Direct answer
To check whether content is AI or human, you run the text through an AI writing detector that scores what share of the submission reads as machine-generated, then inspect the sentence-level highlights rather than trusting the headline number. Turnitin's AI writing indicator does exactly this: it segments the text, flags sentences it classifies as AI-written, and reports an overall percentage [1]. That percentage is a proportion of flagged text, not a probability that a person cheated, so the honest answer to "AI or human?" is usually "mostly one, with some of the other."
How Can I Tell If Content Was Written By AI Or By A Human?
You cannot tell reliably by reading. Human readers — including experienced tutors — perform close to chance on short passages, because the surface features people associate with AI (smooth transitions, tidy structure, balanced paragraphs) are also features of competent academic prose. Turnitin's own guidance frames the indicator as a signal that supports a conversation and a review, not as proof of authorship [2]. That framing matters: the tool produces evidence, and a human produces the judgement.
What the detector actually does is statistical pattern-matching. It looks at how predictable each token is given the words around it, then aggregates those signals across sentences. Text drafted by a language model tends to sit in a narrower band of predictability than text a person typed while thinking. When a submission mixes both — a human outline with AI-polished paragraphs, say — the report reflects that mixture, which is why partial percentages are common and why a 34% result is not the same accusation as a 100% result [2].
The practical consequence is that you should read the report as a map, not a verdict. Look at which sections were flagged, then read those sections yourself and ask whether the flagged prose is genuinely generic or whether it is technical writing that happens to be formulaic. Turnitin explicitly acknowledges that false positives occur, particularly with heavily edited drafts and with writing that follows rigid conventions [2]. If the flags cluster in your methods section or your literature summary, that pattern is informative in itself.
What Does The Turnitin AI Score Percentage Actually Mean?
The percentage is the proportion of your qualifying text that the detector classified as AI-generated — nothing more and nothing less [3]. It is not a confidence score, not a plagiarism measure, and not a grade. A 62% result means roughly three-fifths of the processed words were flagged, which is a very different claim from "62% likely to be misconduct." Instructors who understand the metric read it as a prompt to open the submission, not to close the case [3].
Two mechanical details change how you should read any number you see. First, only submissions that clear the minimum word count are processed at all, so very short pieces may return no AI result rather than a low one [3]. Second, when the detected share falls below Turnitin's confidence threshold, the report displays an asterisk in place of a number — the *% band. That asterisk is a low-confidence signal, not a zero, and treating it as "clean" overreads the tool in the opposite direction [3].
The score also depends on the text you feed it. Formatting stripped from a PDF, a reference list pasted into the body, or a document that has been through several rounds of paraphrasing will all shift the result. If you are comparing two versions of your own draft, keep the input conditions identical, or you will be measuring your export settings rather than your writing [3].
Can A Student Check Their Own Draft For AI Before Submitting?
Yes, and doing so is the single most useful thing you can do with an AI score. Turnitin recommends being transparent with students about what the indicator can and cannot prove, precisely because a student who understands the metric can act on it before a deadline instead of after [4]. Checking your own draft turns an opaque institutional judgement into a concrete editing task: here are the flagged sentences, here is what I do about them.
The most effective response to a flag is not synonym-swapping. It is returning to the flagged passage and rebuilding the reasoning in your own words — adding the specific example, the counter-argument, or the source detail that a model would not have supplied [4]. Flagged text is usually flagged because it is generic, so specificity is the fix. Students who rewrite at the level of argument rather than vocabulary consistently see the flagged share fall, and they end up with a better paragraph in the process.
It also helps to know where the decision actually sits. Turnitin's guidance is clear that institutional policy, not the tool, determines consequences, and that the report is designed to open a discussion between student and instructor [4]. That means a pre-submission check is not a way to game a system — it is a way to arrive at that discussion with a draft you can defend, and with the confidence that you have already reviewed the passages most likely to draw a question.
If you want that same pre-submission certainty on your own draft — the real Turnitin AI report and the similarity report, side by side, before your deadline rather than after it — turnitin0 runs the check for you and returns both PDFs in one order, usually in under 15 minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is there a free way to check if content is AI or human?
Free detectors exist, but most give you a single opaque percentage with no sentence-level evidence and no similarity report alongside it. Turnitin's indicator is the one instructors actually use, so a check against it tells you what your marker will see rather than what an unrelated model thinks [1].
Can Turnitin tell the difference between AI-written and AI-edited text?
Not cleanly. The detector scores predictability, so a human draft that has been heavily polished by a model can flag much like generated text, and a generated draft edited into a personal voice can score low [2]. This is why the report shows highlighted sentences instead of a single verdict.
What does an asterisk instead of a percentage mean?
An asterisk means the detected AI share fell below Turnitin's confidence threshold, so the report withholds an exact figure [3]. Read it as a low-confidence signal rather than a confirmed zero, and check whether the flagged passages still make sense to you on review.
If my draft is flagged, should I rewrite the whole thing?
Usually not. Start with the highlighted sentences and rebuild the reasoning in your own words, adding specifics a model would not have produced [4]. Targeted revision at the level of argument is more effective than paraphrasing every sentence, and it preserves the parts of your draft that were never the problem.
Does checking my draft before submission count as misconduct?
Turnitin frames the indicator as a prompt for discussion between student and instructor, with consequences set by institutional policy rather than by the tool [4]. Reviewing your own work for clarity and originality before submitting is ordinary academic practice.