Does Turnitin Detect AI in Arabic? The Official Answer

Short answer: no. Turnitin's official documentation, updated in June 2026, limits AI writing detection to long-form English, Spanish and Japanese. A file written in an unsupported language is not processed at all — no percentage, no report, an empty indicator.

That gap is where most students misread their situation. The question is not "does Turnitin catch AI in Arabic." The real question is "which of Turnitin's systems is running on my file" — because AI detection is one of three independent systems, and two of the others have supported Arabic for years.

Does Turnitin scan Arabic submissions for AI writing?

No. The official AI writing detection FAQ names three supported languages and no more: English, Spanish and Japanese, and only for long-form prose. AI paraphrasing detection and AI bypasser detection are narrower still — English submissions only.

The documentation also spells out what happens with an unsupported language. The detector does not process the submission. The instructor sees an empty state with in-app guidance explaining the feature works for supported languages only. No report is generated at all.

Decision rule: if your paper is entirely in Arabic, any number a third-party site shows you labelled "your Turnitin AI score" did not come from Turnitin. There is no number to show.

What does a grey indicator or an asterisk actually mean?

A grey indicator with no percentage (- -) does not mean cleared. It means the file was never processed, and the documentation lists the specific reasons:

• The file is in an unsupported language.
• It contains fewer than 300 words of prose.
• It exceeds 30,000 words, or 100 MB in size.
• The file type falls outside the accepted list (docx, pdf, txt, rtf).
• It was submitted before the feature launched and needs resubmitting.

A separate case confuses people just as often: the asterisk (*%). When detection lands between 1% and 19%, Turnitin displays no percentage at all and shows an asterisk instead. That is a deliberate choice to avoid false positives in the low band, not a glitch.

If my Arabic text isn't scanned, what is?

Two systems, neither of which cares which language you wrote in.

Text matching, including translated matching. The matching index covers Arabic alongside dozens of other languages. Picture the practical case: a student asks ChatGPT for a paragraph in English, runs it through machine translation, and pastes the Arabic into a paper. The AI detector never sees it. Translated matching may still link that Arabic passage back to the English source it came from.

Authorship analysis. Turnitin describes this as using file metadata plus forensic linguistic analysis. Crucially, it does not claim text was AI-generated. It flags that a submission may not be the student's own work. Different mechanism, different finding, different conversation.

One more detail worth internalising: the Similarity score and the AI writing percentage are completely independent and do not influence each other. Treating a clean score on one as protection on the other is a misreading of the report.

When does your paper move from "unscanned" to fully scanned?

The moment it becomes an English document. That is not an edge case across Arab universities — many graduate programmes, and most medicine, engineering and business departments, require submissions in English.

At that point three models run together on the same file: one for likely AI-generated text, one for text that appears to have been run through an AI paraphraser, and a third for text processed by a humanizer or bypasser tool. The last two work on English only.

A detail many people miss: analysis covers prose exclusively. Bullet lists, tables, code and annotated bibliographies sit outside the calculation, which is why the reported percentage sometimes looks out of proportion to the amount of highlighted text.

Turnitin's own testing draws another useful boundary. Grammar and spelling corrections from Grammarly were generally not flagged as AI writing. Output from Grammarly's generative features — draft generation, paraphrasing, summarising — likely will be.

Can you see your score before you submit?

No. The AI writing indicator and report are visible to instructors and administrators only, never to students. An instructor can download the report as a PDF and share it with you if they choose. That is the entire access path.

This is why sites promising a "pre-submission Turnitin check" deserve scepticism. Nobody outside a licensed institution has access to that model. What any other tool gives you — ours included — is a probabilistic style signal, not a copy of your university's report. You can check your Arabic text's style with our detector to see where your writing reads as unusually uniform, but treat the number as input rather than verdict. For how the two approaches differ, see our comparison with Turnitin.

Why do detectors misfire on Arabic specifically?

Detection output is always probabilistic, and error rates run higher for people writing in a second language. Arabic adds structural obstacles on top of that:

• Diacritics. A peer-reviewed study in the journal Information tested how the marks above and below Arabic letters affect detection, and recorded 62.7% accuracy for GPTZero on the AIRABIC benchmark — close to coin-flip territory — while its own models trained on diacritised text exceeded 98%.
• The regularity of Modern Standard Arabic. Careful academic Arabic tends toward balanced structures and recurring vocabulary. That same regularity is precisely what detectors read as a machine signature.
• Root-based morphology. Models trained on Latin-script languages handle Arabic roots and their derivations with logic that does not fit the language.

This applies to our own detector too, and we say so plainly on our methodology page. For a deeper look at the linguistic problem, read our piece on the Arabic AI detection challenge, and for what "accuracy" actually means in this field, see what AI detector accuracy really means.

What should you actually do before submitting?

Four steps, in order of impact:

1. Ask your department a specific question: what does the policy count as permitted assistance, and what does it count as authorship? The line between those two is what gets referred to a disciplinary committee, and it moves from institution to institution.
2. Keep your version history. A Google Docs or Word file with weeks of edit history is the strongest practical evidence that you wrote the thing yourself — and it works regardless of which language you wrote in.
3. Disclose your use of AI in whatever form your regulations require. Disclosure converts the issue from an accusation into a documented procedure.
4. Review your style, not your score. If your paragraphs are all the same length and your vocabulary keeps circling the same words, the problem is in the writing before it is in the detector.

A fuller set of steps lives in our responsible pre-submission checklist.

Frequently asked questions

Does translating from English into Arabic hide the source?

Not reliably. Translated matching is an existing Turnitin feature, Arabic is among the languages it covers, and its whole purpose is linking translated text back to its original source.

Will Turnitin support Arabic later?

Turnitin staff indicated on their official forum in 2025 that work on additional languages including Arabic was underway. As of the June 2026 documentation update, the supported list still holds at three. Check the official page before building any assumption on it.

Does my university even use AI detection?

Not necessarily. The feature requires a separate licence, and an administrator can disable it from account settings. Several institutions have switched it off after internal debate about false positives. Ask your department directly rather than guessing.

Is a percentage enough to prove misconduct?

Turnitin says it is not. Its documentation repeats that the company makes no determination of misconduct, that the percentage should not be the sole basis for any action, and that the final decision rests with a human reviewer applying institutional policy.