Do AI Detectors Work in Spanish, French and Other Languages?
Which AI detectors support Spanish, French, German and more? Compare Turnitin, GPTZero, Copyleaks and others, and see why translated text gets flagged.

Key takeaways
- Turnitin detects AI writing in English, Spanish, Japanese, Arabic, Dutch, German and Turkish, but not French.
- Turnitin's paraphrasing and humanizer detection work only on English submissions.
- GPTZero, Copyleaks, Originality.ai and Winston AI each list more than 20 languages, Sapling lists 16 and QuillBot six.
- Most non-English accuracy figures come from the vendors themselves, and machine translation can raise false positives.
- Check text in its original language with a detector that supports it, and keep your drafts.
In this guide
- Which languages do AI detectors support?
- Does Turnitin detect AI in Spanish, French or German?
- Do GPTZero and Copyleaks work in other languages?
- What about Originality.ai, Winston AI, QuillBot and Sapling?
- What if your school's detector doesn't support your language?
- Are AI detectors less accurate in other languages?
- What happens when text is translated?
- Why are non-native English writers flagged more often?
- How should you check writing in another language?
- How does The Humanizer handle other languages?
- The bottom line
Many do, with limits. Turnitin detects AI writing in English, Spanish, Japanese, Arabic, Dutch, German and Turkish, but not French. GPTZero, Copyleaks, Originality.ai and Winston AI each cover more than 20 languages. Accuracy can vary by language, though, and translated or second-language writing is more likely to be misread.
Here's which detectors support which languages, what's known about accuracy outside English, why translation causes problems, and how to check writing in another language fairly.
Which languages do AI detectors support?
| Detector | AI detection languages | Worth knowing |
|---|---|---|
| Turnitin | English, Spanish, Japanese, Arabic, Dutch, German, Turkish | Paraphrase and bypasser detection are English only |
| GPTZero | English plus 20+ others, including Spanish, French, German, Italian, Portuguese, Japanese and Hindi | Its non-English accuracy figures come from its own benchmark |
| Copyleaks | 30 languages | Phrase-level AI Logic explanations in six of them |
| Originality.ai | 31 languages | Its own study found round-trip translation raised false positives |
| Winston AI | 23 languages | Says performance can vary by language |
| QuillBot | English, French, Spanish, German, Dutch, Portuguese | Free under 1,200 words |
| Sapling | 16 languages | Other languages go through its English model, which it says is less reliable |
| ZeroGPT | Says it supports all languages | We found no independent tests outside English |
Sources: Turnitin's AI writing FAQ, GPTZero's language list, Copyleaks' documentation, Originality.ai's help center, Winston AI's help center, QuillBot's AI detector page, Sapling's detector page and ZeroGPT's homepage. These lists change often, so check the tool's own page before you rely on it.
Does Turnitin detect AI in Spanish, French or German?
Spanish and German, yes. French, no.
Turnitin has added languages one at a time, each with its own model:
| Language | AI writing detection added |
|---|---|
| English | April 2023 |
| Spanish | September 2024 |
| Japanese | April 2025 |
| Arabic | August 2026 |
| Turkish, German and Dutch | October 2026 |
According to Turnitin's FAQ, papers in unsupported languages aren't processed: the indicator shows an error state and no report is generated. That currently includes French, Italian and Portuguese. Turnitin's paraphrasing and bypasser detection run only on English, so its non-English detectors look for likely AI-generated text but not text altered by paraphrasers or humanizers. We explain bypasser detection in can Turnitin detect AI humanizers.
Each language model is trained separately. Turnitin's release notes say its Spanish model was first trained to detect GPT-3.5 and GPT-4 output, and a May 2026 update improved detection of newer models. If you write in Spanish, Japanese or Arabic, Turnitin may well score your work, so the same advice about drafts and version history applies.
Do GPTZero and Copyleaks work in other languages?
Yes. GPTZero lists English plus more than 20 languages, including Spanish, French, German, Italian, Portuguese, Russian, Japanese, Korean, Chinese, Arabic and Hindi. In a May 2025 update, GPTZero published internal benchmark results for French, Spanish, German and Portuguese, with false positive rates under 1% in each. Those are GPTZero's own numbers on its own test set. For how it performs in English, see is GPTZero accurate.
