AI Detector for Tagalog: Which Tools Actually Work on Filipino and Taglish?

Short answer: as of September 2026, none of the major AI detectors officially supports Tagalog or Filipino. Turnitin doesn't run AI-writing detection on Tagalog at all, so a Tagalog paper gets an empty AI indicator instead of a score. Originality.ai, GPTZero, Copyleaks and Pangram publish language lists that leave Tagalog out, and the tools that claim "all languages" never name it.

So any "AI score" you get on Tagalog or Taglish text comes from a model working outside its documented languages. That doesn't make every result wrong, but it does mean nobody has told you how accurate it is. Our own tests, described below, show how far those scores can swing.

Below: what each detector actually documents, what code-switching does to language detection, what we measured on Tagalog, and what CHED's new AI order means if you study at a Philippine university.

Which AI detectors list Tagalog? (checked September 2026)

Every row below comes from the vendor's own page, opened in September 2026.

The details matter more than the ticks.

Turnitin says its AI writing detection works on "long-form English, long-form Spanish, long-form Japanese, and long-form Modern Arabic." If a paper in any other language is submitted, "the detector will not process the submission," the indicator shows an empty or error state, and "no report will be generated" (Turnitin AI writing FAQ). Its checks for AI-paraphrased or "bypassed" text are English only. Turnitin describes the Similarity Report and the AI percentage as "completely independent."

Originality.ai lists 31 languages for its multi-language model, including Indonesian, Hindi, Thai and Vietnamese, but not Tagalog or Filipino (Originality.ai help centre, updated 21 September 2026).

GPTZero says it "fully supports English, German, Portuguese, French and Spanish" (GPTZero). Its FAQ adds that "the majority of our dataset is in English prose, written by adults" (GPTZero FAQ). A separate help article lists English plus 23 further languages, from Arabic to Vietnamese; Tagalog isn't one of them (GPTZero support).

Copyleaks advertises "over 30 supported languages" (Copyleaks AI Detector). Its API documentation lists 30 ISO codes, from English to Vietnamese, with no Tagalog or Filipino code (Copyleaks API docs).

Pangram lists 24 languages, including Hindi, Persian and Urdu, but no Philippine language (Pangram knowledge hub).

QuillBot says its AI detector "supports 20+ languages" and names English, Spanish, French and Portuguese "and many others," without publishing a full list (QuillBot AI Detector); a QuillBot guide revised in July 2026 lists only six: English, French, Spanish, German, Dutch and Portuguese (QuillBot AI Detector guide). ZeroGPT's homepage says it supports "all the languages with the highest accuracy rate of detection" but names none and shows no per-language results there (ZeroGPT).

Leaving a language off a list doesn't prove a tool fails on it. It means the vendor hasn't claimed it works, and hasn't told you how often it's wrong.

Why a Tagalog score is hard to trust

Detectors don't read meaning. They learn statistical patterns from training text. Turnitin says its classifiers are trained to spot differences in word probability between human and AI writing, and GPTZero says most of its training data is English. A model that learned what "human" looks like mostly from English has little basis for judging Tagalog word choices.

English detectors already misjudge people writing in their second language. In a Stanford study published in Patterns, seven GPT detectors flagged more than half of 91 human-written TOEFL essays as AI-generated. The average false-positive rate was 61.22%, and 97.8% of the essays were flagged by at least one detector (Liang et al., 2023). If detectors stumble on fluent non-native English, a language most of them weren't built for deserves more caution, not less. Our guide to why ESL writers get falsely flagged covers that mechanism.

What Taglish does to language identification

Before a detector scores anything, something has to decide which language the text is in. Copyleaks' API does this automatically: if no language is specified, "our system will automatically detect the language of the content." Turnitin doesn't document how it decides, but its English report requires that the file "must be written in English."

Taglish breaks the assumption behind that step. Researchers who built TweetTaglish, a dataset of more than 20,000 tweets for studying Tagalog-English code-switching, note that "traditional NLP techniques tend to perform poorly on mixed-language data." They add that word-level language identification is harder than document-level, and that most studies of document-level classification "focus on monolingual classification or do not consider the possibility of code-switching within the document." Even their own dictionary-based tagger mislabeled Spanish cognates such as para as Tagalog (Herrera et al., LREC 2022).

No vendor says what happens to a Taglish essay, so here is the plain inference. A mostly-English Taglish text may be routed to an English model, which then scores Tagalog stretches it was never trained on. A mostly-Tagalog one may be treated as unsupported and skipped. Either way, a Taglish score is untested territory. Formal coursework is usually written in one language anyway, so the practical fix is to write each assignment in the language it asks for.

What we measured on Tagalog, and what it doesn't prove

On 26 September 2026 we ran two small checks on our own detector, the one built into SupWriter. We did not use Turnitin or GPTZero for these checks.

Method. We used 4 distinct AI-written Tagalog drafts, scored before and after a rewrite. The first draft went through each rewrite mode twice, so each mode has 5 runs on 4 drafts. Separately, we scored 9 human-written Tagalog passages taken from Wikipedia revisions older than 2020, well before ChatGPT. Each passage was 80 to 200 words, in encyclopedic style.

AI-written drafts, before rewriting. Three of the four drafts scored 100% AI. The fourth scored only 35.9%, although it too was AI-written.

Human-written passages. The nine scores were 0, 7.1, 0, 0, 2.1, 0, 0, 100 and 23.3% AI. Seven of nine scored 7.1% or lower. One scored 23.3%, and one, the passage about Cebu, scored 100%: a clear false positive on text a person wrote years before ChatGPT existed.

