Direct answer
An AI detector "text guard" is the screening layer that scans your prose and flags passages it believes were machine-generated, and it works at sentence-segment level rather than issuing one verdict on the whole document [1]. What it actually measures is the statistical predictability of your writing, so a guard can flag polished human prose as readily as it flags ChatGPT output [3]. The practical consequence is that a guard is a review signal, not a ruling — which is why reading your own segment-level flags before submission matters more than chasing a single headline percentage [4].
What Does an AI Detector Text Guard Actually Check in Your Writing?
A text guard does not read for meaning. It reads for predictability — how surprising each next word is given the words before it — and it scores that at the level of individual sentences and segments rather than across your whole essay [2]. Turnitin's report separates two distinct categories, AI-generated and AI-paraphrased, and reports a percentage for each, because text pasted from a model and text run through a paraphraser leave different statistical fingerprints [2]. That distinction is the reason a guard can tell you "this paragraph looks paraphrased" rather than only "this paragraph looks artificial."
The guard also applies a qualifying-text filter. Only prose above a minimum length is counted toward the percentage, so reference lists, block quotations, headings and short strings are excluded from the calculation entirely [2]. This is why two students with identical AI use can show very different headline numbers: the denominator is not the document, it is the subset of your writing the detector considers analysable [2]. If a large share of your word count sits in quotes or tables, the percentage is computed on a much smaller base.
Finally, the guard has a confidence floor. When detection confidence falls below the 20% threshold, Turnitin displays *% instead of an exact figure, signalling a low-confidence result rather than a precise measurement [2]. Reading a *% as "zero" or as "safe" is a misreading of the report — it means the guard could not reach a confident determination on that text [1]. The segment highlighting is the part worth your attention, because it shows where the signal concentrates [2].
Why Does a Text Guard Flag Human Writing as AI-Generated?
Detectors model two properties of prose: perplexity (how predictable word choices are) and burstiness (how much sentence length and rhythm vary). Human academic writing is frequently low on both, because the genre rewards formulaic transitions, nominalisation and evenly paced sentences — the exact profile a detector associates with machine output [3]. Non-native English writers are over-represented in false positives for the same structural reason: careful, rule-following academic English reads as unusually regular [3].
Editing behaviour compounds the problem. Grammar checkers and "polish" passes tend to smooth out the irregular rhythm that detectors rely on as evidence of human authorship, so a document can become more flaggable after a round of legitimate proofreading [3]. This is not a flaw you can proofread your way out of by adding synonyms; the signal lives in sentence-level variation, not vocabulary.
The institutional framing matters here. Turnitin's own guidance treats a flag as a prompt for review and conversation rather than a finding of misconduct, precisely because the underlying measurement is probabilistic [1]. A guard that says 60% is reporting a statistical pattern in your text, not an assessment of how the text came to exist [3]. Knowing that changes what you do with the result: you investigate the flagged segments instead of panicking at the number [4].
How Can You See Your Own AI Detection Flags Before You Submit?
You cannot self-submit into your institution's Turnitin assignment — submission is controlled by the instructor's assignment settings, so there is no official student-side button that returns your real report [4]. That constraint is exactly why pre-submission review exists as a separate step: you need a report you can read before the deadline rather than a flag you discover after it [4]. Treat any draft-stage result as indicative of what the institutional report will show, not as a guarantee of the identical number [4].
The productive way to use a preview is segment by segment. A headline percentage tells you nothing actionable; the highlighted passages tell you which paragraphs to work on, which is usually two or three out of twenty [4]. Rewriting those specific segments for specificity — concrete evidence, your own data, varied sentence length, a claim only you could make — addresses the actual signal rather than the surface wording [4].
It also helps to re-read your draft for the two properties detectors weight: are your sentences all roughly the same length, and are your transitions the ones every essay in the genre uses? Where the answer is yes, you have found the passages most likely to be flagged, whether or not a model wrote them [3]. Fixing rhythm and adding source-specific detail is legitimate revision, and it is the same work a guard's segment view is pointing you toward [2].
Reading a segment view is only useful if the report you are reading is the one your institution will actually generate. Turnitin0 lets you upload your draft and see the same Turnitin AI and similarity reports your professor sees — the AI score, the flagged segments and the match highlights, delivered as two downloadable PDFs in one checkout, typically within 5–15 minutes and without adding your file to Turnitin's student paper database. That turns a guard's verdict into a revision list while you still have time to act on it.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is a text guard the same thing as an AI detector?
A text guard is the screening function — the part that scans and flags — while an AI detector is the broader system producing the report. In practice the terms are used interchangeably for the segment-level flagging you see in a Turnitin AI writing report [1].
What does an asterisk score mean on an AI report?
It means detection confidence fell below Turnitin's 20% threshold, so no exact percentage is shown. It is a low-confidence signal, not a confirmed clean result [2].
Can a guard flag text I wrote entirely myself?
Yes. Formulaic academic prose, heavy editing and non-native English patterns all reduce the sentence variation detectors rely on, which is why flags are treated as a review prompt rather than proof [3].
Can I check my draft without my institution seeing it?
You cannot self-submit to an instructor's Turnitin assignment. A non-repository pre-check keeps your file out of the student paper database and returns a report you can read before the deadline [4].
Should I rewrite everything a guard flags?
No. Work at segment level and revise the specific paragraphs carrying the signal, adding concrete evidence and varying sentence rhythm rather than paraphrasing the whole document [4].