Direct answer
A "rilevatore AI testo" is a tool that estimates how much of a text was generated by an AI model rather than written by a person. It does this by scoring statistical patterns in the writing — chiefly perplexity and burstiness — and then reporting the share of the document that looks machine-generated [1]. Turnitin's AI writing detection works this way: it reads the text only, flags segments that resemble AI output, and expresses the result as a percentage of the qualifying text [1]. The number is a signal about the writing, not a verdict about who wrote it.
How Does an AI Text Detector Decide Whether a Piece of Writing Was Generated by AI?
Detectors do not "recognize" AI writing the way a person recognizes a friend's handwriting. They measure how predictable and how uniform the text is. Perplexity captures how surprising each next word is to a language model, and burstiness captures how much sentence length and structure vary across a passage [2]. Human drafts usually wobble: a long clause, then a short blunt sentence, then a fragment. Machine output tends to be smoother and more even, which is exactly what these two signals pick up [2].
Turnitin applies that logic at scale. It splits the submitted document into overlapping segments, scores each window, and then aggregates the window-level results into a single document-level percentage [2]. That is why a report can show "12%" rather than a simple yes or no — the figure describes the proportion of qualifying text that the model flagged, not the probability that the student cheated [1].
Two practical consequences follow. First, detection depends on the text itself, so paraphrasing a few words rarely changes the underlying statistical profile [2]. Second, because the analysis is text-only and language-limited, formatting, images, and non-supported languages simply fall outside the score [1]. A high percentage therefore points to a passage worth rewriting, not to a person worth accusing.
Why Does Turnitin Sometimes Show *% Instead of an Exact AI Percentage on a Text?
Turnitin does not publish a precise number when its signal is weak. If the detected AI-writing proportion sits below the tool's confidence threshold — 20% — the report displays *% in place of a figure [3]. The asterisk is a confidence marker, not a hidden score: it tells the reader that the detector saw some AI-like patterns but not enough evidence to stand behind an exact percentage [3].
This matters because *% is easy to misread in both directions. Students sometimes assume it means "0% — you're safe," while instructors sometimes treat it as a soft accusation. Neither reading is supported. Turnitin's own guidance is that low-confidence results should be handled cautiously and never presented as proof of misconduct [3]. A *% result is best understood as "inconclusive — the text did not trigger a confident detection."
From a revision standpoint, the practical question is whether the flagged passages are passages you can defend. If a paragraph was drafted by ChatGPT, Claude, or Gemini and then lightly edited, the statistical smoothness often survives the edit, and the score can move back above the threshold on a re-check [3]. If the paragraph is genuinely your own reasoning with your own citations, a low-confidence reading is usually the correct one and no rewrite is needed.
Can a Student Preview the Turnitin AI Detection Result for a Text Before Submitting It?
Not always through official channels. Student visibility into AI detection depends on how the instructor has configured the assignment, and many institutions restrict or disable student preview of AI results entirely [4]. Where preview is allowed, it is typically tied to the institutional Similarity report rather than a full AI-detection view [4]. So the honest answer is: sometimes, and you cannot rely on it being available when your deadline arrives.
That gap is why independent pre-submission checks exist. A useful pre-submission check has three properties: it returns the same two reports an instructor would see (AI detection and similarity), it completes fast enough to be useful before a deadline, and it is non-repository so the draft is not added to the student paper database [4]. Turnitin0 is built around exactly that workflow — you upload a .docx, .pdf, or .txt file, and you receive a Turnitin AI detection report plus a similarity report you can download and read before you submit.
Reading the report correctly is the other half of the job. Look at which passages were flagged, not just the headline number, because a document-level percentage can be driven by one long AI-drafted section while the rest of the paper is clean [4]. If the flagged text is yours, you can leave it and cite your process. If it is not, the flagged passages are the ones to rewrite in your own voice before the real submission.
If you would rather see the actual report before your professor does, turnitin0 lets you run the same AI and similarity check on your draft and read the flagged passages yourself — so the only version your instructor ever sees is the one you have already reviewed.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a high AI detection percentage mean I cheated?
No. The percentage describes the share of text segments that matched AI-like statistical patterns, not your intent [1]. Turnitin itself frames the result as a signal that should be interpreted alongside context, and low-confidence results in particular should not be treated as evidence of misconduct [3].
Can I lower an AI score just by swapping synonyms?
Usually not meaningfully. Detectors score perplexity and burstiness across segments, so surface word swaps leave the underlying rhythm intact [2]. Changing sentence length, adding your own analysis, and restructuring arguments does more than replacing individual words.
Is Turnitin's AI detection available for every language and file type?
No. The detector is text-only and limited in the languages and file types it supports, which is why non-English or image-heavy submissions may return no AI signal at all [1]. Check the supported set before assuming a blank result means a clean one.
Will a pre-submission check put my draft into the student paper database?
That depends on the service. A non-repository check does not add your file to Turnitin's student paper database and does not share the report with third-party databases, which is the behavior you want for a draft you intend to revise and submit [4].
What should I do if only one section of my paper is flagged?
Treat the flagged section as the target. A document-level percentage can be driven by a single long AI-drafted passage, so read the report at the segment level and rewrite only what needs rewriting [4].