Direct answer
An artificial intelligence checker is a tool that scans a piece of text and estimates what share of it was likely produced by a large language model such as ChatGPT, Claude, or Gemini. It does this by measuring statistical properties of the writing — chiefly how predictable each next word is and how much sentence rhythm varies — then reporting a percentage of "qualifying text" it believes was machine-generated. Turnitin's own AI writing detection works this way and is built into institutional accounts, where the resulting AI Writing Report is visible to instructors rather than students [1]. A checker score is an indicator, not proof of authorship, and it should be read alongside the flagged passages rather than as a single verdict [1].
How do artificial intelligence checkers actually detect AI-written text?
Detectors are, at heart, probability engines. A language model generates text by repeatedly choosing a likely next token, so its output tends to sit in the high-probability region of the model's vocabulary distribution. A checker runs the submitted text back through a comparable model and measures perplexity — how surprised that model is by each word choice. Low perplexity, meaning the text was highly predictable, is one of the strongest single signals of machine generation [2].
The second signal is burstiness, which is the natural variance in sentence length and structural rhythm that human writers produce without thinking. People write a long, winding sentence and then a short one; they interrupt themselves, drop in an aside, or shift register mid-paragraph. Model output is measurably more uniform, and detectors score that uniformity as evidence of AI involvement [2].
Turnitin's implementation returns a document-level percentage representing the share of prose that met its detection criteria, and it highlights the specific sentences that drove the score so a reader can inspect them [1]. That segmentation matters: a 34% result usually means roughly a third of the qualifying text was flagged, not that a third of the ideas were stolen. It also means the score moves when you edit, because changing sentence structure changes the underlying statistics [2].
It is worth being honest about the limits. Formulaic academic writing — literature review scaffolding, methodology boilerplate, heavily templated lab reports — can score as machine-like even when a student typed every word, and non-native English writers are disproportionately affected because their prose is often more regular than a native speaker's [2]. Turnitin itself frames the report as a prompt for conversation rather than a disciplinary finding [1].
Why does an AI checker show an asterisk instead of a percentage?
When the detected AI share falls below Turnitin's 20% confidence threshold, the report displays *% in place of an exact number [3]. This is the single most misunderstood element of the whole system, and it causes a lot of unnecessary panic in the week before a deadline. The asterisk does not mean zero, and it does not mean the tool failed — it means the signal was too weak and too unstable for Turnitin to stand behind a precise figure [3].
The reasoning behind the threshold is statistical honesty. Very small percentages are the least reproducible part of any detector's output: a handful of reworded sentences can swing a 4% result to 12% and back again, so publishing a hard number there would imply a precision the model does not have [3]. Collapsing that range into *% communicates "low-confidence, low-level signal" without inviting students and instructors to over-read noise.
Institutional policy is where the asterisk gets its teeth. Because *% still indicates that some qualifying text was detected, an instructor can choose to treat a *% report as something worth discussing, and some departments do exactly that [3]. The practical reading is therefore: *% is good news relative to a double-digit score, but it is not a clean bill of health, and it is not the same as a confirmed 0%.
For anyone checking a draft, the useful takeaway is that a *% result tells you the detector found little to react to — it does not tell you the exact margin you have left. If you need to know how close to the threshold you are sitting, the only way to find out is to look at the flagged passages themselves and judge whether the phrasing is genuinely yours [3].
Can I check my own draft with Turnitin before I submit it?
Usually not through Turnitin directly. Whether a student can see a Similarity Report — and by extension the AI Writing Report attached to the same submission — depends entirely on whether the instructor has enabled student access for that assignment [4]. Many courses leave that setting off, which means the first time a student sees their AI score is after the work has already been graded. Where access is switched on, it typically comes through a dedicated draft-check assignment rather than the final submission box, so the report you generate there is not the one that counts [4].
This gap is the reason independent pre-submission checkers exist. Students who want to see the actual report before the deadline — the same score panel, the same flagged sentences, the same similarity matches a marker would see — use a third-party service that runs the document through Turnitin and returns the reports as downloadable PDFs. Turnitin0 does this without adding the file to Turnitin's student paper database, so a draft check does not create a match against your own later submission. Turnaround is under 15 minutes in 98% of cases, and each check returns both the AI detection report and the similarity report in one order.
The workflow that actually reduces risk is a loop rather than a single check. Run the draft, read the flagged sentences rather than the headline number, rewrite the passages that are genuinely templated or over-polished, then re-check to confirm the score moved. Students who repeat that cycle once or twice before submitting consistently end up with a far clearer picture than those who check once and guess at what the number meant [4].
If you have read this far, you probably already know which draft you want to run — and you probably want the real report, not a rough estimate from a free tool that never shows you the flagged sentences. Turnitin0 gives you the actual Turnitin AI and similarity reports on your own document, delivered as PDFs you can open and read line by line, so the number you see before the deadline is the number your instructor will see after it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an artificial intelligence checker accurate enough to rely on?
It is accurate enough to be useful and not accurate enough to be final. Detectors reliably separate clearly human prose from clearly model-generated prose, but they produce false positives on formulaic academic writing and on the work of non-native English writers [2]. Treat the score as a screening signal and always read the flagged sentences yourself.
What does it mean if my report shows an asterisk instead of a number?
It means the detected AI share fell below Turnitin's 20% confidence threshold, so the tool declined to publish a precise percentage [3]. The signal is low-confidence rather than absent — the asterisk is not the same as zero, and some instructors still review *% reports [3].
Will checking my draft count as a real submission?
Not with a non-repository checker. Turnitin0 runs the document without adding it to Turnitin's student paper database and does not share reports with third-party databases, so a draft check does not create a self-match against the version you later submit. Institutional draft-check assignments behave differently depending on how your instructor configured them [4].
Can I see my AI score before my instructor does?
Only if your instructor has enabled student access to reports for that assignment [4]. Many courses never turn that on, which is why students use an independent pre-submission check to see the same report format their marker will see.
Does a low AI score mean my writing will pass?
It means the detector found little machine-like text in the passages it evaluated [1]. It does not certify originality, and it does not replace the similarity report — which is why the two reports are best read together.