Direct answer
An AI test is an automated check that estimates how much of a piece of writing was likely produced by a generative AI model rather than a person. In universities it usually means a Turnitin AI writing detection scan, which splits a submitted document into segments, scores each one, and reports the share of qualifying text that looks machine-generated [1]. It is a screening signal for instructors, not a verdict, and it is read alongside a similarity report and the student's own drafting history [1].
How Does an AI Test Actually Detect AI-Written Text?
An AI test does not search for a fingerprint hidden inside the text. The detector evaluates the statistical texture of the prose — how predictable each next word is given the words before it — and classifies segments of the document on that basis, with only segments above an internal confidence threshold contributing to the final figure [2]. Because the judgement is made segment by segment, a single document can come back with a mixed picture: some paragraphs reading as clearly human, others reading as machine-like [2].
This is why the model is described as a probability estimate rather than a measurement. Turnitin's own documentation notes that the percentage reflects the proportion of qualifying words flagged, not a statement that a specific sentence was written by a specific tool [2]. The detector is also scoped to long-form prose; short answers, reference lists, and code sit outside what it was built to assess [1].
Practically, that means the AI test is sensitive to style rather than to provenance. Heavily formulaic academic register, repeated sentence templates, and uniform paragraph rhythm can all push a segment toward the flagged side, even when the drafting was human [2]. Conversely, a student who rewrites AI-drafted passages in their own voice often moves the number substantially without changing a single citation.
The honest framing for students is that the AI test answers a narrow question — "does this text read like a language model wrote it?" — and leaves the wider question of authorship to the people who set the assignment [1].
What Do the AI Scores and Percentages in an AI Test Report Mean?
The headline number is a share, not a grade. If a report shows 34%, the reading is that roughly a third of the qualifying text in that document was flagged as likely AI-generated — not that the student is 34% guilty, and not that 34% of the assignment should be penalised [3]. Turnitin displays an asterisk instead of a number when the detected share falls below its low-confidence threshold, because at that level the signal is too weak to report as a figure [3].
The most common misreading is to compare the AI percentage with the similarity percentage. The two measure unrelated things: similarity compares the text against a database of sources and publications, while the AI score compares the text against a model's expectation of machine-like phrasing [3]. A paper can show 28% similarity purely from correctly cited quotations and 0% AI, or the reverse, and neither number explains the other [3].
Context matters as much as the digits. A high similarity match to a properly referenced journal article is normal academic practice; a low similarity score on a paper with no references is not automatically reassuring [3]. Instructors are trained to read the two reports together and to weigh them against drafts, version history, and any conversation they have had with the student [3].
For students, the useful takeaway is to treat any single number as one data point. If a report surprises you, the productive response is to open the flagged passages and look at what the detector reacted to — usually a run of predictable, evenly paced sentences rather than the ideas themselves [2][3].
Can You Run an AI Test on Your Own Draft Before You Submit It?
You can, and many institutions now encourage it. Turnitin's guidance to teaching staff recommends discussing AI writing detection openly with students and, where policy allows, letting them see how reports work before final submission, because transparency reduces appeals and deadline panic [4]. The catch is access: the institutional checker most students rely on is tied to a course submission, so running a test usually means either submitting for real or using an independent pre-submission service [4].
Whether a pre-submission check is permitted is a policy question, not a technical one, and it varies by institution and by module. Some departments explicitly allow students to review their own reports; others treat any external check as outside the assessment process [4]. Checking your own handbook before you upload anything is the step most students skip and later regret.
Where pre-submission checking is allowed, the practical value is diagnostic rather than defensive. Reviewing flagged passages lets you see which of your own habits — a stock transition phrase, a repeated sentence shape, a paragraph that summarises rather than argues — are triggering the detector, so you can revise the prose yourself [4]. That is a genuinely different activity from trying to disguise text, and it is the version of the workflow that holds up if an instructor asks what you did.
One more detail worth knowing: the file you check should be the file you submit. Formatting, headings, and reference lists all affect how the document is segmented, so a test run on a rough draft tells you less than a test run on the near-final version [1][4].
If you want that pre-submission AI test to mirror what your marker will actually see, turnitin0 runs the same check on your draft and returns the two PDFs together — the AI detection report and the similarity report — usually in under 15 minutes. Files are checked without being added to Turnitin's student paper database, so nothing is stored against your name, and you can delete the upload afterwards. More than 100,000 reports have been delivered to 20,000+ students across the US, UK, Canada, Australia, New Zealand, and Ireland.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI test the same as a plagiarism check?
No. A plagiarism or similarity check compares your text against published sources and previously submitted papers, while an AI test estimates whether the phrasing looks machine-generated [3]. Turnitin reports them separately, and a document can score high on one and zero on the other.
What does an asterisk instead of a percentage mean?
It means the detected share of AI-like text fell below Turnitin's low-confidence threshold, so the system reports a low-confidence signal rather than an exact figure [3]. Treat it as "not enough evidence to show a number," not as a clean pass.
Can a human-written essay be flagged by an AI test?
It can, particularly when the prose is highly formulaic or has been heavily edited into a uniform register [2]. That is why reports are read alongside drafts and version history rather than acted on in isolation [1].
Does checking my draft first count as cheating?
That depends entirely on your institution's policy, which varies by department and module [4]. Reviewing your own report to find habits worth revising is different from disguising text, but confirm what your handbook permits before uploading anything.
Which file types can be tested?
Turnitin's detector is built for long-form prose documents such as.docx,.pdf, and.txt, and it does not assess short-form answers, reference lists, or code [1]. Submit the near-final version, since headings and structure affect how the document is segmented [1][4].