Can Another AI Detector Tell You What Turnitin Will Say?

Not in a way you can rely on, and a detector company that tested exactly this says so on its own help page. In an August 2026 study, Originality.ai found that its closest setting, called 40% AI Allowance, made the same human-or-AI call as Turnitin on 69.7% of 300 essays. The help page then adds: "The 40% AI Allowance setting does not predict or reproduce a Turnitin result, and the scores from the two tools should not be compared directly." The study's other headline figure, 96.3%, sounds like a forecast. Mostly it reflects how rarely Turnitin crossed the study's 50% line on essays that people wrote.

Two more things are worth knowing before you go looking for a second opinion. Turnitin does not show its AI result to students at all, although an instructor is able to send you the report as a PDF. And at least one university now tells students not to use detector websites, while others bar staff from uploading student work to them. Both are covered further down.

The study is "What AI Detector Is Most Similar to Turnitin? We Tested 300 Student Essays", posted on Originality.ai's blog on 20 August 2026 under the name of Jonathan Gillham, the company's founder and CEO. A help-centre article summarising it is marked "Updated on: 19/08/2026".

Keep in mind whose study this is. Originality.ai sells the detector being compared, the article closes by inviting readers to "Check your work with Originality.ai", and it is a company blog post rather than a peer-reviewed paper. The page states that Originality.ai "is not affiliated with or endorsed by Turnitin."

How the test worked:

• 300 English essays from GEDE, a public research dataset for testing AI detection in education, released with a 2025 paper by Lukas Gehring and Benjamin Paaßen. 100 were fully human, 100 were human essays polished with AI, and 100 were fully AI-generated.
• Six scans per essay. One by Turnitin, and five by Originality.ai's AI Allowance mode at its 0%, 5%, 15%, 25% and 40% settings.
• Every result became a yes or no. For Turnitin, a score of 50% or more counted as AI. The study says itself that this line "is not an official Turnitin Human/AI threshold".
• Some things the page does not give. It does not say when the scans were run or which Turnitin model version was used. It mentions a "supporting Google Sheet" with the essay-by-essay data, but when we read the page on 2 October 2026 there was no link to one.

It is 209 essays out of 300 on which the 40% setting and Turnitin landed on the same side of that yes/no line. That was the best of the five settings. The 25% setting agreed on 55.3%, and the three strictest settings on fewer than half (45.3% to 46.7%, from the study's Figure 1).

Where the two tools split tells you more than the total does. Of the 91 essays they disagreed on, 76 were fully AI-generated: the 40% setting labelled all 100 of those as Likely AI, and Turnitin put 24 of them at 50% or above. The page does not say where the other 76 fell below 50%, so this is not a count of zeros. The remaining 15 disagreements were AI-polished essays. On the 100 fully human essays, neither tool called anything AI.

The study's own verdict on the headline figure is blunt: "69.7% is not close enough to call the tools substitutes". The same sentence goes on to call the setting "an excellent option" for students anyway.

One detail about the setting itself. Originality.ai's AI Allowance page labels 40% as "Generous" and describes it as "Hybrid AI writing is permitted (but not fully AI-generated content)". For students, that page suggests something else: "Check assignment or school policies (typically 0% AI or 5% AI)". Those were the settings that matched Turnitin least in this test. The two pages are answering different questions (one is about what your course allows, the other about resembling Turnitin). Still, the setting picked for resembling Turnitin is the one the company describes as permitting hybrid AI writing.

This is the number the study leads with. When the 40% setting called an essay Original, Turnitin scored it under 50% in 182 of 189 cases, or 96.3%. The page's opening section turns that into a "96.3% chance of also being treated as human by Turnitin."

That sentence needs one piece of context, and the page has the numbers for it in other sections: Turnitin rarely went over 50% on essays a person wrote. Across the 100 human and 100 AI-polished essays, it crossed 50% on 10. So for an essay in those two groups, simply guessing that Turnitin would land under 50% was right 190 times out of 200, before anyone ran a second detector.

We worked these rows out from the counts on the study page. All 189 Original calls came from these 200 essays, because the setting labelled every fully AI essay as Likely AI, so the three rows describe the same group. An Original result moved the odds from 95% to 96.3%.

Now read it the other way round, which is the direction a worried student cares about. Turnitin put 10 of the polished essays at 50% or more. For 7 of those 10, the 40% setting had said Original. Ten essays is far too few to build a rule on, but it is exactly the case a pre-check exists to catch, and here it caught three. The study's own tips for students put it plainly: "Use the 40% AI Allowance as a comparison point, not a Turnitin preview."

If you open the study yourself, one line near the top describes 96.3% as the agreement on human-written essays. The breakdown lower down calculates it over all 189 essays the setting called Original, human and polished together. On the 100 fully human essays alone, the two tools agreed on every one.

Even a perfect match at 50% would not tell you what your instructor sees. Turnitin's FAQ describes the low end of its scale: "no score or highlights are attributed for AI detection scores in the 1% to 19% range", and an asterisk appears instead. From 20% upward, the report shows a percentage. So an essay Turnitin scored at 30% or 45% counts as "treated as human" in this study, while the instructor's report shows a number. What the asterisk on a Turnitin AI score means covers the low band in detail.

