Direct answer
An AI writing detector is software that analyses a document sentence by sentence and estimates how much of the text carries the statistical fingerprint of machine generation, then reports that estimate as a percentage of qualifying text. Turnitin's AI writing report, the version most universities use, displays an overall score plus sentence-level highlighting, and it shows an asterisk instead of a number when the signal falls below its confidence threshold [1]. The number is an indicator for a human reviewer, not a verdict of misconduct [1].
How Does an AI Writing Detector Decide Whether Text Was Written by AI?
Detection starts with segmentation. Before any scoring happens, the detector splits the submitted document into sentence-level spans so that a mixed draft — your own paragraphs plus a rewritten introduction — can be scored in pieces rather than as one lump [2]. That is why two students with the same overall percentage can receive very different reports: the pattern of where the flagged spans sit matters as much as the total.
The model then looks at patterns rather than at content. It evaluates word choice, sentence structure, phrasing and the predictability of transitions across those spans, and it compares what it observes against the characteristics of text it has learned to associate with generative models [2]. Nothing in that process checks whether an idea is correct, well cited, or appropriate for your discipline — a factually excellent paragraph can still be flagged if its prose is unusually uniform.
What comes out is a percentage of qualifying text, not a probability that you cheated [2]. This distinction is the single most misunderstood part of the whole system: a 40% score does not mean "40% chance this student used AI," it means that roughly 40% of the assessed text carried the detected pattern. Reviewers are expected to read the highlighted spans and weigh them against drafting history, version files, and the assignment's expectations [1].
Finally, the detector has boundaries that shape what it can see. Non-English text and very short submissions are outside the reliable assessment range, and the model does not train on student papers, so a document submitted once is not recycled into the next student's score [3]. Knowing those limits is what lets you read a report as evidence instead of as an accusation.
Why Does an AI Writing Detector Flag Human-Written or AI-Polished Work, and What Do the Scores Actually Mean?
Formulaic writing is the main source of surprising flags. Academic conventions reward predictable structure — topic sentence, evidence, analysis, link — and the detector is trained on patterns, so a paragraph that follows the template very tightly can resemble generated prose even when every word is yours [2]. Students writing in a second language, or under pressure to sound formal, produce exactly that kind of uniform prose.
Scores below the confidence threshold are deliberately vague. Turnitin displays an asterisk rather than an exact figure when its AI detection is below the 20% confidence mark, because a low-confidence signal should not be read as a precise measurement [3]. Seeing *% on a report is therefore not a hidden number you failed to unlock; it is the system telling you the signal is too weak to quantify responsibly.
Interpretation is contextual by design. Guidance for staff is explicit that the report is one input among several and should be read alongside drafting evidence, and that a highlighted span is a prompt for a conversation rather than a finding [1]. The same guidance notes that mixed authorship — human drafting polished by an assistant — is exactly the situation where a score is least decisive on its own [3].
Practical reading order for your own report: check the overall percentage, then look at which spans were flagged, then ask whether those spans are the ones you actually drafted with assistance. If the flagged text is your own and simply formulaic, you have a defensible explanation. If it is the section you asked a model to "tighten up," you now know precisely what a reviewer will see — which is the whole point of checking early.
Can You Check Your Own Draft With a Real AI Writing Detector Before You Submit?
Yes — and institutions increasingly expect students to be transparent rather than to guess. Published guidance for teaching staff frames AI writing as something to discuss openly with students, with formative checking and conversation positioned ahead of punitive responses [4]. That framing matters: the report is meant to be seen and discussed, not discovered by surprise after the deadline.
The reason to preview is format familiarity. When you have already seen the layout of a Turnitin AI report — the overall score, the sentence highlighting, the asterisk convention, the accompanying similarity summary — a result that arrives later is no longer ambiguous [1]. You can tell the difference between a low-confidence asterisk and a genuine flagged block without panicking, and you can prepare an explanation for the specific spans that were marked [4].
Previewing also lets you act while you still can. If a section is flagged and you know it was model-assisted, you have time to rewrite it in your own words, keep your citations intact, and re-check; if it is flagged and it is entirely yours, you have time to collect your drafts and notes as evidence before submission [4]. Either way the decision stays with you rather than with an inbox notification.
One caution about tool choice: an "AI writing detector" that returns a bare percentage with no sentence-level breakdown gives you almost nothing to work with, because it does not tell you which passages a reviewer would question [2]. What you want is the report structure your institution actually uses, so the preview and the real submission are directly comparable [1].
If you would rather see that report on your own draft tonight than on your marked submission next month, turnitin0 runs the same check for you: upload a .docx, .pdf or .txt file of 300–30,000 words and you receive two PDFs in one order — a Turnitin AI detection report and a similarity report, laid out the way your professor sees them. Most orders finish in 5–15 minutes, files are never added to Turnitin's student paper database, and more than 100,000 reports have been delivered to students in the US, UK, Canada, Australia, New Zealand and Ireland.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI writing detector the same thing as a plagiarism checker?
No. A similarity checker compares your text against published and submitted sources, while an AI writing detector looks for machine-generation patterns inside your own prose [2]. Turnitin0 delivers both reports together because institutions review them side by side [1].
What does an asterisk instead of a percentage mean on the report?
It means the AI signal fell below the detector's confidence threshold, so Turnitin shows *% rather than an exact figure [3]. Treat it as a low-confidence result, not as a hidden high score.
Can a detector flag text I wrote myself?
It can, especially when the writing is highly formulaic or follows a rigid academic template [2]. That is why the report is meant to be read with context and discussed with your instructor rather than treated as a final judgement [1].
Does checking my draft affect my institution's records?
A non-repository check is not added to Turnitin's student paper database and is not shared with third-party databases, so your draft stays with you [3]. Turnitin0 also lets you delete files from your account after download.
How do I lower a flagged score without losing my meaning?
Rewrite the flagged spans in your own voice, keep your citations and headings, then re-check the revised draft [4]. Turnitin0's AI humanizer is built for exactly that loop on ChatGPT, Claude or Gemini drafts, and 98.2% of humanizer orders are re-checked with Turnitin.