Direct answer
There is no single tell that proves a passage was written by AI. Modern detectors such as Turnitin's AI writing indicator work statistically: they classify text sentence by sentence and report the share of qualifying text that matches machine-generated patterns, which is why the output is a percentage rather than a yes/no verdict [1]. A high score is a signal worth investigating, not proof of misconduct — and a low score, or a *% result, does not certify that text as human [1]. The reliable way to answer the question for your own draft is to run it through a detector that reports at sentence level, then read the flagged sentences yourself.
How can you tell if a piece of text was AI generated?
Detectors do not look for a watermark or a hidden signature. Turnitin's model evaluates the statistical regularity of the text and returns a document-level percentage alongside sentence-level highlighting, so you can see exactly which sentences drove the score rather than guessing from a single number [2]. That design matters: a 40% AI score on a 2,000-word essay typically means a minority of sentences were classified as AI-generated, not that 40% of your ideas came from a chatbot [2].
The signals that move a score are structural rather than topical. Machine-generated prose tends to be unusually uniform in sentence length and rhythm, leans on predictable transitions, and repeats phrasing patterns with low lexical surprise; human writing is messier, with uneven cadence, idiosyncratic word choices, and occasional awkwardness [2]. Because the model is measuring pattern regularity, a human writer who drafts in a very even, formulaic register can score higher than expected — the detector is not reading your intent, only your surface patterns [2].
This is why the honest answer to "is this text AI generated or not" is probabilistic. Turnitin itself frames the indicator as one data point that institutions should weigh alongside drafting history, version files, and conversation with the student, not as a standalone finding of academic misconduct [2]. If you want a usable answer, treat the score as a triage tool: it tells you where to look, and you supply the judgment about what actually happened.
Why do AI detectors sometimes flag human writing as AI?
False positives are a documented limitation, not a myth. Turnitin's own guidance tells institutions that the AI writing score should not be used as the sole basis for adverse decisions, precisely because the indicator can misfire on legitimate human writing [3]. The most commonly reported pattern is formulaic academic prose: heavily templated essays, lab reports with rigid section structures, and text written to a marking rubric can all display the uniformity that the model associates with machine generation [3].
Second-language writers are disproportionately affected. When a writer is working hard to produce grammatically safe, conventional sentences, the resulting text is smoother and more repetitive than a native speaker's free-flowing draft — and smoother, more repetitive text is exactly what the detector's features reward [3]. Several institutions have publicly revised their policies after flagging non-native English speakers, and Turnitin has acknowledged the issue in its own limitation notes [3].
There is also a display convention worth knowing. When Turnitin's confidence falls below its threshold, the report shows *% instead of an exact figure, which signals a low-confidence detection rather than a clean zero [3]. Reading *% as "definitely human" is a misreading of the report; reading it as "the model could not make a confident call" is accurate, and it is the correct basis for deciding whether further review is warranted [3].
How can you check your own draft for an AI score before you submit it?
In most institutions you cannot self-submit to the Turnitin assignment your instructor created — the submission point is tied to the graded dropbox, and a test upload may consume your one permitted attempt or land in the class similarity database [4]. That structural constraint is why pre-submission checking has become a separate workflow: students run their draft through an independent checker that returns the same style of report without touching the official submission [4].
What you want from that check is the full report, not a headline number. A useful preview shows the overall AI percentage, sentence-level flags so you can see which passages triggered detection, and a similarity report so you can catch source-matching problems in the same pass [4]. Reviewing flagged sentences individually is the highest-value step, because it separates genuine machine-drafted passages you should rewrite from false positives caused by formulaic phrasing you can vary instead [4].
Timing is the other practical variable. Checking early enough to revise is the whole point, so turnaround matters more than marginal accuracy differences between tools; a report that arrives after the deadline has no decision value [4]. Once you have the flagged sentences in front of you, the question stops being abstract — you are no longer asking whether the text "feels" AI-generated, you are looking at specific sentences and deciding whether to rewrite, rephrase, or leave them alone [4].
Reading your own flagged sentences is where most students discover the real problem: not that they cheated, but that their drafting habits produced text regular enough to look machine-made. turnitin0 exists for exactly that moment — it returns the same Turnitin AI detection report and similarity report your instructor would see, sentence-level flags included, without adding your file to the student paper database. You get both PDFs in one check, usually in under fifteen minutes, so there is still time to revise the sentences that actually need it.
※ Turnitin0.com - Actual [Turnitin AI Report](https://www.turnitin0.com/) Cover, Score, Flag And Similarity Summary
FAQ
Can I tell if text is AI generated just by reading it?
Not reliably. Human readers pick up on tone and rhythm, but the cues that feel obvious — em dashes, tidy transitions, balanced sentences — appear constantly in human academic writing too [2]. Reading is a useful first filter; it is not a verdict.
What does a *% result on a Turnitin AI report mean?
It means the detection confidence fell below Turnitin's reporting threshold, so the system shows an asterisk instead of a precise percentage [1]. Treat it as a low-confidence signal rather than a confirmed zero, and look at the sentence-level highlighting for context [3].
Does a high AI score prove a student cheated?
No. Turnitin advises institutions not to base misconduct findings on the AI indicator alone, because formulaic or second-language writing can trigger false positives [3]. The score identifies text worth reviewing; it does not establish how the text was produced.
Will checking my draft with a third-party tool get me in trouble?
A non-repository check does not add your file to Turnitin's student paper database and does not notify your institution [4]. That is the practical difference between previewing your own draft and submitting to the graded assignment dropbox.
If a sentence is flagged, should I rewrite it or leave it?
Rewrite the ones that genuinely came from a model, and vary the ones that are merely formulaic — break up uniform sentence lengths, replace stock transitions, and add specific detail [2]. Re-checking after revision is the only way to know whether the change moved the score [4].