Turnitin uses its own proprietary AI detector called Turnitin Similarity

Turnitin does not use OpenAI's tools, Google's systems, or any other third-party AI detection software. Instead, the company built and maintains its own detection engine called Turnitin Similarity, which scans submitted work for patterns that suggest AI generation. This detector runs automatically when a student or instructor uploads a document to Turnitin's platform.

The system works by comparing submitted text against Turnitin's database of student papers, published sources, and web content — the same comparison that has always been Turnitin's core function. The AI detection layer sits on top of that existing similarity check. When you submit a paper, you get back a similarity percentage (matching existing sources) and a separate indicator showing whether sections appear to be AI-generated.

Turnitin has not published the exact technical details of how Similarity identifies AI text, which is standard practice in the detection industry. Revealing the method would make it easier for people to work around it. The company has said the detector trains on known AI outputs from tools like ChatGPT, Claude, and Gemini, but the specific model architecture remains proprietary.

Key Takeaways

  • Turnitin Similarity is Turnitin's own AI detector, not a licensed version of someone else's tool.
  • The detector runs automatically on all submissions and flags sections that appear AI-generated separately from plagiarism matches.
  • Turnitin has not disclosed the exact method it uses to identify AI text, which is typical for detection tools.
  • The system trains on outputs from major AI tools like ChatGPT, Claude, and Gemini, but detection accuracy varies depending on how the AI text was written.
  • Instructors see AI detection results in the same report as plagiarism results, but can choose whether to weight them in grading.

How Turnitin Similarity appears in the report you receive

When an instructor uses Turnitin, the report shows a color-coded similarity score at the top — this is the plagiarism check, unchanged from before. Below that, Turnitin now displays an AI writing indicator that shows what percentage of the document appears to be AI-generated. The indicator uses a scale: green (likely human-written), yellow (mixed or unclear), or red (likely AI-generated).

The report also highlights specific passages in the document itself. Sections flagged as AI-generated appear in a different color so the instructor can see exactly which parts triggered the detector. This lets an instructor read the flagged sections and decide whether the detector is correct — because false positives do happen, especially with technical writing or non-native English speakers.

Instructors can see this information but are not required to act on it. Some schools have policies that treat AI use as academic dishonesty; others allow AI with disclosure; still others have not yet decided. The detector provides data; the institution decides what that data means for grading or consequences.

Why Turnitin built its own detector instead of licensing someone else's

Turnitin already had the infrastructure to do this. The company has been comparing text against massive databases for over 20 years, which means it has the computing power, the data pipelines, and the expertise to add a new detection layer. Building in-house also meant Turnitin could integrate AI detection directly into the existing report format that millions of instructors already use.

Licensing a third-party detector would have meant either paying another company per submission or integrating a separate tool into the workflow. For Turnitin, developing Similarity in-house was faster and kept the entire process under one platform. It also meant Turnitin could update the detector as AI tools evolved without waiting for another company to release updates.

The downside is that Turnitin's detector is only as good as its training data and algorithms. Like all AI detection tools, Similarity has limitations — it can miss AI text that was heavily edited by humans, and it can flag human writing as AI if it happens to match patterns the detector learned. No detector is perfect, which is why most institutions treat AI detection as one signal among many, not as proof of wrongdoing on its own.

What detection accuracy actually means for Turnitin Similarity

Turnitin has published limited information about how often Similarity correctly identifies AI text. The company has said the detector achieves high accuracy on unedited AI outputs, but accuracy drops when a student edits the AI text, mixes it with human writing, or uses multiple AI tools in one document. This is true for all AI detectors, not just Turnitin's.

The real-world accuracy also depends on the writing itself. Technical papers, academic writing, and non-native English speakers sometimes produce text that looks similar to AI outputs even though it was written by a human. Conversely, an AI tool that was prompted to write in a casual, conversational style may produce text that looks more human-like and harder to detect.

Because of these limitations, most schools that use Turnitin treat the AI indicator as a starting point for conversation, not as final proof. An instructor might see a high AI score and ask the student to explain their process, or ask them to rewrite sections in their own words. The detector helps flag suspicious work, but human judgment still decides what happens next.

How Turnitin Similarity compares to other AI detectors

Several other companies offer AI detection tools: Originality.AI, GPTZero, Copyleaks, and others. Each uses a different method and trains on different data, which means they do not always agree on the same document. A paper might score as 40% AI on one detector and 20% on another. This inconsistency is one reason schools often use multiple tools or treat any single detector as incomplete evidence.

Turnitin's advantage is integration — instructors who already use Turnitin for plagiarism checking see AI detection in the same report without switching tools. Standalone detectors like GPTZero require uploading the document separately. For schools already paying for Turnitin, adding AI detection does not cost extra; for schools considering a new tool, the all-in-one approach can be simpler.

The disadvantage is that Turnitin's detector is not independently audited the way some competitors have been. Originality.AI and Copyleaks have published third-party testing results; Turnitin has released less public data about how often Similarity is correct. This does not mean Turnitin's detector is worse, only that less independent verification is available.

What happens if Turnitin Similarity flags your work as AI

Being flagged does not automatically mean you will face consequences. The flag goes to your instructor, who decides what to do with it. Some instructors will ask you to explain your writing process. Others might ask you to rewrite the flagged sections to prove you can write the material yourself. Some institutions have formal policies that treat AI use as a violation; others do not.

If you used AI and disclosed it to your instructor, the flag is not a surprise — you already told them. If you used AI without telling them, the flag gives them a reason to ask questions. If you did not use AI and got flagged anyway, you can explain that to your instructor and ask them to read the flagged sections and decide whether the detector is correct.

The key point is that the flag is information, not a verdict. What matters is what your school's policy says about AI use and what your instructor decides to do with the information Turnitin provides.

Frequently Asked Questions

Does Turnitin use ChatGPT to detect AI?

No. Turnitin Similarity trains on outputs from ChatGPT and other AI tools to learn what AI-generated text looks like, but the detector itself is Turnitin's own system. Turnitin does not use ChatGPT or OpenAI's tools to run the detection — it uses its own proprietary algorithm.

Can I see what Turnitin flagged as AI in my paper?

Yes. The Turnitin report highlights the specific passages that triggered the AI detector, usually in a different color or with a note. You can read those sections and see why the detector flagged them. Your instructor can also see this breakdown and can decide whether the detector is correct.

Is Turnitin Similarity accurate?

Turnitin Similarity works well on unedited AI text but is less reliable on heavily edited AI, mixed human-and-AI writing, or human writing that happens to match patterns the detector learned. Like all AI detectors, it has false positives and false negatives. Most schools treat it as one signal, not as definitive proof.

What if my school does not use Turnitin?

Other schools use different tools like Originality.AI, Copyleaks, or GPTZero. Some schools use multiple detectors. If you are unsure what your school uses, ask your instructor or check your school's academic integrity policy.

Can I fool Turnitin Similarity by editing AI text?

Editing AI text makes it harder for any detector to identify, but Turnitin Similarity can still catch heavily edited AI if enough of the original patterns remain. The more you rewrite and personalize the text, the harder it is for any detector to flag — but that also means you are doing more of the work yourself, which is usually the point of the assignment.