Does Turnitin Detect QuillBot? (What Turnitin Actually Documents)

Reviewed and updated August 2026 against Turnitin’s AI writing detection capabilities FAQ and its AI writing product page.

This page used to say the opposite. An older version of it told readers that QuillBot “cannot be detected” by Turnitin. That was written before Turnitin shipped AI writing detection, and it is no longer true — so it has been rewritten from the vendor’s own documentation rather than from what people repeat on forums.

Turnitin runs two different checks, and people confuse them

Almost every argument about QuillBot and Turnitin goes wrong in the same place: it treats Turnitin as one thing. It is two. Understanding which check you are actually asking about resolves most of the confusion by itself.

The similarity check is the old one, the one that produces the percentage everyone recognises. Turnitin is blunt about what it does: “Turnitin does not check for plagiarism in writing. We do check all submissions against our database.” It looks for text in your submission that matches text it has already seen — web pages, journals, publications, and papers other students have submitted. The score is the proportion of your writing made up of those matches.

The AI writing check is the newer one, and it works on an entirely different principle. It does not compare your text against a database at all. It reads the text and predicts, sentence by sentence, whether a language model wrote it.

This distinction matters enormously for a paraphrasing tool. Rewording a source really can reduce a similarity score, because the matching strings are gone. That is the grain of truth buried in the old advice. But it does nothing about the second check — and the second check is the one built for exactly this behaviour.

Does Turnitin detect QuillBot?

Turnitin does not name QuillBot specifically, and no honest page should claim it does. What Turnitin names is the category, and it names it explicitly on its own product page: the AI content checker “helps educators, publishers, and researchers identify when an AI writing tool such as ChatGPT, text spinners, or AI bypassers (also called humanizers) may have been used in students’ submissions.”

A paraphrasing tool that rewrites your sentences with an AI model sits inside that category. Turnitin goes further and builds the distinction into the report itself, describing output as “likely AI-generated from ChatGPT or likely AI-generated and likely AI-paraphrased or bypassed.”

So the accurate answer is not “yes, Turnitin detects QuillBot” and it is certainly not “no, QuillBot is undetectable.” It is this: Turnitin ships a detection capability aimed squarely at AI paraphrasing, it reports paraphrased text as its own category, and it makes no promise of catching every instance.

How the AI paraphrasing detection works

Turnitin describes the mechanism in reasonable detail. A submission is first broken into segments of roughly a few hundred words — about five to ten sentences — and those segments are deliberately overlapped so each sentence is scored in context.

Every sentence then receives a score between 0 and 1. A sentence the model believes a human wrote scores 0; one it believes was entirely AI-generated scores 1. The document-level percentage is built from the average of those segment scores.

The underlying model was originally trained on GPT-3 and GPT-3.5 and has been extended since. Turnitin lists the coverage openly, and the list is long and current — GPT-4 and GPT-4o, the GPT-5 family, several Gemini versions, Claude Sonnet 4.5, LLaMA, and tools built on top of those models. That last clause is the one that matters for paraphrasers, because a paraphrasing product is a tool built on top of a model.

One important limit: AI paraphrasing and bypassing detection is English-only. Turnitin’s base AI writing detection also supports Spanish and Japanese, but it states that the paraphrasing and bypassing capabilities “are only available for English submissions.”

Reading the report: cyan versus purple

If you ever see the AI writing report, the colours carry specific meanings that Turnitin defines:

Purple is the paraphrasing category. Its existence is the clearest possible answer to whether Turnitin thinks about paraphrasing tools: the company built a dedicated colour for them.

One detail students often get wrong — the AI writing indicator and report are not visible to students. Only instructors and administrators see them. An instructor can download the report as a PDF and share it with you, but you cannot check your own AI score inside Turnitin before submitting.

The 20% threshold, and what the asterisk means

Turnitin deliberately suppresses low scores rather than reporting them. When the AI detection score falls in the 1% to 19% range, no percentage is attributed at all — the report shows an asterisk instead. The company built this in specifically to avoid acting on weak signals.

Above that band, Turnitin publishes its error target: it aims to keep the false positive rate — human-written text incorrectly flagged as AI — under 1% for documents with over 20% AI writing. Before each model release it tests against a further 700,000 academic papers written before ChatGPT existed to validate that figure.

Turnitin is also candid about the trade-off it chose. Keeping false positives that low means accepting misses: “there is a chance that we might miss some AI written text in a document. We’re comfortable with that since we do not want to incorrectly highlight human-written text as AI-written.” Its own example is that a document scored at 50% could actually contain as much as 65% AI writing.

What the checker will not process

The detector only analyses what Turnitin calls qualifying text: long-form prose, meaning blocks written in standard grammatical sentences. Lists, bullet points and other short non-sentence structures are excluded. This is why the reported percentage is explicitly “not necessarily the percentage of the entire submission.”

Short documents behave differently too. Turnitin warns that in a paper of only a few hundred words the prediction becomes “mostly all or nothing,” because there is a single segment and no opportunity to overlap. A mixed short piece can be flagged as entirely AI-generated.

Submissions in unsupported languages are not processed at all — the indicator shows an empty or error state and no report is generated. And a paper submitted before the AI detection feature launched will not carry a report unless it is resubmitted.

