Direct answer
A content checker cannot prove who wrote a text; it estimates the likelihood that a passage was AI-generated by scoring statistical patterns in the writing. Turnitin's AI writing detection reports a percentage of qualifying text flagged as AI plus highlighted sentence-level segments, and it is explicitly designed as an indicator that prompts human review rather than a verdict of misconduct [1]. That is why a checker can say "82% AI" about a paragraph you typed yourself — and why the number should be read as a signal to investigate, not a finding.
How Does a Content Checker Decide Whether Text Is AI or Human?
Detectors do not "recognize" an AI author or match your draft against a database of chatbot outputs. They measure properties of the text itself. The two classic signals are perplexity — how predictable each word choice is given the preceding words — and burstiness, the variation in sentence length and rhythm across a passage. Human academic writing tends to be uneven and occasionally surprising; machine-generated prose tends to be smooth, evenly paced, and statistically unsurprising. Turnitin's model segments a document and scores sentence-level spans, then aggregates those spans into an overall percentage of AI-generated qualifying text [2].
The scoring is deliberately scoped. Only "qualifying text" — continuous prose of sufficient length — is evaluated, which is why reference lists, block quotations, tables of contents, and short fragments are excluded from the percentage [2]. A 30% score therefore does not mean 30% of your document is fake; it means 30% of the qualifying prose looked machine-like to the model.
The output is probabilistic by design. A high score means "likely AI-generated," and Turnitin frames the report as evidence for an instructor's judgement rather than an automatic penalty [1]. The company also states plainly that its detector cannot definitively determine authorship, and that results must be interpreted with context about the assignment, the student, and the drafting process [1]. In practice this means two drafts with identical scores can deserve completely different conclusions.
Finally, the tool communicates uncertainty rather than hiding it. When the AI score falls below Turnitin's confidence threshold, the report shows *% instead of an exact figure — a low-confidence signal that the model is not willing to commit to a number [1]. Reading *% as "zero" or as "caught" are both misreadings; it is the detector declining to make a strong claim.
Why Do AI Detectors Flag Human Writing as AI?
False positives are not random noise — they cluster around specific writing habits. Formulaic, highly structured, or heavily edited prose carries many of the same statistical traits detectors associate with machine output: uniform sentence length, predictable transitions, and low lexical surprise. A student who writes to a rigid essay template, or who runs their own draft through a grammar tool until every sentence is polished flat, can accidentally produce text that scores like a chatbot's [3].
The risk is unevenly distributed. Non-native English writers, students working from provided scaffolds, and anyone who used AI for proofreading or paraphrasing rather than generation are disproportionately likely to be flagged, because the editing process itself removes the irregularities detectors read as human [3]. This is the single most common reason a genuinely hand-written submission comes back with an alarming number.
Accuracy also varies with the material. Detector performance shifts with discipline, text length, and whether the draft was paraphrased, translated, or assembled from multiple sources — a short reflective paragraph is far harder to classify than a 3,000-word literature review [3]. A score is therefore not a portable fact about your writing ability; it is a measurement taken under particular conditions.
Because of this, Turnitin advises institutions to treat the score as one data point among several and to discuss flagged work with the student rather than act on the number alone [3]. If you are ever flagged, the productive response is process evidence — drafts, version history, notes, source files — not an argument about the percentage.
How Can I Check My Own Draft's AI Score Before Submitting It?
Most students cannot run a self-check inside their institution's account. The similarity and AI reports are generated when work is submitted to the course assignment, so there is no button in the student view that returns a report on demand [4]. That gap is exactly why pre-submission checking services exist, and why students search for one the night before a deadline.
An independent pre-submission check works by running your file through the same detection pipeline and returning the reports in Turnitin's format — an AI detection report and a similarity report — without adding your document to the student paper repository [4]. That non-repository detail matters: a draft checked this way will not later match itself as 100% plagiarised when you submit the real assignment. Turnitin0, for example, returns both PDFs in one checkout for English .docx, .pdf, or .txt files between 300 and 30,000 words, typically in under 15 minutes, and does not share files with third-party databases.
Re-checking after revision is the part students skip. Scores move with every edit, so a single reading tells you where you stand at that moment, not where you will stand at submission [4]. The practical loop is: check, revise the specific flagged spans, then check again to confirm the number actually dropped — otherwise you are guessing about the effect of your own edits.
One caveat worth stating plainly: institutional policies differ on whether pre-submission checks are permitted, and some departments treat them as a form of academic misconduct [4]. Confirm your own course rules before you upload anything, and treat the report as diagnostic information about your draft rather than a target to be gamed.
Reading a percentage is one thing; seeing the actual report — the cover, the score band, the highlighted sentence-level flags, and the similarity summary side by side — is what tells you whether your draft is genuinely at risk. That is precisely what turnitin0 returns: the same AI and similarity reports your professor sees, on your own file, before the deadline rather than after it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a content checker accurate enough to prove I used AI?
No. Turnitin describes its AI writing detection as an indicator of likelihood, not proof of authorship, and the report is built to support human review [1]. Any decision made on the percentage alone is being made on incomplete evidence.
Why does my own writing show a high AI score?
Usually because the prose is unusually uniform — template structure, heavy proofreading, or translation can all flatten the variation detectors look for [3]. Non-native English writers are flagged more often for exactly this reason.
What does an asterisk score like *% mean?
It means the AI score fell below Turnitin's confidence threshold, so the tool reports a low-confidence signal instead of an exact percentage [1]. It is not zero and it is not a detection.
Can I check my draft before I submit it to my university?
Not through your institution's student account, which only generates reports on submitted coursework [4]. Independent pre-submission services return Turnitin-format AI and similarity reports on a draft instead — subject to your course's own rules.
Do flagged sentences mean I will be penalised?
Not automatically. Because detector accuracy varies by discipline and text length, institutions are advised to weigh the score alongside discussion and process evidence rather than act on it in isolation [3].