The study says as much: "A score below 50% can still contain text Turnitin marked as likely AI; a score above 50% is not proof of misconduct." What it does not report is how many essays landed between 20% and 49%. For a student trying to guess whether a number will appear at all, that is the missing figure.

At some universities, whether to run your draft through a detector website is now a written rule. Our earlier piece on uploading your draft to an AI detector goes through three of them as they stood on 18 September 2026. In brief:

• Leeds (Academic Integrity and Assistance Policy, approved 8 July 2026 for 2026/27) says its Turnitin tutorial explains "why students should not use AI detector or plagiarism detector websites". This one is addressed to students.
• Leeds' guidance to staff gives three reasons not to use online detection tools: they are "insufficiently accurate for the University to use them as evidence", uploading "may represent a data security breach", and "the document being checked becomes available as training data for other AI tools". It is written about staff checking other people's work, and the third reason is stated in general terms rather than about any named service.
• Reading tells the staff who design and mark assessments that "Student work must not be uploaded into GenAI tools or AI detection tools under the current interim approach." UT Austin prohibits third-party detection software for evaluating student work unless a university contract is in place, and frames it partly as the student's copyright. Its guidance is addressed to the people who teach and mark, although the prohibition itself does not say who it applies to.

None of the three calls it misconduct to check your own work. But at a university with a line like Leeds', a pre-check can go against written guidance and still tell you little you can rely on about Turnitin. Look up your own institution's current page before you paste anything. Originality.ai says a scan with it does not send your paper to Turnitin. What any checking site does with your text after that is set out in its own terms.

1. Know what you can't see. Turnitin's FAQ says "The AI writing detection indicator and report are not visible to students." Right after it comes a way around that: "However, with the PDF download feature, instructors can download and share the AI report with students." Can students see their Turnitin AI score sets out both views side by side.
2. Ask your instructor two things. Is the AI indicator switched on for this assignment, and what is it used for on this course? That answer decides how much any score matters. How to check your work before you submit it to Turnitin lists who else to ask and in what order.
3. If you use another checker anyway, read it on its own scale. Originality.ai explains its own result with an example: a "Likely Original" at the 15% setting "should be read as" AI use that "appears to be 15% or less". Applied to the 40% setting, that reading gives you 40% or less, which says little about an assignment that allows no AI. It also cannot tell you whether Turnitin will show a number at all.
4. Keep the record that answers the real question. Both companies point to the same thing. Originality.ai's help page: "AI detection results should always be considered alongside other evidence of authorship, such as drafts, notes, sources, and document history." Turnitin's FAQ says its percentage "should not be used as the sole basis for action". Drafts and version history stay in your hands, whatever anyone else's settings are.
5. If a score is raised after you submit, ask for the report itself. The full AI Writing Report PDF shows the result and any highlighted passages. A figure from a different detector cannot stand in for it, in either direction.

Will scanning my essay with Originality.ai put it into Turnitin's database? Originality.ai says no: "Running an Originality.ai scan does not submit the paper to Turnitin." It also says it "is not affiliated with Turnitin" and produces only its own reports. What Originality.ai itself keeps is a question for its own terms.

My checker says "Likely AI". Does that mean Turnitin will flag me? Not necessarily. In this study, the 40% setting labelled 11 of the AI-polished essays as Likely AI, and Turnitin put 3 of those 11 at 50% or more.

Would a stricter setting be closer to Turnitin? Not in this test. The 0%, 5% and 15% settings flagged 92% to 95% of the AI-polished essays, while Turnitin put 10% of them at 50% or more, and those settings agreed with Turnitin on fewer than half of all 300 essays.

Does this show which detector is more accurate? No, and the study says so: it "estimates behavioural similarity. It does not establish which detector is more accurate." For what each company documents about its own tool, see Originality.ai vs Turnitin AI detection.

• Originality.ai, What AI Detector Is Most Similar to Turnitin? We Tested 300 Student Essays, Jonathan Gillham, 20 August 2026; read 2 October 2026 (text and Figures 1 to 4).
• Originality.ai help centre, Which Originality.ai AI Detector Is Most Similar to Turnitin?, updated 19 August 2026; read 2 October 2026.
• Originality.ai, Introducing AI Allowance; read 2 October 2026.
• Lukas Gehring and Benjamin Paaßen, Limits of LLM Text Detectors in Education, arXiv 2508.08096, the paper that introduced the GEDE dataset (first version August 2025).
• Turnitin, AI writing detection capabilities FAQs and Using the AI Writing Report; read 2 October 2026.
• University of Leeds, Academic Integrity and Assistance Policy and guidance to staff on the use of artificial intelligence, as quoted in our 18 September 2026 article.
• University of Reading, assessment and feedback guidance, and UT Austin Office of the Provost, AI detection software guidance; read 2 October 2026.