Can Turnitin detect AI in non-English submissions?
The language question splits into three that get answered as one, and the wrong one usually gets answered first. There is what language the file is in. There is what the detector supports. And there is what happens when you wrote in one language and submitted in another, which is the situation most people asking are actually in.
HumanPen Team
· 6 min read
The short answer
Seven long-form languages, as of 6 October 2026. That is when Turnitin added Dutch, German, and Turkish alongside English, Spanish, Japanese, and Modern Arabic in its official FAQ. Actual availability depends on whether your institution licenses the capability and has it enabled. The 7 October 2026 snapshot of its file requirements page listed English, Spanish, Japanese, and Arabic, while the page now lists seven, so check the date on anything you read about this. Submissions in languages outside those seven fall outside what an AI Writing Report is documented to handle. And the rule applies to the file you submit, so a paper drafted in Portuguese and submitted in English is an English submission.
The supported-language list, verbatim
From Turnitin's file requirements for an AI Writing Report, as that page stood on 7 October 2026:
"File must be written in a supported language: English, Spanish, Japanese, Arabic"
Alongside it, on the same short list: "Accepted file types: .docx, .pdf, .txt, .rtf". That file requirements page lags behind Turnitin's product releases and should not be taken as the full current list.
Turnitin's product updates page notes that on 6 October 2026, AI writing detection expanded to Turkish, German, and Dutch for existing customers licensing Turnitin Originality or iThenticate 2.0 with AI detection enabled, running alongside its English, Spanish, Japanese, and Arabic detectors. Turnitin's official AI writing detection capabilities FAQ explicitly lists all seven long-form languages: English, Spanish, Japanese, Modern Arabic, Dutch, German, and Turkish. If you click through to the 7 October 2026 snapshot of the requirements page and count four (English, Spanish, Japanese, and Arabic), the page now lists seven. Languages like Chinese, Portuguese, and Korean remain unlisted across these official pages.
The detectors are not identical to each other
This is the part that rarely makes it into the summaries. Turnitin's official documentation notes that these detectors run as separate models alongside one another, with the system selecting the appropriate model based on the submission's language. The main FAQ page covering AI writing detection has historically scoped its core answers to English submissions, while the vendor maintains distinct models for non-English languages rather than routing all text through a single pipeline.
Crucially, the capability that looks for text modified by paraphrasing or bypasser tools has documented limits. Turnitin's documentation on using the AI Writing Report states that only its English AI detector includes AI paraphrasing and AI bypasser detection capabilities; stated conservatively, the cited documentation explicitly names English, Spanish, and Japanese, and the page does not establish coverage for the newer Dutch, German, or Turkish models.
Do not read that as an opening. Read it as a reminder that "Turnitin's AI detection" is not one uniform thing, and that anything you have read about how it behaves was almost certainly written about the English detector.
Translating into English makes it an English submission
The requirement is about the file. If you drafted in your first language and the document you upload is in English, it is the English text that gets segmented and scored, regardless of where the ideas were first written down.
Which raises the question people are really asking: does machine-translated prose look like model output? Turnitin has not published anything that answers it directly, and nobody outside Turnitin has the access it would take to answer it properly. What Turnitin has published is the list of properties that turn up in its own false positives:
"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."
Translated academic prose can land on all three, not because the writer did anything wrong, but because translation tends to regularise sentence shape. That connection is our inference, not Turnitin's statement.
Turnitin's own next sentence is the one to carry into a meeting with a supervisor: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated." The vendor is on record telling the people who read these reports to factor that in.
What Turnitin says it did about bias
Its FAQ says the training sample deliberately included "statistically under-represented groups like second-language learners, English users from non-English speaking countries, students at colleges and universities with diverse enrollments and less common subject areas such as anthropology, geology, sociology, and others to minimize bias when training our model".
That is the company's account of its own process, and it comes with no published bias figures. Treat it as a stated intention. The separate question of who actually gets flagged more often has its own evidence, and we went through it in why non-native English writers get flagged more often.
Short documents behave worse, in any language
Before you assume a two-page reflection is the low-risk case:
"In shorter documents where there are only a few hundred words, the prediction will be mostly 'all or nothing' because we're predicting on a single segment without the opportunity to overlap."
The line after that: "This means that some text that is a mix of AI-generated and original content could be flagged as entirely AI-generated." Short pieces are where translation-heavy writing and all-or-nothing scoring meet, and that combination produces the reports that are hardest to argue with.
Frequently asked questions
My paper is in Chinese. Will it get an AI score? Chinese is not on the supported-language list for an AI Writing Report. As of 6 October 2026, the official FAQ lists seven long-form languages: English, Spanish, Japanese, Modern Arabic, Dutch, German, and Turkish, and availability depends on institutional licensing and configuration. Whether your institution's setup does anything else with unlisted languages is a matter for the institution, not something Turnitin's documentation guarantees.
I wrote it myself, then used a translation tool. Is that AI-generated? Those are two different questions and only your institution answers the first one. Disclosure rules increasingly cover translation as well as drafting, so read the actual policy rather than assuming translation is exempt.
Does the Spanish detector work the same way as the English one? Not identically, and neither do the other non-English detectors (such as Japanese, Arabic, German, Dutch, or Turkish). Turnitin documents these as separate models operating alongside the English detector; the cited documentation explicitly names English, Spanish, and Japanese regarding paraphrase and bypasser detection, without establishing coverage for newer models like Dutch, German, or Turkish.
My university uses Turnitin, but I have never seen an AI percentage. You would not. Turnitin says only instructors and administrators can see the indicator, though instructors can download the report as a PDF and share it. And AI detection requires a specific licence, so some institutions have none at all.
Does the model try to identify which AI tool was used? No. It classifies segments by probability of being likely human or likely AI-generated (more on what that measurement is).
KEEP READING