Direct answer
Direct Answer - ** A ChatGPT detector works by scoring how predictable your text is: it measures perplexity (how likely each next word is) and burstiness (how much sentence length and structure vary), then reports the percentage of qualifying text that looks machine-generated [1]. Turnitin's AI writing indicator only evaluates long-form, prose-style English, and any result under its 20% confidence threshold is displayed as *% rather than an exact figure [1]. Crucially, a detection score is a statistical signal, not proof of authorship — which is why the same paragraph can be flagged on one draft and cleared on the next [2].
How Do ChatGPT Detectors Decide Whether Text Is AI-Generated?
Detectors do not keep a database of ChatGPT sentences to match against. Instead they model language statistically: a large language model produces text by choosing high-probability continuations, so AI prose tends to be low perplexity — each word is the obvious next word — and low burstiness, with sentences of similar length and rhythm [2]. Human writing is measurably messier: it mixes a five-word sentence with a thirty-word one, repeats itself, and takes detours that a model would rarely choose [2]. That difference in variance is the core signal, and it is why detectors can flag text they have never seen before.
The second stage is segmentation. Rather than returning one number for a 3,000-word essay, the tool divides the document into passages, scores each one, and reports the share of qualifying text that crosses its threshold [2]. This is why a report often shows a headline percentage alongside highlighted segments — the percentage is an aggregate, and the overlay is where the actual evidence lives [1]. It also explains a common confusion: a document can be 30% flagged while the flagged passages are concentrated in one section that was drafted differently from the rest.
Finally, detectors are scoped. Turnitin states that its AI writing feature is designed for long-form, prose-style English and is not reliable on short answers, bullet lists, code, or heavily technical text [1]. Feed a detector 80 words and you will get noise; feed it 1,500 words of flowing argument and you will get something meaningful [1]. Reading a detector result without knowing that scope is the single most common way students misjudge their own risk [2].
Why Does Text You Wrote Yourself Sometimes Get Flagged as ChatGPT?
Because detectors measure style, not origin, and human academic writing is often stylistically identical to machine output. The classic essay template — a definitional opening, three signposted body paragraphs, "Furthermore" and "Moreover" as connective tissue, a summary conclusion — is exactly the kind of high-probability, low-variance prose a detector associates with AI [3]. Students who follow the rubric closely are, statistically, writing more like a model.
Two further amplifiers matter. First, drafting tools: grammar correctors, paraphrasing apps, and translation tools smooth out the irregularities that make human text human, pushing perplexity down [3]. Second, revision: a draft rewritten three times for clarity becomes more uniform, and uniformity is the signal being measured [3]. This is why multilingual writers and heavily edited dissertations appear disproportionately in false-positive discussions — not because they used AI, but because their final text is unusually regular [3].
The threshold is the practical detail most people miss. Turnitin displays *% instead of a precise number when AI detection falls below its 20% confidence level, meaning the system is not confident enough to assert AI involvement [3]. A *% is therefore a weak signal, not a clean bill of health and not an accusation — and it should never be read as "0% AI" [1]. If your instructor sees *%, the honest interpretation is "no confident AI signal detected," which is a materially different statement [3].
How Can You Preview Your Turnitin AI Score Before You Submit?
Previewing is a two-part job: get the same report your instructor would see, then read it at segment level rather than headline level. Turnitin produces two separate outputs — a similarity (plagiarism) report and an AI writing report — and they answer different questions, so a clean similarity score tells you nothing about your AI score [4]. Requesting both is the only way to know where you actually stand.
When you open the AI report, work from the overlay inward. The flagged passages show you precisely which sentences triggered detection, which is far more actionable than the aggregate percentage [4]. If 22% of your text is flagged and all of it sits in one literature-review paragraph, you have a targeted revision problem, not a document-wide rewrite problem [4]. Segment-level reading also tells you whether the flags cluster around templated transitions — a fixable stylistic habit — or around genuinely generic phrasing [2].
Then re-check. Detection scores move when text moves, and the only way to confirm a revision worked is to run the revised draft through the same detector again [4]. Treat the first report as a diagnostic and the second as verification; students who skip the second step are guessing. Keep both PDFs, because comparing the before-and-after overlays shows you exactly which edits changed the signal [4].
Reading a report is straightforward once you have one in front of you — the harder part is getting a real Turnitin AI and similarity report on a draft that has not been submitted yet. That is what turnitin0 was built for: upload your .docx, .pdf, or .txt (English, 300–30,000 words, under 20 MB) and receive both PDFs — the AI detection report and the similarity report — in one checkout, checked without adding your file to Turnitin's student paper database. Most orders finish in 5–15 minutes, and 100,000+ reports have been delivered to 20,000+ students. A single check is $3.80, with no subscription.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does a ChatGPT detector prove I used AI?
No. Detectors produce a statistical confidence signal, not evidence of authorship, and Turnitin advises that results be read alongside context and professional judgement [1]. Treat a score as a risk indicator on your draft.
What does *% mean on a Turnitin AI report?
It means the AI detection fell below Turnitin's 20% confidence threshold, so an exact percentage is withheld [1]. It signals low-confidence detection rather than a confirmed zero [3].
Can human writing really be flagged as AI?
Yes. Formulaic, heavily edited, or tool-smoothed academic prose can score as low-perplexity text, which is why multilingual writers and repeatedly revised drafts show up in false positives [3].
Does a low similarity score mean my AI score is fine?
No. Similarity and AI writing are separate reports measuring different things, so you need to read both [4].
Can I check my draft before submitting it?
Yes — that is the point of a pre-submission check: you get the AI and similarity reports on your own draft before your instructor sees anything [4].