AI detectors ranked by languages supported

Field sorted: the language count on each card, from the vendor's own page read on 8 November 2026. Direction: more ranks higher. Tie rule: more languages named individually ranks first, so Copyleaks (11 named) sits above Originality.ai (0 named) at 30; Undetectable AI's 50+ stood alone at the top for part of 9 November 2026 until Smodin's 100+ was indexed with the fourth wave; at 30, Copyleaks (11 named) and Originality.ai (0 named, 5 study records) sit above TruthScan (4 named, 0 records) on study records and then name; at 1, BrandWell (2 records) sits above Scribbr (1) and Surfer (0). Qualification rule: the vendor states a count or a countable list; ZeroGPT's claim of all languages cannot be counted, so that card and the eighteen with no count are unranked. "30+" is recorded as 30 and "50+" as 50, with the plus shown. 13 of 28 qualified after the third wave; 19 of 38 qualified after the ten cards of the fourth wave, six of them (Smodin, Compilatio, TruthScan, Quetext, BrandWell and Surfer) with a count; 31 of 72 qualify after the fifth wave, Resemble Detect (54) and Pindrop Pulse (40) joining with counts stated for spoken audio. Full rules: methodology page; both studies cited are in the studies index. What the two studies and Orenstrakh 2023 measured on machine-translated and Spanish text is set out, with bases, on the question page on translated text.

The non-English evidence ranking, added 9 November 2026, counts the other side of this field: the study records that print a result on non-English, translated or non-native text, GPTZero first with five, and only nine cards appear on both pages.

Vendors publish different counts on different pages. Walter Writes states 20+ on the detector page and over 80 in the homepage FAQ, where the figure covers the humanizer bundle (walterwrites.ai). Pangram's 20+ headline sits above a 24-name list ending "and more" (pangram.com). Both read 8 November 2026. The index takes the detector product page, so each count is reproducible from one URL.

One vendor dates its languages: Turnitin's 7 carry release notes for English (4 April 2023), Spanish (12 September 2024) and Japanese (2 April 2025), set out on the Turnitin languages question, added 9 November 2026. The localisation teams view, added the same day, reads the 14 cards stating 10 or more languages beside their non-English study records.

Resemble Detect's 54 and Pindrop Pulse's unnamed 40 or more are set side by side in the Pindrop vs Resemble Detect comparison, read 10 November 2026.

What do studies say about non-English and translated text?

No study in the index tests a detector in a non-English language; two test English-language tools under non-native conditions. On non-native English text, Liang 2023 reports an average false positive rate of 61.22% across seven detectors on 91 human-written TOEFL essays, against near-perfect accuracy on 88 US essays (arXiv 2304.02819, read 8 November 2026; Liang 2023 record). On machine-translated text, Weber-Wulff 2023 placed 9 translated human documents among 18 and found 6 of GPTZero's 9 false accusations on them (arXiv 2306.15666, read 8 November 2026; Weber-Wulff 2023 record). Proofademic states a 0.5% non-native false positive rate on its research page, a vendor figure (proofademic.ai, read 8 November 2026).

Smodin

• Verdict: Fits readers whose language appears on no other card, with the count read as a claim.
• Value: 100: the detector page header states "100+ languages" and names English, Spanish, Portuguese, Swedish, Norwegian, Dutch and Greek; the FAQ on the same page says Smodin supports 12 languages, and the API language parameter lists 104 codes (smodin.io, docs.smodin.io, read 9 November 2026).
• Best for: A first check on text in a language the other cards do not state.
• Main limitation: Three figures on two pages (100+, 12 and 104); no indexed study tested Smodin in any language.
• Link: Smodin card

Compilatio

• Verdict: Fits institutions whose students write in French, Spanish, Italian or Portuguese.
• Value: 35: the AI detector page states more than 35 languages and names French, English, Spanish, Italian, Portuguese, Russian, Arabic and Hindi; the support article lists 24 languages used in the vendor's reliability test (compilatio.net, support.compilatio.net, read 9 November 2026).
• Best for: Universities in France, Spain and Italy buying through an LMS.
• Main limitation: Weber-Wulff 2023, the one record, tested English documents only; the Magister page states +40 languages for the product as a whole, a different count.
• Link: Compilatio card