Copyleaks lists 30 languages for AI text detection and says it detects the language automatically. Its AI Logic feature, which shows the phrases behind a result, works in six of them: English, Spanish, French, Portuguese, German and Italian.
What about Originality.ai, Winston AI, QuillBot and Sapling?
- Originality.ai lists 31 languages, from Bulgarian and Danish to Urdu and Vietnamese.
- Winston AI supports 23 languages and says "detection performance can vary by language because writing patterns, available training data, and AI model behavior differ across languages."
- QuillBot supports six languages: English, French, Spanish, German, Dutch and Portuguese. We look at its accuracy in is QuillBot's AI detector accurate.
- Sapling scores 16 languages and runs anything else through its English model, which it says is less reliable.
What if your school's detector doesn't support your language?
If your school uses Turnitin and you write in French, Italian or Portuguese, Turnitin won't produce an AI writing report for that paper. That doesn't mean nobody checks. An instructor might run the text through another tool that supports the language, compare it with your earlier work or ask about your process.
So assume your work can still be reviewed, and keep the same evidence you would for an English essay: outlines, notes, sources and version history. If you're not sure which tools your course uses, ask. Knowing the tool tells you which language rules and limits apply to you.
Are AI detectors less accurate in other languages?
They can be, and the honest answer is that there's much less independent evidence outside English. Most of the non-English accuracy numbers you'll find come from the detector companies. Winston AI and Sapling both say plainly that results can vary or weaken by language.
There are good reasons to expect differences. Detectors learn from examples of human and AI writing, and the mix of examples differs from one language to the next. The AI models a detector was trained on may not match the ones people use in a given language. And formal academic prose in any language leans on standard phrasing, which can look predictable to a model. We explain that mechanism in why AI detectors have false positives.
What happens when text is translated?
Translation is where the clearest problems show up.
- Machine translation can trigger false positives. In a 2023 study of 14 detectors, accuracy on human-written English was 96%, but it dropped by 20% for human texts written in other languages and machine-translated into English. The authors warned this leaves second-language students "at risk of being falsely accused" when they translate their own work.
- Round trips are worse. Originality.ai's own 2026 translation study found that human writing translated directly into Spanish or Portuguese was rarely flagged, but text translated there and back into English was falsely flagged about 28% of the time.
- Scan the original. Winston AI advises that "if authorship matters, scan the original-language document whenever it is available."
Translation can also raise a policy question. Some courses count machine translation as AI assistance, and the MLA Style Center recommends acknowledging AI tools used for "translating words." Check your policy before you translate your own essay with AI.
Why are non-native English writers flagged more often?
When you write in a second language, you tend to use common words and safe sentence structures. Detectors read that as predictable, which is a core signal of AI text. A Stanford study found seven detectors labeled an average of 61.22% of TOEFL essays by non-native English speakers as AI-generated, while scoring essays by US-born eighth-graders almost perfectly.
If that's you, our page for ESL writers explains how to keep your writing clear while giving it more natural variation, and why is my writing flagged as AI covers other common causes. The underlying mechanics are in how AI detectors work.
How should you check writing in another language?
- Use a detector that supports your language. If it doesn't, the result may be meaningless or come from an English model.
- Check the original, not a translation. Translating to English just to run a check adds noise.
- Test full sections. Short passages swing more in every language.
- Compare tools. Agreement between two detectors that support your language means more than one score. Our comparison of AI detectors covers the main options.
- Keep your drafts. Version history is evidence in any language.
For more research on detector errors, see our AI detection statistics.
How does The Humanizer handle other languages?
The Humanizer works with English, Spanish, French, German, Russian, Japanese, Hindi and more. It rewrites sentence rhythm, structure and word choice while keeping your meaning, facts, citations, key terms and paragraph count. It's designed to read naturally and pass common detectors, though results vary by text and language. Our multilingual AI humanizer page has the details.
The bottom line
AI detectors increasingly work beyond English, but coverage varies: Turnitin handles seven languages and not French, while GPTZero, Copyleaks, Originality.ai and Winston AI each list more than 20. Independent accuracy evidence outside English is thin, translation can raise false positives, and second-language writers are flagged more often. Check in the original language and keep your drafts.
To make AI-assisted writing in your language read naturally, try The Humanizer free with 350 words a month.
Frequently asked
Questions, answered
Written by
The Humanizer Team
Editorial


