AI watermark detectors: what you can check in 2026, and what you can't

The best AI watermark detector depends on what you are checking. For images, video and audio, Google's SynthID Detector is the broadest public check: since 7 October 2026 anyone can use it, and it recognises SynthID watermarks from Google, OpenAI, NVIDIA, Kakao and Apple. For text there is no public watermark checker yet: Anthropic, OpenAI and Google keep text watermark detection for vetted organisations, so for a pasted essay or article the practical check is an AI detector that reads the writing itself.

AI watermark detectors compared

The public checkers as of 1 November 2026, from each tool's own page.

Two kinds of signal are being read here. A watermark such as SynthID is woven into the pixels, sound or words themselves. Content Credentials, built on the C2PA standard, are a signed record attached to a file as metadata: detailed when present, but easy to lose, because re-saving or uploading a file can strip it.

A free first check for text: IsThisAiWritten

IsThisAiWritten is our own tool, and it is listed here without a number because it does a different job. It is a classifier, not a watermark reader: it reads the writing itself and shows how AI-like it is, so it works on any text you paste, from any model, watermarked or not, with no key needed.

It is free, with no sign-up, no email and no daily cap, and there is no word limit, so a whole thesis or manuscript goes in at once. The check runs in your browser, so the text is never uploaded, stored or used for training. The report gives an AI-likeness score, sentence-by-sentence highlights and style notes with plainer wording, and a cleaner for hidden characters and look-alike letters. Paste a text into the free AI detector to try it.

1. SynthID Detector, the SynthID checker

Best for: images, video and audio that may come from Google's AI or its partners.

Google launched the SynthID Detector in 2025 for journalists and media professionals and opened it to everyone on 7 October 2026, worldwide in English. Upload a file and it tells you whether it carries a SynthID watermark from Google or one of its partners: OpenAI, NVIDIA, Kakao and Apple. Google says it has watermarked over 180 billion images and videos and 240,000 years of audio. The same checks are built into Search, Chrome and the Gemini app, where you can upload an image and ask whether it was made with Google AI.

It only recognises SynthID, so a file from a generator that doesn't use it shows nothing either way.

Is there a SynthID checker for text? Not a public one. Google has marked text in the Gemini app with SynthID Text since May 2024, but the public detector covers images, video and audio only. Google has released SynthID Text as open source, which lets developers watermark and check text from their own models with their own keys.

2. OpenAI Verify

Best for: images and audio that may come from ChatGPT, Codex or the OpenAI API.

OpenAI's verify page checks an uploaded image or audio file for the two signals OpenAI adds: Content Credentials and an invisible SynthID watermark. It accepts PNG, JPG, WEBP and GIF images and common audio formats, and the same check is available through an API. OpenAI says detected signals are reliable and false positives rare. If nothing is found, the file may still be OpenAI's: metadata can be removed, a watermark can degrade, and older output predates the signals.

OpenAI's text watermark detector is separate, and access is limited to approved researchers and expert organisations.

3. Claude Content Checker

Best for: files you think were made or edited with Claude.

Anthropic's free checker looks for a Content Credential issued by Claude in images, video and audio files up to 100 MB. The file never leaves your device, and the tool reads only the attached credential, not the content. Anthropic is clear that "the tool does not check text": detection for Claude's text watermark runs through an API in private preview for eligible organisations.

4. Content Credentials Verify

Best for: reading the full Content Credentials record of a file, from any issuer.

Run by the Content Authenticity Initiative, Verify opens the C2PA record in a file and lays out its details, including the changes made to the file over time. It takes common image, video and audio formats and PDF. It reads metadata only, and the site itself warns that Content Credentials are still rolling out, so many files have nothing to show.

Who can check text for a watermark?

Only vetted organisations, for now. Anthropic's detection API is open to regulators, law enforcement, media, fact-checkers, independent researchers, educational organisations, EU civil society groups and businesses with their own duty to verify watermarks. OpenAI grants access to its detector case by case to approved researchers and expert organisations, and says it won't make the tool public at launch because of the risk of missed watermarks and false positives. The EU's Code of Practice on AI-generated content allows this: since detection for text watermarks is less reliable, signatories may restrict it, for a limited time, to verified expert users with a legitimate need.

Neither company offers public access, so be wary of any website that claims to read ChatGPT or Claude watermarks in pasted text. How the watermarks themselves work, and where they apply, is covered in does ChatGPT watermark text?

What a watermark check can and can't tell you

A positive result is strong evidence: it means a specific company's system made or processed the file. A negative result means much less. Metadata gets stripped, watermarks weaken with heavy editing, older output was never marked, and most generators outside these partnerships don't use SynthID or Content Credentials at all. Both Anthropic and OpenAI stress that a missing mark doesn't show a person made something. And no check, positive or negative, tells you whether the content is true.

How we chose

We included every watermark or Content Credentials checker that a member of the public can use today, and ordered them by how much content they cover: SynthID Detector reads several companies' watermarks, OpenAI Verify and the Claude Content Checker read one company's signals each, and Content Credentials Verify reads metadata only. IsThisAiWritten comes first, unnumbered, because it is our own tool and the free first check for text; it is a classifier rather than a watermark detector, so it isn't ranked among them. We opened each tool's page on 1 November 2026. We did not test the checkers ourselves.

Sources

1. Google, Google expands SynthID Detector for AI content (7 October 2026), checked November 2026
2. Google DeepMind, SynthID, checked November 2026
3. Google, AI image verification in the Gemini app (20 November 2025), checked November 2026
4. Google DeepMind, Watermarking AI-generated text and video with SynthID (14 May 2024), checked November 2026
5. Google AI for Developers, SynthID Text, checked November 2026
6. OpenAI, Verify content generated by OpenAI, checked November 2026
7. OpenAI, Our approach to EU text provenance rules (5 October 2026), checked November 2026
8. Anthropic, Check if a file was made with Claude, checked November 2026
9. Claude Help Center, How Claude marks AI-generated content, checked November 2026
10. Anthropic, Claude text watermarking (14 August 2026, updated 1 September 2026), checked November 2026
11. Content Authenticity Initiative, Content Credentials Verify, checked November 2026
12. European Commission, Code of Practice on Transparency of AI-Generated Content (10 June 2026), checked November 2026