detection report and a similarity report as two downloadable PDFs from a single check, formatted the way instructors see them in the LMS. Documents must be English, between 300 and 30,000 words, under 20 MB, in .docx, .pdf, or .txt; most orders complete in 5–15 minutes, with delivery guaranteed inside 30 minutes even during queue spikes. Checks are pay-per-use with no subscription — $3.80 for a single check, or $2.75 each in a 10-check pack — and the service has delivered 100,000+ reports to 20,000+ students across the US, UK, Canada, Australia, New Zealand, and Ireland.
Knowing how the tester scores your draft is only useful if you can see your own report before the deadline — and that is the gap turnitin0 closes, returning the same AI and similarity PDFs your instructor would open, so you revise against real highlights instead of guessing.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an artificial intelligence tester the same as a plagiarism checker?
No. A plagiarism or similarity checker compares your text against published and previously submitted sources, while an AI tester estimates whether the prose itself was machine-generated [3]. Turnitin reports them as two separate documents because they answer two separate questions.
Can an AI detector prove I used ChatGPT?
No. Detection is probabilistic and Turnitin presents the score as an indicator for human review rather than proof of misconduct [2]. That is why institutions pair the report with draft history and discussion rather than acting on the number alone.
Why does my report show *% instead of a number?
That display indicates the result fell below the detector's confidence threshold, so it is reported as a low-confidence signal rather than an exact percentage [3]. It is not a zero and not a hidden score.
Does editing a few words lower the score?
Usually not. Light synonym swaps and reordered clauses leave the predictability pattern largely intact, so substantive rewriting of flagged segments is what actually changes the result [2].
Will testing my draft get it added to Turnitin's database?
Not with a non-repository check, which does not add the file to the student paper database or share reports with third-party databases [4]. That is what allows you to revise a draft and still submit it as your own original work.