Does your school even have it switched on?

This is the question most pages skip, and it is the one that changes the answer for any individual student. AI writing detection is an administrator setting. Turnitin states that admins “have the option to enable/disable the AI writing feature from their admin settings page,” and that disabling it removes the indicator and report from the Similarity Report entirely, for instructors and admins alike.

That places Turnitin in exactly the same position as every other tool in this category. The vendor builds the capability; the institution decides whether it runs. It is the same pattern as Proctorio collecting data only “if the Institution chooses to collect it,” and the same reason a blanket answer about what “Turnitin can see” is never quite right for your specific course.

There is a second layer of judgement on top. Turnitin repeats, in the same words it uses about similarity scores, that it “does not make a determination of misconduct, rather it provides data for the educators to make an informed decision.” The percentage “should not be used as the sole basis for action.” The tool records; a person judges.

Is using QuillBot itself against the rules?

Separate question from detection, and the honest answer is that it depends on a policy Turnitin does not write. Paraphrasing software is not inherently prohibited, and many institutions treat a grammar or clarity tool very differently from one that rewrites arguments wholesale.

What is broadly consistent across academic-integrity policies is the principle underneath: the submitted work is supposed to represent your own thinking, and rewording someone else’s uncited source does not make it yours. Turnitin notes this indirectly — where a similarity match is suppressed by rewording, the AI report may still indicate “only that the content is not the student’s own work.”

Your course handbook, module guide, or programme regulations are the authoritative source here. If they are silent or ambiguous, asking the instructor before submission is a genuinely better use of your time than trying to reason from what a detector may or may not catch.

QuillBot now ships detection tools of its own

A detail that undercuts the whole “beat the checker” framing: QuillBot’s own product line now includes an AI Detector and a Plagiarism Checker alongside its paraphraser. Its AI detector page advertises detection of content generated by GPT-5, Gemini, Claude and Llama, and QuillBot itself frames AI content detection as something “used by educators, editors, SEO professionals, and businesses to make sure content is original.”

The company selling the paraphraser also sells the thing that catches paraphrasers. That is a reasonable signal about which way this arms race is pointed.

If you have been flagged and you did write it

False positives are real — Turnitin publishes a rate rather than denying one — and the company names the text most likely to trigger them: content “without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas.” Formulaic academic prose can look statistically flat.

Practical steps that actually help:

• Ask to see the report. You cannot see it yourself, but an instructor can download and share the PDF. Which sentences are highlighted matters far more than the number.
• Produce your drafting history. Version history in Google Docs or Word, notes, and outlines are the most persuasive evidence available to you, and they only exist if you kept them.
• Check the score band. A flag under 20% is not attributed a percentage by design, which is useful context to raise.
• Point to the vendor’s own framing. Turnitin states its score should not be the sole basis for action, and that it does not determine misconduct.
• Follow your institution’s appeal process. Nearly every institution has one, and it is a normal thing to use.

If you are the instructor looking at a flag

The vendor documentation is unusually helpful here, because it tells you what the number is not. It is not a misconduct finding, it is not a percentage of the whole submission, and below 20% it is not even a percentage — the asterisk exists precisely so that weak signals do not get treated as evidence.

The report is more informative than the score. Purple highlighting is the paraphrasing category specifically, so a submission that is mostly purple is telling you something different from one that is mostly cyan. And because only prose sentences are analysed, a heavily bulleted or tabular submission may produce a score drawn from a small fraction of the actual work.

Two structural cautions worth holding on to. Turnitin names the writing most likely to produce a false positive — repetitive text, text with little structural variation, and text paraphrased without developing new ideas — which describes a fair amount of legitimate undergraduate writing. And in short submissions of a few hundred words, the model itself becomes close to all-or-nothing. Neither of those is a reason to distrust the tool; both are reasons to open the report before opening a case.

Frequently asked questions

Can Turnitin tell which tool I used?

No. Turnitin predicts that text is likely AI-generated or likely AI-paraphrased. It does not identify a specific product, and it does not claim to. Any page that says Turnitin names QuillBot by name is describing something the documentation does not support.

Does paraphrasing lower my similarity score?

It can, because the similarity check looks for matching strings and rewording removes them. That is a genuinely different check from AI detection, which is why the two questions produce different answers — and why a low similarity score is no longer evidence of much on its own.

Can Turnitin detect copy and paste?

Directly copied text is the case the similarity check handles best, because the matching strings are intact. If the source is anywhere in Turnitin’s database — including papers other students have submitted — the passage is highlighted.

Does Turnitin keep my paper?

Submissions can be stored in Turnitin’s repository and used in later similarity checks, which is how the student-paper database exists at all. Repository behaviour is a setting institutions control. Notably, Turnitin states the AI indicator does not create a separate repository: “There is no separate repository for AI writing detection.”

Is a high AI score proof of cheating?

No, and Turnitin says so itself. The company does not make a determination of misconduct; it provides data for an educator to interpret against institutional policy. The score is a starting point for a conversation, not a verdict.

For a wider view of the tools themselves, and when using one is a legitimate study aid rather than misconduct, see our guide to paraphrasing tools and academic integrity.