After rewriting (AI score before → after, per run):

Runs 1 and 2 are the same draft. Put through the same mode twice, it came out at 100% one time and 0% the other. In each mode, 4 of the 5 runs landed at or under 11%, and aggressive mode did no better than default.

What this shows. On this small sample, our detector separated Tagalog AI drafts from human passages reasonably well: most AI drafts scored 100%, and most human passages scored under 8%. It still scored one AI draft at only 35.9%, flagged one human article as fully AI, and the same draft, rewritten twice in the same mode, ended at opposite scores.

Limitations. That's 13 source texts, one detector and one day. The human passages share one register (encyclopedia prose), and we didn't design a Taglish test. None of this says anything about how Turnitin or GPTZero would score the same text. So treat any single Tagalog score, ours included, as a rough signal, never as proof either way. Our testing methodology page explains what we have and haven't measured.

The rules that matter more: CHED CMO No. 21, s. 2026

For students at Philippine universities, the binding rules come from CHED and your school, not from a detector. CHED Memorandum Order No. 21, series of 2026, sets a national framework for AI in higher education. It was adopted under Commission en Banc Resolution No. 503-2026, dated 21 July 2026, and applies to all public and private higher education institutions under CHED (CMO No. 21, s. 2026, PDF). CHED announced it publicly on 22 September 2026 (Philippine News Agency).

Four provisions matter most if you write in Tagalog or Filipino:

• Disclosure is required, but your school designs it. Section 9 says HEIs define the disclosure rules and processes, CHED will provide model templates, and schools may set responses for non-disclosure "consistent with their academic policies and due process." Section 12 requires disclosure of AI use in theses and dissertations too.
• Detection can't decide a case alone. Section 10(2) encourages authentic, process-based, oral, reflective and performance-based assessment, and says: "AI detection tools, where used, shall not be the sole basis for determining academic misconduct."
• Detection isn't mandatory. Section 19 lists "mandatory AI detection for every submission" among obligations CHED does not require.
• Your syllabus sets the rule. The order says syllabi should state when AI use is appropriate, when it is restricted and how it must be disclosed. The PNA report quotes it: "No AI system should autonomously determine academic outcomes without meaningful human review."

Schools get a default 12-month transition period, which they can extend to 18 to 24 months, so the details will vary by campus for a while.

What to do: Filipino students and freelancers

If you're a student:

1. Read each syllabus for its AI rule. Under the CMO, that's where the line is drawn, and it can differ between courses.
2. Don't assume a Filipino-language paper goes unchecked. De La Salle University's library, for example, publishes a Turnitin guide that covers AnimoSpace assignments and includes a page on accessing AI writing detection (DLSU Libraries). A Tagalog paper won't get an AI score, but it still goes through the school's Turnitin workflow, and your instructor still reads it.
3. Keep your process. Save your notes, outlines and drafts, and keep your document's version history. With a detector that can't read your language, that record is your strongest evidence.
4. If you draft in Tagalog and submit in English, the English is what gets scored. English detectors can misflag non-native writers, so keep your Tagalog draft as a record. And note that the CMO's own definition of "AI-assisted work" includes translation, so AI-translated text may be something you have to disclose.
5. Don't take a free detector's number too seriously. A Tagalog score from a tool that doesn't list Tagalog isn't validated. Ours flagged a pre-2020 Wikipedia article at 100%.
6. If you're accused, point to Section 10(2) of the CMO, bring your drafts and version history, and ask what else supports the finding. Our guide to proving your writing is your own walks through the evidence.

If you're a freelancer:

• Ask which detector a client uses and what score they act on. If the deliverable is in Tagalog, show them the vendor's own language list.
• Offer process evidence instead of scores. Shared drafts and version history settle authorship questions better than a number no vendor has validated for Tagalog.
• Follow your contract on AI. If AI helped with a localization or draft, say so if the contract requires it.

If you rewrite Tagalog drafts for flow, see our Tagalog rewriting page; the Philippines guide compares tools in pesos. No rewrite tool can promise how any detector will score Tagalog, because the detectors themselves don't claim to support it.

FAQ

Is there an AI detector that officially supports Tagalog? Not among the major ones, as of September 2026. Turnitin, Originality.ai, GPTZero, Copyleaks and Pangram all publish language lists that leave out Tagalog and Filipino. QuillBot and ZeroGPT claim broad multilingual support but don't name Tagalog.

Can Turnitin detect AI in a Filipino or Tagalog essay? No. Turnitin says a paper in an unsupported language isn't processed for AI writing: the indicator shows an empty or error state and no AI report is generated. Turnitin treats its Similarity Report as a separate, independent check.

Does CHED require Philippine universities to use AI detectors? No. Section 19 of CMO No. 21, s. 2026 says CHED does not require mandatory AI detection for every submission. Section 10(2) says detection tools, where used, can't be the sole basis for a misconduct finding.

Does Taglish confuse AI detectors? No vendor documents how it handles code-switched text. Research on Tagalog-English code-switching finds that mixed-language text is harder for standard language-identification tools, so a Taglish score comes from a setup nobody has validated.

Can an AI detector flag my own Tagalog writing as AI? Yes. In our check of nine human-written Tagalog passages from pre-2020 Wikipedia, our detector scored seven at 7.1% or lower but flagged one at 100%.

---

Check a draft with SupWriter's free AI detector before you submit: 300 words free, no credit card required. For Tagalog, treat the score as a rough pre-check, not proof.