TruthScan

• Verdict: Fits API users who want a stated count beside an English-first caveat.
• Value: 30: the FAQ states support for over 30 languages and names English, Spanish, French and German, adding that the detector "provides the highest accuracy for English text" (truthscan.com, read 9 November 2026).
• Best for: Teams that want a results-based free plan and a multi-language text API.
• Main limitation: No list of the 30; the same FAQ pairs the count with a 95%+ figure that carries no base; no indexed study tested TruthScan.
• Link: TruthScan card

Quetext

• Verdict: Fits readers checking short Japanese, Korean or Arabic text free.
• Value: 25: the detector page and pricing FAQ state 25 supported languages and name Spanish, French, German, Arabic, Hindi, Japanese and Korean; a bullet on the pricing page says 17+ (quetext.com, read 9 November 2026).
• Best for: Students and editors checking 1,000 words at a time in a named language.
• Main limitation: Two counts on two pages; no indexed study tested Quetext.
• Link: Quetext card

BrandWell

• Verdict: Fits English text only, as the archived page states.
• Value: 1: the archived FAQ of 9 August 2026 says "Currently, the AI Detector is just for the English language"; the live Moxby Mod listing names no language (web.archive.org, read 9 November 2026).
• Best for: English content teams reading the Content at Scale record.
• Main limitation: The detector page redirects to a Moxby Mod since 29 August 2026, so the English-only statement is from an archived page.
• Link: BrandWell card

Surfer

• Verdict: Fits English marketing copy, as the FAQ states.
• Value: 1: the detector page FAQ says "Surfer's AI detection tool has only been trained on content written in English" (surfer.ai, read 9 November 2026).
• Best for: SEO teams already working in Surfer's editor.
• Main limitation: English only, and no indexed study tested Surfer.
• Link: Surfer card

Resemble Detect

• Verdict: Fits teams that check spoken audio and want the languages named rather than counted.
• Value: 54: the benchmarks page states "54 languages detected from Arabic to Spanish", names 30 and adds "over 20 additional languages", with the test described as "Validated against MLAADv10" (resemble.ai/benchmarks, last updated August 2026, read 9 November 2026).
• Best for: Trust-and-safety and contact-centre teams screening recordings in several languages.
• Main limitation: The count is for speech, not written text, and the validation is the vendor's own; no indexed study tested Resemble Detect.
• Link: Resemble Detect card

Pindrop Pulse

• Verdict: Fits contact centres that take calls in many languages and buy through sales.
• Value: 40: the Pulse Inspect launch release of 15 August 2024 states the model was trained on "40+ languages, covering over 90% of the internet's spoken languages"; no list is published (pindrop.com, read 9 November 2026).
• Best for: Enterprises adding synthetic voice detection to an existing Pindrop deployment.
• Main limitation: The figure describes training data for speech, with no language named; no public price, no free tier, and no indexed study tested Pindrop Pulse.
• Link: Pindrop Pulse card

Isgen

• Verdict: Fits readers who want the second-largest stated count, with 16 languages named on the vendor's word alone.
• Value: 80+ languages stated, 16 named: English, Portuguese, German, French, Spanish, Ukrainian, Filipino, Arabic, Chinese Simplified, Japanese, Korean, Chinese Traditional, Dutch, Italian, Russian and Turkish (isgen.ai, read 9 November 2026).
• Best for: Readers who check short texts in a named non-English language free, up to 5,000 characters per scan before sign-in.
• Main limitation: 64 of the 80+ are unnamed, the vendor's 96.4% accuracy claim names no benchmark, size or date, and no study in the index tested Isgen in any language.
• Link: Isgen card

AI Detector Pro

• Verdict: Fits readers who want a named list of seven, with the caveat that the vendor's own pages disagree on the count.
• Value: 7 languages named in the FAQ: English, Arabic, French, Portuguese, Korean, Spanish and German (aidetector.pro/pages/faq, read 10 November 2026); the pricing page's feature line names five and the home banner six plus English, so the card records the longest list.
• Best for: Readers checking a document in one of the six non-English languages named.
• Main limitation: Three pages print three counts; no study in the index tested AI Detector Pro; the free plan is 3 AI scans a month (aidetector.pro/pricing, read 10 November 2026).
• Link: AI Detector Pro card

PlagiarismCheck

• Verdict: Fits Spanish-language institutions that want the detector inside an LMS-integrated plagiarism suite.
• Value: 2 languages, English and Spanish (plagiarismcheck.org/ai-detector, read 9 November 2026).
• Best for: Spanish and English coursework through Canvas, Moodle, Blackboard, Brightspace, Schoology or Google Classroom.
• Main limitation: No Spanish result exists in the index; the one record, Weber-Wulff 2023, tested English documents, 9 of them machine-translated from seven other languages, and ranked the tool 13th of 14 (arXiv 2306.15666, read 9 November 2026).
• Link: PlagiarismCheck card

