Direct answer
An AI generator checker reads your text and estimates how much of it was produced by a large language model such as ChatGPT, Claude, or Gemini. It does not compare your words against a stored database of AI outputs; it measures how statistically predictable the wording is, sentence by sentence, and reports the share of qualifying text that looks machine-generated [1]. The number it returns is a probability-based signal about your draft, not a verdict about you as a writer.
How Do AI Generator Checkers Decide Whether Text Is AI-Generated?
Detection models are trained on two large bodies of text: writing known to be human-authored and writing known to be produced by AI generators. From that training, the model learns to recognise the statistical fingerprint of machine output — most notably low perplexity (the next word is highly predictable) and low burstiness (sentence length and complexity stay unusually uniform). When you paste your text in, the checker scores those properties across the document rather than searching for a copy of your paragraph anywhere [2].
Turnitin's implementation works at the segment level. It divides the submitted document into sentence-sized spans, evaluates each one, and then reports the proportion of qualifying text that its model flags as AI-generated [1]. That is why two documents with the same overall score can look completely different inside the report: one may be uniformly flagged, while another has three isolated flagged sentences inside otherwise human prose. Reading the highlighted spans is far more useful than reading the headline number [2].
It also explains why different checkers disagree on the same paragraph. Each vendor trains on different data, uses a different confidence threshold, and defines "qualifying text" differently, so scores are not interchangeable between tools [2]. A checker is a measurement instrument with a known error band, not an oracle. Treat any single score as one data point, and treat the flagged spans as the actual evidence you can act on.
What Does an AI Detection Score Like 100%, 45%, or *% Actually Mean?
The percentage is the share of your qualifying text that the model classifies as AI-generated — not the probability that you cheated [3]. A 45% result means roughly 45% of the analysed text was flagged, which usually corresponds to specific sections rather than the whole document. A 100% result means every qualifying span was flagged, which is typical of text pasted straight out of a generator with no substantive editing.
Turnitin displays an asterisk (%) instead of a precise figure when the detected AI writing falls below its 20% confidence threshold [3]. Students often misread this as an error or a hidden high score. It is the opposite: % is a low-confidence signal, meaning the detector did not find enough flagged text to report a reliable number. If you see *% on your own draft, the honest interpretation is "little to no AI writing detected with confidence," not "something is being concealed."
Context matters as much as the number. Turnitin itself advises institutions to treat the AI writing indicator as a prompt for discussion rather than proof of misconduct, because formulaic academic conventions, heavily templated lab reports, and writing by multilingual authors can all raise false-positive risk [3]. The score tells you where to look; the highlighted passages plus your own drafting history tell you what actually happened.
Can Flagged AI-Generated Text Be Revised So It Passes a Turnitin AI Check?
Yes — but revision has to target the flagged spans, not the whole document. The productive workflow is to open the report, note exactly which sentences were flagged, and rewrite those passages from your own understanding of the source material rather than paraphrasing the flagged wording [4]. Swapping synonyms inside a machine-shaped sentence usually leaves the underlying predictability intact, so the same span tends to get flagged again.
What actually moves a score is restoring the properties detectors measure. Ground each claim in a specific citation you have read, add concrete detail that only someone familiar with the material would include, and vary sentence length deliberately — a short declarative sentence next to a longer analytical one raises burstiness [4]. Where a flagged passage is genuinely your own analysis, rebuilding the argument in your own sequence of steps often clears it, because your reasoning order is not the order a generator would produce.
Confirm the result rather than assuming it. Re-run the revised draft through a checker and compare the flagged spans against the previous report; a drop in the headline number with the same spans still highlighted means the rewrite did not address the actual problem [4]. Because Turnitin's report reflects what an instructor sees in the LMS, previewing with an equivalent report before final submission is the only reliable way to know where you stand — and it is far cheaper than discovering a flag after the deadline.
If you would rather see the exact report your instructor will open — the AI indicator, the flagged spans, and the similarity summary side by side — turnitin0 produces both PDFs from a single pre-submission check, with results typically back in under 15 minutes and no addition to Turnitin's student paper database.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is an AI generator checker the same as a plagiarism checker?
No. A plagiarism checker compares your text against a corpus of published and previously submitted work, while an AI generator checker estimates whether the wording was machine-produced [1]. Turnitin reports them as two separate indicators, and a draft can score 0% on similarity and still carry an AI flag.
Why did my AI score change between two checks of the same text?
Detectors are probabilistic and threshold-based, so borderline passages can fall on either side of the cut-off, and different vendors use different models and thresholds [2]. Small edits, or a different tool, can shift the reported percentage without any change in authorship.
Does a *% result mean my text is flagged?
No. Turnitin shows *% when detected AI writing is below its 20% confidence threshold, which is a low-confidence display rather than a hidden high score [3].
Can I lower a flagged score without changing my meaning?
Usually, yes, if you rewrite the specific flagged spans with your own citation-grounded reasoning and varied sentence structure rather than swapping synonyms [4]. Always re-check the revised draft instead of assuming the score dropped.