Is It AI or Human Text?

Detectors do not recognise "AI words." They measure how predictable a passage is. The two signals behind almost every commercial detector are perplexity — how surprised a language model is by each word choice — and burstiness — how much sentence length and structure vary across a document [2]. Human academic writing is uneven: a long qualified sentence followed by a short blunt one, an unexpected concrete example, a slightly awkward transition that reflects a real train of thought. Machine text tends toward a smooth, evenly weighted rhythm where every sentence is roughly the same length and every clause resolves the way the reader expects [2].

That is why the practical signals you can read yourself map onto the same two properties. Uniform paragraph lengths, repeated triadic lists ("clear, concise, and effective"), hedged generalities with no named source, and transitions that announce structure rather than carry meaning are all low-burstiness symptoms [2]. Conversely, specific dates, named studies, idiosyncratic examples, and sentences that occasionally break their own pattern raise both perplexity and burstiness. Turnitin's own documentation is explicit that the underlying model was trained to recognise patterns in how AI writing is produced, which is why formulaic human writing — lab reports, annotated bibliographies, templated cover letters — can score unexpectedly high [1][2].

The honest limit is that these are statistical properties, not authorship marks. A careful human editor who smooths a draft can reduce burstiness and push a genuinely human document toward a higher AI score, and a writer who deliberately varies output can lower one. Turnitin states plainly that its detection is an estimate and that no result should be treated as definitive evidence of misconduct on its own [1]. Anyone asking "is it AI or human" should therefore expect a confidence band, not a yes/no.

The score is useful as a signal and unreliable as a verdict, and the asterisk is the clearest illustration of why. When Turnitin's AI detection confidence falls below its 20% threshold, the report displays *% instead of a number, because the system does not have enough confidence to publish a figure [3]. Students often misread *% as either a clean pass or a hidden accusation. It is neither: it is a low-confidence signal, meaning the detector found some AI-like patterns but not enough to quantify them [3].

Above that threshold, the report separates a document-level percentage from sentence-level highlighting, and the two must be read together. A 30% document score built from a handful of scattered sentences tells a very different story from 30% concentrated in one continuous section — the first is often consistent with formulaic or heavily edited human prose, the second with a generated block that was pasted in [3]. Turnitin's guidance to instructors stresses that the report is designed to prompt a conversation and a closer look at the flagged passages, not to function as a disciplinary finding [3].

For a student reading their own report, the practical rule is to treat the score as a location tool rather than a grade. Open the flagged sentences, compare them against your drafts and notes, and ask whether the flagged passage is genuinely yours and whether your own writing process explains its regularity. If it does, you have a defensible explanation; if it does not, you have something to rewrite before submission. Trusting the number without reading the highlights is the single most common way students misjudge their own risk [3].

The most useful pre-submission check is one that returns both indicators at once, because AI and similarity problems are diagnosed differently. A draft check that does not add your file to the student paper database lets you see the AI indicator and the similarity matches without consuming your institutional submission or creating a record that follows the file [4]. Reviewing both together matters: a passage can be entirely human-written and still match an existing source, and a passage can be original in wording while still reading as machine-generated [4].

Work through the report in a fixed order. Start with the similarity summary to confirm every match is properly quoted and cited, then move to the AI indicator and open each flagged sentence rather than reading only the headline percentage [4]. For each flagged sentence, decide which of three situations applies: you wrote it and can point to your notes, you wrote it but paraphrased a source too closely, or it came from a tool and needs rewriting in your own voice. That classification turns a number into an action list, and it is also the evidence trail you would need if an instructor ever questioned the passage [4].

Timing matters as much as method. Checking the night before a deadline leaves no room to rewrite flagged sections, and rewriting under pressure is exactly when prose becomes more formulaic and scores climb. Running a check early enough to allow a genuine revision pass — and re-checking after that pass — is the difference between using a detector as a planning tool and using it as a source of panic [4].

Reading the signals yourself gets you part of the way, but the sentence-level report is what actually tells you where the flags sit — and that is what turnitin0 delivers. Students upload a .docx, .pdf, or .txt file (English, 300–30,000 words, under 20 MB) and receive two downloadable PDFs in one checkout: a Turnitin AI detection report and a similarity report, matching what instructors see in their LMS. Files are checked without being added to Turnitin's student paper database, most orders finish in 5–15 minutes, and the service is independent of Turnitin, LLC. Over 100,000 reports have been delivered to 20,000+ students across the US, UK, Canada, Australia, New Zealand, and Ireland, rated 4.9/5.0.

※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary

Get Real Turnitin AI & Similarity Report