Direct answer
An AI content detector is software that estimates how likely a piece of writing was generated by a large language model such as ChatGPT, Claude, or Gemini, usually by analysing statistical patterns in the text rather than by proving authorship. The best-known academic example is Turnitin's AI writing detection, which is built into the similarity report and flags AI-generated passages segment by segment [1]. It returns a percentage of the submission that reads as AI-written, and scores below Turnitin's 20% confidence threshold are displayed as an asterisk (*%) because they are low-confidence signals [1]. This article explains how detection actually works, what the score does and does not mean, and how to preview your own draft before you submit.
How Does an AI Content Detector Actually Decide Whether Text Was Written by AI?
An AI content detector does not "know" who wrote a sentence. It measures how statistically typical that sentence is, then compares it against patterns learned from large bodies of human and machine text [2]. Two signals dominate that comparison: perplexity, which captures how predictable each next word is, and burstiness, which captures how much sentence length and structure vary across a passage [2]. Human academic writing tends to be uneven — a short blunt sentence followed by a long qualified one — while model output often settles into a smoother, more uniform rhythm [2].
That is why detection is best understood as a probability estimate rather than a verdict. A detector can flag a passage that a student wrote by hand if the prose is highly formulaic, heavily templated, or repeatedly polished by a grammar tool [2]. Equally, a lightly edited AI draft can slip under the threshold because the edits reintroduce the variation detectors look for [2]. Turnitin states plainly that its indicator is designed to support instructor judgement, not to replace it [1].
The practical consequence for students is that the score is a signal about the text, not a moral judgement about the writer. A high percentage means large stretches of the submission look machine-typical; it does not by itself establish misconduct [1]. Understanding that distinction is the first step toward using a detector productively instead of fearfully.
What Does the Turnitin AI Detection Percentage Really Mean, and How Accurate Is It?
The percentage in a Turnitin AI writing report describes the share of the submission's qualifying text that the model classifies as AI-generated [3]. It is not the probability that the student cheated, and it is not a measure of how much of the essay's ideas came from AI [3]. Turnitin reports the figure alongside highlighted segments so a reader can see exactly which passages drove the number, which is why the report is read segment by segment rather than as a single headline score [3].
Accuracy is deliberately bounded by design. When Turnitin's detection confidence falls below its 20% threshold, the score is shown as *% rather than an exact figure, and Turnitin treats those results as low-confidence signals that should not be over-interpreted [3]. Turnitin also cautions institutions against using the indicator as the sole basis for an academic misconduct decision [3]. False positives are a documented risk, particularly for non-native English writers and for text that has been run through paraphrasing or translation tools [2].
For students, the useful takeaway is that a single number is a starting point for a conversation, not a final answer. If your draft shows a flagged percentage, the productive response is to inspect the highlighted passages, check whether your citations and phrasing are doing their job, and revise the sections that read as machine-typical [3]. Reading the report that way turns a stressful number into an editing checklist.
How Can I Check My Own Draft for AI Content Before I Submit It?
The most reliable way to preview AI detection is to run your own document through a detector before the deadline, using a check that mirrors what your instructor will see. A non-repository check is the key detail here: the file is analysed without being added to Turnitin's student paper database, so your draft does not become a future match for anyone else [4]. That makes it safe to test early versions, revise, and test again as the submission date approaches [4].
Reviewing the flagged passages early is what actually changes outcomes. Students who see which paragraphs read as AI-typical can fix the underlying cause — thin paraphrase, missing citation, or overly uniform sentence rhythm — rather than rewriting blindly at the last minute [4]. Because detection looks at patterns across the whole submission, targeted revisions to the flagged segments usually move the score more than a wholesale rewrite does [2].
Timing matters too. Preview reports are most useful when there is still room to act on them, so checking a draft a few days before submission is far more valuable than checking it the night before [4]. A preview is a diagnostic, not a permission slip: it tells you where your writing reads as machine-generated so you can make it genuinely yours [4].
If you want that same preview on your own draft — the AI report and the similarity report, delivered together and without your file entering Turnitin's student paper database — turnitin0 gives you exactly the view your instructor will see, so you can revise the flagged passages while there is still time to act.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI content detector the same as a plagiarism checker?
No. A plagiarism checker looks for matches against existing sources, while an AI content detector estimates whether text was machine-generated [1]. Turnitin's AI writing detection sits alongside the similarity report rather than replacing it, so the two answer different questions [1].
Why does Turnitin show an asterisk instead of a percentage?
An asterisk (*%) means the result fell below Turnitin's 20% confidence threshold, so the system treats it as a low-confidence signal rather than an exact score [3]. Turnitin advises against drawing firm conclusions from those low-confidence results [3].
Can a human-written essay be flagged as AI?
Yes. Detectors rely on statistical patterns such as perplexity and burstiness, so highly formulaic or heavily tool-polished human writing can look machine-typical [2]. That is why Turnitin frames its indicator as supporting evidence for instructor judgement rather than proof of misconduct [1].
Does checking my draft with a detector add it to Turnitin's database?
Not with a non-repository check, which analyses the file without adding it to Turnitin's student paper database [4]. That lets you test and revise early drafts without creating future matches for other students [4].
Can I lower a flagged AI score before submitting?
Yes — the practical route is to inspect the flagged passages and revise the underlying causes, such as thin paraphrase or uniform sentence rhythm [2]. Previewing the report while there is still time to edit is what makes those revisions effective [4].