Does Turnitin Detect QuillBot Paraphrasing?

Turnitin says it detects AI content that has been run through a paraphrasing tool, that the check runs automatically on every submission where AI writing detection is enabled, and that it will not publish which tools it catches. So the honest answer for QuillBot specifically is that Turnitin neither confirms nor denies it by name, and has said outright that naming names is something it refuses to do. What it does describe is the category, and QuillBot sits squarely inside it.

There is a second answer underneath that one, and it is the one that actually catches people. Paraphrasing a source you did not write is a similarity problem, not an AI problem, and Turnitin has been good at that particular job since long before language models existed.

What Turnitin says about paraphrasing tools

Turnitin's own FAQ, read on August 15, 2026, describes its capabilities as including "detection of likely AI-generated content that may have been modified using a word spinner/AI paraphrasing or bypassing tool to evade detection". It adds that this "is run automatically for all submissions made to Turnitin by institutions that have the AI writing indicator enabled for their accounts".

Two details are easy to skim past. The first is that this is not a separate report, a separate setting or a separate score. Turnitin says the paraphrasing and bypassing detection "is integrated within our AI writing indicator", so a submission is checked for AI writing, AI bypassing and AI paraphrasing in a single pass and the result lands in one percentage.

The second is a limit worth knowing. The main AI detector runs on long-form English, Spanish and Japanese, but Turnitin states that the paraphrasing and bypassing layer is narrower: those capabilities "are only available for English submissions". If your paper is in Spanish, the AI writing check applies and the paraphrase-detection layer does not.

Why Turnitin will not say whether it catches QuillBot

This is the part that separates what is knowable from what is marketing. Turnitin has been asked directly which paraphrasing tools its model detects, and its published answer is that it is "unable to disclose the names of these tools", because sharing a list "would make it easier for students to evade our system".

That single refusal does more work than any test result you will find online. It means nobody outside Turnitin has a list, so any page telling you confidently that QuillBot is or is not detected is guessing. It also means the situation is not static: Turnitin re-tests its detector before every model release and consolidated its multi-model ensemble into a single model in July 2026, so a result someone posted in 2024 describes a system that no longer exists.

You also cannot check it yourself. Turnitin's AI indicator is visible to instructors and administrators only, so there is no experiment a student can run and no feedback loop for anyone to learn from. Whether your institution even has the feature switched on is a separate question with its own answer, which we work through in does Turnitin check for AI writing.

Does QuillBot get detected by Turnitin?

Split the question by what you actually put into it, because the risk is completely different in each case.

The middle row is the one students underestimate. Similarity matching does not need AI detection to catch a paraphrased source, and it never did. Turnitin also states that the AI percentage is "different from and independent of the similarity score", so a clean AI reading tells you nothing about the similarity reading and the reverse is equally true. Two separate numbers, two separate ways to have a problem.

Why paraphrasing can make an AI score worse

This is counterintuitive enough that it is worth stating plainly. Turnitin publishes the kinds of writing its detector is most likely to misread, and the list is three items long: "content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas".

Read that third one again. Paraphrasing is on Turnitin's own list of things that raise a false positive risk, for human and AI text alike. A synonym swapper changes the words and leaves everything the detector actually measures intact: the sentence lengths stay even, the paragraph shapes stay uniform, the transitions stay predictable, the ideas stay restated rather than developed. You end up with prose that is statistically flatter than what you started with.

The detector is not reading vocabulary. It scores how predictable each stretch of writing is, segment by segment, and pools those scores into one number. Swapping "significant" for "substantial" does not move a predictability measurement, because both words were the expected word in that slot. We unpack the scoring mechanism in how Turnitin detects AI writing.

QuillBot ships its own AI detector, which tells you something

Worth knowing before you rely on a paraphraser as a detection strategy: QuillBot sells both sides of this. Alongside the paraphraser it offers its own AI humanizer and its own AI content detector, and on the humanizer page the feature that scores how human your text sounds is marked Premium, with a link across to the detector. The free humanizer stops at 125 words a pass and six passes a day, which we break down on the QuillBot AI humanizer page. Checked August 22, 2026.

None of that is a criticism of QuillBot, which is a good tool for what it is. The point is what the product structure implies. A company that builds a paraphraser, a humanizer and a detector understands perfectly well that paraphrasing and detection-avoidance are different jobs, and prices them separately. If reworded text reliably cleared detectors, the humanizer would not need to exist as a distinct product.

It is also worth separating academic use from everything else. Outside a submission box, the same tools get used to stretch one piece of writing across a blog post, a newsletter and a handful of social posts, which is a genuinely different task from disguising a source and is better served by reshaping a piece properly for each channel than by a synonym pass.

What actually changes a detection reading

If rewording is not the lever, what is? Structure and variation, which are the things a predictability model is measuring in the first place.

That means sentences of genuinely different lengths, including short ones. Paragraphs that are not all the same shape. Specifics rather than category words, since a concrete detail is by definition the less predictable choice. Arguments that go somewhere instead of restating a premise three ways. And ideas developed rather than reworded, which is the exact distinction Turnitin drew when it named paraphrasing without new ideas as a false-positive risk.

That is the work our humanizer does, and it is a different operation from synonym substitution: rewriting flat, evenly-paced drafting into prose with the rhythm of something a person wrote. What it cannot do, and what nobody honest will promise, is tell you what number Turnitin will report. There is no public Turnitin checker to test against, as we set out on our Turnitin AI checker page, and the paraphrase layer is deliberately undocumented.

Keeping evidence of how you wrote something remains the strongest position available, because it answers the question a flag actually raises. We cover what holds up in how to prove you did not use AI, and the wider question of whether rewriting tools beat detectors at all in can AI detectors detect paraphrasing.

The short version

• Turnitin detects AI text "modified using a word spinner/AI paraphrasing or bypassing tool", and runs that check automatically on every submission where AI detection is enabled.
• It will not name the tools it catches, saying a list "would make it easier for students to evade our system". Nobody outside Turnitin has that list.
• The paraphrase and bypass layer is English only, narrower than the main detector's English, Spanish and Japanese.
• Paraphrasing a source is a similarity problem, separate from AI detection, and Turnitin has caught it for years.
• Turnitin names "text that has been paraphrased without developing new ideas" as writing its detector is most likely to misread, so rewording can raise your reading rather than lower it.
• QuillBot sells a paraphraser, a humanizer and an AI detector as three separate products, which is itself a statement about what paraphrasing does and does not accomplish. What its humanizer caps on the free tier and charges to unlock is a separate decision from the paraphraser.
• What moves a predictability score is structural variation and developed ideas, not swapped vocabulary.