If your business uses AI to write anything your customers read, that text can now be identified as machine made. From 7 November 2026, the text Grok writes through its developer API carries an invisible watermark by default. Here is what changed, what it means for a small business, and what to check.
What happened
xAI, the company behind Grok, switched on invisible watermarking for text produced through the Grok API. The change covers seven models: grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three versions of grok-4.20. It applies on every Grok API endpoint, and it needs no change to the code that calls the API.
The watermark is invisible. It adds no characters, symbols or hidden text to the reply, and it does not change what the model writes. It works by shaping which words the model picks as it generates text, so the result can later be recognised as machine generated. Developers can check whether a reply was marked by reading the x-watermark header on the API response.
The company states the reason plainly in its developer documentation: the watermark is there to comply with the EU AI Act. Team administrators can switch it off model by model in the console, but doing so asks them to confirm that a legal exception covers their use.
This is the first time a major AI provider has made text watermarking the default for its API. Anthropic has announced a second round of text watermarking starting 16 November, so Grok is unlikely to stay alone for long.
What it means for a small business
Most small businesses never call an AI API directly. But many of the tools a small business uses every day do: the chatbot that answers customers on WhatsApp, the app that drafts product descriptions, the tool that writes social captions. If any of those tools run on Grok, the text they produce for your business now carries a mark that identifies it as AI written.
That matters in three ways.
First, honesty is becoming checkable. Until now, a business could publish AI written text and nobody could prove it was machine made. A watermark changes that. Customers, platforms and regulators can increasingly tell the difference between text a person wrote and text a model wrote.
Second, the rules behind this are already in force. The EU AI Act transparency rules took effect on 2 August 2026, and providers of AI systems already on the market have until 2 December 2026 to comply. If you sell to customers in Europe, the marking of AI content is not a future problem. It is a current one.
Third, your chatbot is part of this. A WhatsApp chatbot that answers customers with AI generated replies is producing exactly the kind of text these rules cover. Whoever built your chatbot should be able to tell you which model it uses and whether its replies are marked.
What to do about it
Run these checks this week. None of them costs anything.
- Find out which AI tools your business uses. List every tool that writes text for you: your chatbot, your product description writer, your caption tool, your email assistant. Ask each provider which model it runs on, or check its settings page. If any of them use Grok through the API, its output is now watermarked by default.
- Keep a human in the loop. Read AI written text before it reaches a customer. Fix errors, check prices and details, and make sure the tone sounds like your business. A watermark does not make wrong text right.
- Label AI assisted content honestly. If a reply, a listing or a post was written with AI help, say so where your customers can see it. Customers forgive AI help. They do not forgive being misled.
- Ask your chatbot builder one question. If Digilistan or anyone else built a chatbot for you, ask whether its replies are watermarked and what that means for your customers. A builder who cannot answer that question is a builder to worry about.
- If you sell to Europe, mark the date. The 2 December 2026 compliance date for systems already on the market is less than a month away. Make sure the tools you use are keeping up, and keep records of what you checked.
What this does not change
The watermark applies to text made through the Grok API. xAI has said nothing about watermarking answers in the Grok app or on X, so do not assume those are marked too. The watermark also does not make AI text better or worse. xAI says it changes neither quality nor speed, and that claim has not been independently tested. And a watermark is a statistical signal, not a stamp on the page. It can be detected, but only with the right tools.
The direction is clear, though. AI written text is becoming identifiable, and the businesses that stay ahead of that change are the ones that treat AI as a tool they are open about, not a shortcut they hide.
Sources
- SpaceXSI developer docs, Output Watermarking, updated 31 October 2026: https://docs.x.ai/developers/advanced-api-usage/watermarking
- Maccus, Grok will watermark API text from November 7: https://maccus.fr/en/grok-text-watermark/
- TeslaNorth, Grok’s API Answers Will Carry a Hidden Watermark Starting in November: https://teslanorth.com/2026/10/31/grok-api-watermark-november/
