What the Turnitin AI Detection Asterisk Means When the Score Is Under 20 Percent

The asterisk is Turnitin’s own reliability warning: when the AI writing score comes back below 20 percent, Turnitin marks it with an asterisk to tell instructors the number falls in a range with a higher rate of false positives and shouldn’t be leaned on. That’s the whole turnitin ai detection asterisk meaning — the company that built the detector flagging its detector’s weakest zone. It’s arguably the most honest element in the entire report, and both students and instructors routinely misread it. Here’s what the asterisk does and doesn’t imply, why the sub-20 band is genuinely noisy, and what to do when your paper draws one.

Key takeaways

• Asterisk = the score is between 0 and 20 percent, where Turnitin’s testing found elevated false-positive rates. It’s a confidence disclaimer, not an extra accusation.
• A score of 0 gets no asterisk; it’s only nonzero-but-low results that carry the caveat.
• The percentage is a *proportion of text flagged*, and in this band the flagged passages themselves are the unreliable part.
• Turnitin tells educators not to act on AI scores alone in general — and to treat asterisked scores with extra skepticism on top of that.
• If you’re a student staring at an asterisked 11%, the report is closer to an all-clear than an alarm. Panic-editing over it usually makes writing worse, not safer.

Where the asterisk comes from

Turnitin’s AI writing indicator reports what share of a submission’s prose the model classifies as likely AI-generated. Early in the feature’s life, Turnitin published accuracy claims alongside an admission: reliability isn’t uniform across the scale. High scores — a document classified as mostly machine-written — reflect a strong, consistent signal across many passages. Low scores reflect the opposite: the model flagged a few slices of an otherwise human-reading document.

Turnitin’s own validation found that this low band behaves differently. When the overall score is under 20 percent, the flagged passages are disproportionately likely to be false positives — ordinary human writing that happens to hit the statistical profile the model watches for. Rather than hide that, Turnitin attached a visible caveat: scores in the band display with an asterisk, and the report notes the percentage may be less reliable. A score of exactly zero displays clean; there’s nothing flagged to be unsure about.

It’s worth pausing on how unusual this is. Detection vendors rarely annotate their own output with “maybe don’t trust this.” The asterisk exists because the alternative — instructors prosecuting 9% scores — was producing exactly the wrongful accusations Turnitin was being criticized for.

Why low scores are noisy, mechanically

The intuition is straightforward once you see the report the way the model does. The indicator carves a document into segments and classifies each. A 60% score means a sustained pattern across most of the essay — hard to produce by accident. A 12% score means a passage or two tipped the classifier while everything around them read as human.

At that resolution, mundane things generate flags. A methods-style paragraph written in deliberately flat, procedural prose. A heavily formulaic introduction. Definitions and boilerplate phrasing. The careful, uniform sentences of a non-native English writer who learned to prize correctness over flair — a group multiple studies have shown detectors misflag at elevated rates. None of this is AI use; all of it can look statistically “too regular” in a short stretch. With only a small slice of the document flagged, there’s no corroborating pattern to check against, so the classification stands on the thinnest evidence the tool ever acts on. That’s the band the asterisk fences off.

We’ve written a fuller guide to how to read the full Turnitin AI report — highlighting, percentages, and all — if you want the complete tour rather than the low-score corner.

What instructors are told to do with it

Turnitin’s guidance to educators has been consistent: the AI score is an indicator for further review, never the sole basis for an integrity action. For asterisked scores it goes further — the display itself communicates that the number may not be dependable. In practice, a responsible instructor reads an asterisked 14% as “nothing here rises to the level of a conversation,” or at most glances at which passages were highlighted to see whether they’re the student’s own voice.

If you’re a student facing questions over an asterisked score, that context is your answer. The document trail helps too — drafts, version history, notes — but the first and simplest point is that the tool’s own maker labeled this result unreliable. A process built on evidence shouldn’t convict on a number its manufacturer asterisked. And if the concern is really about the similarity score sitting next to it, remember the two measure unrelated things; we’ve covered why similarity and AI scores disagree separately.

What a low score should change about your writing (usually nothing)

The tempting response to any nonzero score is to start sanding: rephrase everything the report might have flagged, swap words at random, break sentences apart. Against an asterisked score this is solving noise. Human-written essays land small nonzero scores all the time; rewrite the essay and a *different* small number may appear, because the band is inherently jumpy. You can churn a good draft into a worse one chasing a zero the instrument was never precise enough to award meaningfully.

The productive version of the impulse is targeted. Before a high-stakes submission, check whether any *sustained passage* of your draft genuinely reads as mechanical — not scattered sentences, but paragraphs of even, predictable prose. That’s what escalates a trivial score into a real one. A sentence-level detector shows you exactly this: see which of your sentences carry the flag, fix the passages that deserve it, and leave your voice alone everywhere else. The free check covers a full essay; pricing covers heavier use, and there are more platform guides on reading what each tool reports.

Frequently asked questions

What does the asterisk next to a Turnitin AI score mean?

It marks a score Turnitin itself considers less reliable. When the AI indicator lands below 20% (but above zero), Turnitin displays the percentage with an asterisk to signal that scores in that range show a higher incidence of false positives, and that the number shouldn’t be treated with the same confidence as higher scores.

Why is under 20 percent specifically less reliable?

Because at low proportions the model is making its call from a small slice of the document. A 12% score means the detector thought a handful of passages looked machine-generated, and with that little signal, ordinary human writing quirks — formulaic sections, careful uniform prose, boilerplate phrasing — are enough to tip the estimate. Turnitin’s own testing found more false positives in that band, hence the flag.

Can I be accused of AI use over an asterisked score?

You can be asked about it, but the asterisk is Turnitin explicitly telling instructors the number is weak evidence. Turnitin’s guidance says AI scores shouldn’t be the sole basis for action in general, and asterisked scores deserve extra caution. If a low, asterisked score is the entire case against you, pointing at Turnitin’s own reliability disclaimer is a legitimate and strong response.

Does an asterisked 15% mean 15% of my essay was AI-written?

It means the model classified roughly 15% of the analyzed text as likely machine-generated — and the asterisk means even that classification is shaky. The percentage is a proportion of flagged text, not a probability that you cheated, and in the under-20 band Turnitin says the flagged passages themselves may well be false positives.

Should I try to get my score from 15% to 0% before submitting?

Chasing zero is the wrong frame. Turnitin marks sub-20 scores as unreliable precisely because that range is noisy — human-written essays routinely land small nonzero scores. The useful pre-submission question is whether any full passages of your draft read as mechanical, which a sentence-level check on an independent detector shows you directly. Fix genuinely robotic passages; don’t sand your writing down over noise.