Direct answer
Partly, but far less reliably than in English. Turnitin's AI writing detection was built and validated on English prose, and its own guidance states that the model's output is only meaningful for English-language submissions [1]. Most other detectors report some multilingual support, yet accuracy drops sharply outside high-resource languages, and short or heavily translated text produces the most unstable results [1]. For a multilingual student, the practical takeaway is that "not English" does not mean "not detectable" — it means the signal is weaker, more variable, and more likely to be misread in either direction.
Which Languages Can AI Detectors Actually Detect?
Coverage is uneven, and the gap is structural rather than a temporary bug. Detection models learn from large corpora of human and machine text; English dominates those corpora, so the classifier has seen millions of English examples and comparatively few Swahili, Bengali, or Tagalog ones [2]. Turnitin's published guidance is explicit that its AI writing indicator is designed for English and that results for other languages should not be treated as a reliable score [1]. That is a deliberate scope decision, not an oversight.
Where vendors do claim multilingual support, the supported list clusters around high-resource European and East Asian languages — Spanish, French, German, Portuguese, Chinese, Japanese, Korean — with markedly weaker performance in low-resource languages [2]. Even inside that list, published evaluations show detection accuracy varying by language and by genre, with non-English recall trailing English by a wide margin [2]. Treat any "supports 100+ languages" badge as a statement about input acceptance, not about equal accuracy.
A third factor is document type. Academic writing in a second language often contains formulaic phrasing, repeated connectives, and standardised section structures — exactly the surface features a detector associates with machine generation [2]. This is why well-documented false positives cluster around non-native English writers, translated abstracts, and reference-heavy methodology sections rather than around fluent native prose.
How Accurate Are AI Detectors on Non-English Text, and Why Do They Fail?
Accuracy on non-English text is the weakest link in the entire detection chain, and the reasons are technical. Classifiers depend on tokenisation and perplexity signals calibrated to one language's statistical patterns; when a model trained mostly on English is pointed at translated or genuinely foreign text, the resulting probability is closer to noise than to evidence [3]. Independent testing has repeatedly found that detector verdicts shift when the same content is rendered in a different language, even though the underlying authorial process did not change [3].
The second failure mode is translation itself. Machine translation output carries its own machine-like regularity, so a human paragraph passed through a translator can acquire the very features detectors look for, while an AI paragraph translated into a low-resource language can lose them [3]. Neither outcome reflects what the student actually did. This is the core reason institutions are advised to treat a detector percentage as one indicator among several rather than as a finding of misconduct [1][3].
Institutional guidance reflects that caution. Universities increasingly instruct staff not to base an academic-integrity decision on an AI score alone, and to consider drafting history, version files, and a conversation with the student — precisely because the tool's confidence is language-dependent [3]. If your draft is not in English, assume the number you see is a rough signal at best.
Can Translating or Rewriting Text Into Another Language Hide AI Use From Turnitin?
Translation is not a safe workaround, and treating it as one is a common and costly mistake. Turnitin's detection operates on the submitted document, so whatever language you submit is the language being assessed; submitting a translated version simply moves your text into a lower-confidence zone rather than out of scrutiny [4]. Institutions that suspect translation or undisclosed machine assistance can and do request drafts, version history, or an oral explanation, and a mismatch between your written fluency and your spoken command of the language is a well-known red flag [4].
There is also a submission-side risk that students underestimate. Turnitin's similarity checking compares your file against its repositories and the wider web, and translated passages can still surface as matches against published sources in the target language [4]. Meanwhile, the AI indicator on non-English text may show nothing at all — which is not the same as a clean result, because a blank or low-confidence signal means the tool could not make a determination [1][4]. Reading that blank as "cleared" is the single most common misinterpretation of a multilingual report.
The defensible approach is to know what your own report says before your department does. That means checking the actual document you intend to submit, in its final language, and reading both the AI indicator and the similarity matches together rather than in isolation [4]. If the AI signal is ambiguous, the honest move is to keep your drafting evidence — outlines, earlier versions, notes — and be ready to explain your process.
If you want to see how your own multilingual draft is actually read before it reaches your department, turnitin0 lets you upload the exact file you plan to submit and returns the same two PDFs your instructor sees — the AI detection report and the similarity report, delivered together, usually in under 15 minutes.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does Turnitin detect AI writing in Spanish or French?
Turnitin's AI writing indicator is designed and validated for English, and the company advises against relying on it for other languages [1]. Spanish and French submissions may be accepted, but the resulting percentage is not a validated measure of authorship.
Why did my non-English essay get flagged when I wrote it myself?
Non-native phrasing, formulaic academic connectives, and translated passages all resemble the surface patterns detectors associate with machine text [2]. A flag is a prompt for review, not a conclusion, which is why institutions are told not to act on a score alone [3].
Does translating AI text into another language make it undetectable?
No. It moves your text into a lower-confidence zone rather than removing it from scrutiny, and translated passages can still match published sources in similarity checking [4]. Translation also creates an explainable gap between your written and spoken fluency.
What does a blank AI score on a foreign-language submission mean?
It usually means the tool could not make a reliable determination, not that the text was cleared [1][4]. Read the AI indicator together with the similarity report before drawing any conclusion.
Can I preview the report for a non-English draft before submitting?
Yes — turnitin0 checks the exact file you plan to submit and returns both the AI detection report and the similarity report as downloadable PDFs, without adding your file to Turnitin's student paper database.