Mostly no. According to the vendor's documentation, updated on 9 November 2026, Grok's text watermark is off by default and team admins turn it on per model. This site does not read the watermark. It checks how a text is written.
What the vendor says
- Watermarking applies to text from the Responses and Chat Completions APIs, for each model a team has switched on: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and the grok-4.20 reasoning, non-reasoning and multi-agent models.
- Team admins set it in the Console under "Model output watermarking": on for all models, off for all models, or custom per model. Changes apply within five minutes.
- Some models can be locked on for a team. Turning a model off requires confirming that an exception under applicable law applies.
- The signal is invisible. It is applied while the text is generated by slightly shaping token choice, with no added characters. The vendor names the method SynthID Text.
- An API response carries the header x-watermark: true when that request's output was watermarked. Any other value, or no header, means it was not.
- grok-build-0.1 is not on the list. The page does not mention the consumer Grok app or X.
The default changed
Coverage in early November reported that the watermark would be on by default from 7 November. On 10 November, The New Stack reported that the default had been switched back to off. The vendor's own page states off and gives no date for the change. Treat any "watermarked by default" claim as out of date, and re-check the vendor's page.
Who can verify it
The vendor says detection is statistical: it scores a text and reports watermarked, not watermarked, or uncertain, and longer text gives a stronger signal. Access to detection is granted by application through a request form. No public detector exists, and the page does not say who has been approved or on what criteria.
What this means in practice
Most Grok text has no watermark to find: teams that left it off, text from the app, and text from other models all carry none. A missing mark proves nothing about who wrote a text. The only check open to everyone is a style-based one.
Checking a text without the key
GrokDetect reports an AI verdict, the passages that carry the signal, and how closely the writing matches Grok compared with ChatGPT, Claude and other families. On the facts page, 100% of fully AI-written xAI texts were detected and 0.5% of human-written texts were wrongly flagged. Style similarity is a signal, not proof of authorship.
For PDF and Word files, Fileverity's detector takes those as well. Version details are on the Grok versions page.
Sources: docs.x.ai, "Watermarking" (advanced API usage), updated 9 November 2026; The New Stack, 10 November 2026.