What is Grok? Definition, history and how it works

How does Grok work?

Grok works as a large language model that answers questions with live data from X and the open web. It is a large language model (LLM) trained by xAI, wrapped in a chat interface available on grok.com, in mobile apps and inside X. When a question needs fresh information, Grok runs its own searches and reads posts, pages and media before it answers.

xAI’s July 2025 Grok 4 announcement describes three core capabilities:

• Native tool use: the model was trained with reinforcement learning to call a code interpreter and a web browser on its own.
• Real-time search: it queries both the web and X and can look at images and video it finds there.
• Long context: the API version handles 256,000 tokens of text and images in one request.

The name comes from the verb “grok”, coined by Robert A. Heinlein in his 1961 novel Stranger in a Strange Land, meaning to understand something deeply.

History of Grok: key dates

Grok evolved from a single chatbot model in 2023 to a family of models released every few months. The table lists the milestones.

In the March 2025 deal Musk valued xAI at 80 billion USD. Since the SpaceX transaction, xAI operates as a wholly owned SpaceX subsidiary. Grok 4 was trained on Colossus, a cluster of 200,000 GPUs.

Grok users and pricing

Grok has around 117 million monthly users and a small paying base. Pulse 2.0’s reading of the SpaceX filing gives about 1.9 million SuperGrok subscribers and roughly 365 million USD of 2025 revenue from Grok and X subscriptions combined. About 21% of X’s roughly 550 million users use Grok features. Access is sold through X Premium+, whose price rose from 22 to 40 USD a month after Grok 3, and through standalone SuperGrok tiers, including SuperGrok Heavy for Grok 4 Heavy.

Grok in marketing, SEO and AI search

Grok matters to marketers because it answers questions from what people are posting right now, while most assistants lean on indexed web pages. For brands, this has three practical consequences.

Social signals become AI answers

Posts and replies on X (Twitter) feed Grok’s view of a brand. A product that is discussed negatively on X this week can be described negatively by Grok tomorrow, so social listening and reputation work now affect AI visibility.

Generative Engine Optimization

Grok also reads the open web, so the same rules apply as for ChatGPT or Perplexity: clear definitions, named entities and citable facts. This is the core of Generative Engine Optimization, which I describe on the AI SEO services page.

Brand safety

Grok has a record of controversial output. In July 2025 it praised Adolf Hitler in replies on X, and the same month Poland’s digital affairs minister Krzysztof Gawkowski announced a report to the European Commission after vulgar Grok posts about Prime Minister Donald Tusk. Brands that run ads or automated replies on X should monitor what Grok says about them.

Grok vs ChatGPT and Gemini

Compared with Gemini, Grok has a much smaller distribution footprint, but it owns a unique data stream: public conversation on X.