Text Watermark Detector
Vendor-neutral provenance verification

One API. Verify all Providers.

AI Text
Watermark Checker

Verify text watermarks and provenance signals from official providers without guessing from writing style.

View provider status
No style-based guessing Text is not retained
Verification console

Text sample

Check for a watermark

OGAM

Tracking the official evidence from

OpenAIGoogleAnthropicMeta

Provider support

Evidence changes. Our answer changes with it.

Mechanism availability and our integration status are tracked separately, with every claim linked to official material.

01

OpenAI

announced/experimental

OpenAI developed a text watermarking method but is still evaluating it alongside metadata and other provenance options.

Research text watermarking method

02

Google

available

Google applies SynthID Text to Gemini text and offers an early-access detector. Its open-source tools are for teams watermarking their own models.

SynthID Text

03

Anthropic

unavailable

Anthropic says it continues to explore watermarking, but it does not document a deployed text watermark for Claude.

04

Meta

announced/experimental

Meta's open-source TextSeal toolkit can add and detect text watermarks. Meta does not say that Meta AI or released Llama services use it.

TextSeal research toolkit

How it works

Signals, not vibes.

Writing can sound machine-made and still be human. Provenance is about evidence.

01

A provider issues a signal

The model embeds or signs a verifiable signal when it generates the text.

02

We check the actual evidence

Each request is routed to supported verification methods—not a style classifier.

03

You get a sourced answer

The result says what was verified, by whom, and where the method has limits.

Developer API

One request. Every supported verification method.

Add provenance checking to moderation, publishing, education, or trust workflows without maintaining provider-specific integrations.

POST /v1/check
{
  "text": "Content to verify…"
}

→ {
  "status": "no_signal",
  "providers_checked": 4,
  "text_retained": false
}

Straight answers

No certainty theater.

Provider researchWatermarking articles
Can this detect any AI-written text?
No. It checks supported provider signals. Without a verifiable signal, we cannot prove who wrote the text.
Why is the checker disabled?
No provider method is ready for us to use in production. We do not accept text until a real check can run.
Will submitted text be stored?
No. Once checking is available, text will be used for the request and then discarded.