Direct answer
AI writing detection works by scoring a document segment by segment and measuring how statistically predictable its language is: uniform, low-variation prose reads as machine-generated, while irregular human phrasing does not. Turnitin's AI detector reports the percentage of qualifying text it flags as AI, and displays an asterisk (*%) rather than a number when the result falls below its 20% confidence threshold [1]. Knowing how those signals are produced is what turns a detection score from a vague threat into something you can actually act on before submitting.
How Does AI Detection Actually Work?
Detection is not plagiarism matching. A similarity checker compares your text against a database of existing documents; an AI detector compares your text against a statistical model of how language is usually produced [2]. Turnitin's AI writing detection divides a submission into smaller units and evaluates each one for patterns characteristic of generated text, then aggregates them into an overall percentage of qualifying text [1]. Because the analysis is segment-level, a single document can contain both flagged and unflagged passages — a mixed result is normal, not a contradiction.
The core signals are predictability and variation. Language models tend to select the most probable next word, which produces prose with unusually low burstiness (little variation in sentence length and rhythm) and high local predictability [2]. Human writers, by contrast, drift: they use unusual collocations, interrupt themselves, vary clause length, and occasionally phrase things awkwardly. Those irregularities are exactly what a detector is measuring for.
A practical consequence is that detection needs volume. Turnitin's AI detection does not operate on very short samples, and short answers, reference lists, and quoted material are excluded from the qualifying-text calculation [1]. This is why a 250-word discussion post can come back with no meaningful AI indicator while a 3,000-word essay returns a clear percentage — there simply is not enough signal to analyze in the shorter text.
How Accurate Are AI Detectors, And Why Do They Produce False Positives?
No AI detector is a verdict machine, and the vendors say so themselves. Turnitin positions its AI score as an indicator that must be interpreted alongside context, drafts, and the instructor's knowledge of the student's writing, not as standalone proof of misconduct [3]. Treating a percentage as conclusive is the single most common misuse of the tool — on either side of the desk.
False positives cluster around a few recognizable conditions. Highly formulaic human writing (lab reports, standard essay templates), phrasing typical of non-native English speakers, and text that has been heavily edited or run through grammar tools can all look statistically uniform enough to trigger a flag [3]. Short documents and passages that closely follow a source's structure carry the same risk, because low variation is the very thing the detector is scoring.
The inverse error also exists: AI text that has been substantially rewritten, restructured, or manually edited can reduce its own detectable signature and score lower than it should [2]. For students, the honest takeaway is that both directions of error are real. A high score does not prove you used AI, and a low score does not prove you did not — which is precisely why checking your own draft before submission is more useful than guessing.
Can You Check Your Own Draft With A Turnitin AI Detector Before Submitting?
Normally, no — institutional Turnitin access belongs to the university, and students cannot run their own papers through the same account their instructor uses [4]. That gap is why independent pre-check services exist. They accept your draft, run it through detection, and return the same two reports a professor sees: an AI writing report and a similarity report [4].
A pre-submission check is most valuable as a diagnostic, not a verdict. It tells you which sections carry an AI signal, how much qualifying text was flagged, and whether your similarity matches are concentrated in a few passages or spread across the document [4]. That is actionable information: you can rewrite the flagged paragraphs, add your own analysis, and re-check before the deadline rather than discovering the problem after the grade is posted.
One detail worth understanding before you check anything: when detection confidence is low, Turnitin shows an asterisk (%) instead of a precise figure, because the system will not report a number it cannot stand behind [1]. A % result is a low-confidence signal, not a clean bill of health and not an accusation. Reading it correctly is part of reading the report correctly.
If you want to see your own numbers before your instructor does, turnitin0 runs the check and hands you the actual reports — AI score, flagged passages, and similarity matches — in one delivery, so you can decide what to revise while there is still time.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin detect AI in every document?
No. Detection requires enough qualifying text to analyze, and very short submissions may return no meaningful indicator at all [1]. Length and the amount of excluded material (quotes, references) both affect the result.
What does an asterisk (*%) mean on a Turnitin AI report?
It means the AI score fell below Turnitin's 20% confidence threshold, so the system reports a low-confidence signal instead of an exact percentage [1]. It is not proof of AI use and not proof of original writing.
Can AI detectors be wrong about human writing?
Yes. Formulaic academic templates, non-native English phrasing, and heavily edited text can all produce false positives [3]. Turnitin itself advises interpreting the score as a signal rather than a conclusion.
Why does one essay get flagged and another with similar content not?
Detection works at the segment level and depends on statistical regularity, so sentence rhythm, variation, and editing history matter more than topic [2]. Two essays on the same subject can score very differently.
Can I check my own paper with Turnitin before submitting?
Not through your university's institutional account [4]. Independent pre-check services return the same AI and similarity reports so you can review your draft ahead of the deadline [4].