Content Detector AI

• Verdict: Fits readers who want the stated scope of a tool whose page is now archived.
• Value: 1 language, English only, on the archived page of 11 May 2026 (web.archive.org, read 9 November 2026); contentdetector.ai returned 301 to a Moxby Mod on 9 November 2026.
• Best for: Readers tracing the three study results of an English-only free detector.
• Main limitation: The live address redirects, so the stated scope is the archived page's; none of the three records tested the tool in another language.
• Link: Content Detector AI card

GPTKit

• Verdict: Fits readers who want one stated language and one outside Spanish result to read against it.
• Value: 1 language, English: "Currently English is supported" (gptkit.ai/faq, read 9 November 2026).
• Best for: English coursework, the text on which Orenstrakh 2023 recorded 0 of 114 false positives.
• Main limitation: Orenstrakh 2023 also ran 10 human and 10 ChatGPT Spanish submissions and GPTKit classified 80% and 20% correctly (arXiv 2307.07411, Table 12, read 9 November 2026), a result on a language the vendor does not claim.
• Link: GPTKit card

Paperpal

• Verdict: Fits readers checking English academic text who accept a vendor count of one.
• Value: 1 language, English; the detector page describes scholarly text from ChatGPT, Claude and Gemini and names no other language (paperpal.com/tools/ai-detector, read 10 November 2026).
• Best for: Researchers checking an English manuscript before submission, 5 scans a day of 1,200 words free.
• Main limitation: No study in the index tested Paperpal, and the vendor's 99% (accuracy) claim on Gemini, Claude and ChatGPT text is its own figure with no base printed.
• Link: Paperpal card

Check For AI

• Verdict: Fits readers who want the stated scope of a tool whose pages exist only in the archive.
• Value: 1 language, English: "Our models are only trained for English text" (web.archive.org, learn more page, 11 December 2023, read 10 November 2026).
• Best for: Readers reading a 2023 risk score on English coursework.
• Main limitation: The service was sunset in July 2023; the one record, Weber-Wulff 2023, tested English documents, 9 of them machine-translated (arXiv 2306.15666, read 10 November 2026).
• Link: Check For AI card

Passed.ai

• Verdict: Fits readers who want the vendor's one-language statement for a closing product.
• Value: 1 language, English: the home page FAQ says the AI detection and plagiarism check "can only be performed on English documents" while the audit tools work regardless of language (web.archive.org, 7 June 2026, read 10 November 2026).
• Best for: Teachers reading an English document audit from an existing account.
• Main limitation: No study in the index tested Passed.ai, the detector was Originality.ai's engine, and every page of passed.ai served a closing notice on 10 November 2026 (passed.ai, read 10 November 2026).
• Link: Passed.ai card

SEO.ai

• Verdict: Fits readers who want a free check in one of 46 named languages, the Undetectable AI reading of a 50+ claim.
• Value: 50 languages: the page states "50+ languages" and names 46, from Arabic, Basque and Bengali to Turkish, Ukrainian and Vietnamese (seo.ai/tools/ai-content-detector, read 10 November 2026); 50 is the stated floor, as for Undetectable AI.
• Best for: Non-English marketing copy checked without an account.
• Main limitation: Walters 2023 tested English papers only and printed 12% of 42 human papers called AI (DOI 10.1515/opis-2022-0158, Table 4, read 8 November 2026); no record tests any of the 46 named languages.
• Link: SEO.ai card

McAfee Deepfake Detector

• Verdict: Fits readers who want the vendor's one-language statement for an on-device voice tool.
• Value: 1 language, English: both press releases say the detector "is available for English language detection" and the site footnote reads "Video Scams is currently trained only on English audio. Scanning non-English audio in videos might give less accurate results" (mcafee.com newsroom, 3 January 2025, mcafee.com/ai/deepfake-detector, read 10 November 2026).
• Best for: English-language video watched in the browser on a supported PC.
• Main limitation: The count is for audio, not text; no study in the index tested the detector, and the 96% figure is a 2024 press release claim (mcafee.com newsroom, 21 August 2024, read 10 November 2026).
• Link: McAfee Deepfake Detector card