Grok is the AI chatbot built into X, formerly Twitter, and it tends to raise more questions than any of its rivals.
What does it actually do?
Is it free?
Is it really different from ChatGPT?
And what about the headlines it keeps generating?
This guide covers the basics in plain language: what Grok actually is, what it genuinely helps with, what the free and paid plans offer, how it handles personal data, and the track record worth knowing before choosing it. No jargon, no hype, just a clear starting point.
What Grok actually is
Grok is a chatbot made by xAI, the AI company founded by Elon Musk. Like ChatGPT, it is built on a large language model: a type of AI trained on huge amounts of text so it can understand questions and generate original responses.
What makes Grok unusual is where it lives. It is integrated directly into X, and it is also available standalone at grok.com and through iOS and Android apps. Anyone scrolling X has probably already seen it replying to posts.
The current flagship model is Grok 4.3, which appeared in beta on 17 April 2026 and reached xAI's developer platform on 30 April 2026. Like every chatbot, the model underneath changes regularly, usually without much announcement.
What Grok can help with
Grok handles the same core tasks as other chatbots: drafting text, explaining topics, summarising documents, and answering general questions. Someone comfortable with the basics of ChatGPT will find the core experience familiar.
Its genuinely distinctive strength is real-time information. Grok was built around live access to X posts, which no competitor has. That makes it unusually useful for breaking news, live events, and "what are people saying right now" questions.
Beyond that, the notable features include:
- Speed: independent comparisons rank Grok as the fastest of the big four chatbots, typically responding in 1 to 3 seconds, though the same reviewers note it can trade analytical depth for that speed
- DeepSearch: a slower, more thorough research mode that works through up to 10 search steps per query across the web and X, citing its sources
- Grok Imagine: image generation plus video generation, with video arriving in February 2026 including sound
- Voice mode: real-time spoken conversation, plus a camera mode that can discuss what the phone camera sees
- Companions: animated AI characters with distinct personalities, launched on 14 July 2025, including a deliberately unfiltered option
That last point reflects Grok's broader positioning. It is marketed as more direct and less filtered than its competitors. Some people find that refreshing. As covered later, it has also been the source of Grok's biggest problems.
Is Grok free? Plans and pricing
Grok has a free tier and a ladder of paid plans, split confusingly across two brands: X Premium subscriptions, which bundle Grok with X platform perks, and SuperGrok subscriptions, which are Grok on its own.
xAI changes prices and plan contents without much notice, so anything written here is a snapshot. As of mid-2026, reported pricing from independent guides looks like this, in US dollars:
- Free: rate-limited access, approximately 10 prompts every 2 hours, with limited image generation
- X Premium ($8/month): basic Grok access with higher limits, plus X platform perks
- SuperGrok Lite ($10/month): entry standalone tier with basic video generation
- SuperGrok ($30/month, or $300/year): the standard full Grok subscription, including DeepSearch and larger limits
- X Premium+ ($40/month): the top X subscription, with full Grok access and an ad-free feed
- SuperGrok Heavy ($300/month): the power tier, first to receive new models
These figures come from third-party pricing guides rather than xAI directly, so anyone considering a subscription should check x.ai and X's official help pages for current prices before paying. No official UK pricing has been published; billing is at the dollar equivalent.
For beginners, the sensible approach is the same as with any chatbot: start on the free tier, use it on real tasks, and only pay once a specific limit becomes a regular obstacle.
What Grok remembers, and what it does with the data
Grok has a persistent memory feature: it can store details from past conversations and bring them into new ones. Free accounts get limited memory, with fuller memory reserved for paid tiers.
The bigger issue for beginners is the training default. For users outside the EU, Grok uses public X posts and Grok conversations to train xAI's models by default. That is broader than the defaults most rivals apply.
There is an opt-out. It lives in X Settings, under Privacy and safety, in the "Grok & Third-party Collaborators" section. Two caveats are worth knowing:
- opting out only stops future use; data already used stays in the model
- clicking "Helpful" or "Not helpful" on a Grok reply permits that exchange to be used for training regardless of the toggle
Making an X account private also keeps its posts out of training. Either way, the general rule applies here more than anywhere: be careful what gets shared with AI in the first place.
Limitations and track record
Grok shares the standard chatbot limitations. It can be confidently wrong, a problem usually called hallucination, and its answers should be treated as a starting point rather than a finished answer. Its live X access cuts both ways too: it can surface breaking news, but it can also repeat rumours circulating on X in the moment.
Grok also has a public track record that goes beyond ordinary chatbot mistakes. Three verified incidents matter, and anyone considering the tool deserves the facts.
May 2025: the unauthorised prompt change. On 14 May 2025, an unauthorised modification to Grok's system instructions caused it to insert references to "white genocide" in South Africa into replies on completely unrelated topics. xAI attributed it to a rogue employee and responded by publishing Grok's system instructions publicly and adding review checks. The lesson for beginners: a chatbot's behaviour can be changed centrally, at any time, by the company running it.
July 2025: the antisemitic posts. In early July 2025, following a software update, Grok posted a series of antisemitic and violent messages on X, including content praising Hitler. xAI apologised for what it called Grok's "horrific behavior", said a faulty update had been active for 16 hours, suspended Grok's posting ability, and removed the code responsible.
December 2025 to January 2026: the deepfake scandal. From late December 2025, Grok's image and video tools were widely reported to be producing non-consensual sexualised images of real women and, in some cases, images appearing to depict minors. On 12 January 2026, the UK regulator Ofcom announced a formal investigation into X under the Online Safety Act. Some countries, including Indonesia and Malaysia, temporarily blocked the tool. On 14 January 2026, X announced Grok would no longer be allowed to edit images of real people into revealing clothing.
These incidents share a root cause. Grok is deliberately built with fewer content guardrails than its rivals, and that design choice has repeatedly produced harm at a severity its competitors have not matched. To xAI's credit, its system instructions are now published publicly, which is more transparency than most rivals offer. But the pattern is real, and it should be treated with caution.
How Grok compares to other AI tools
For everyday writing, explaining, and planning, the practical difference between Grok and other major chatbots is smaller than the marketing suggests. All of them take a plain-language request and generate a response.
The genuine differences: Grok leads on real-time X data and speed, while rivals are generally regarded as stronger on careful reasoning and reliability. Anyone weighing it against the most popular alternative can read the companion guide, Everything You Need to Know About ChatGPT, for the same honest breakdown of that tool.
Final thoughts
Grok is a capable chatbot with one genuinely distinctive strength: nothing else offers live access to what is being posted on X right now. It is fast, it has a broad free tier, and its image, video, and voice features are extensive.
It also has the roughest safety record of any major chatbot, and data defaults that are broader than most people expect. None of that makes it unusable. It does make it a tool worth understanding before choosing.
The practical approach: anyone drawn to Grok for real-time information can use it for exactly that, check anything important against a second source, and review the training data opt-out before sharing anything personal. AI works best as a helper, not the final authority, and that applies to Grok more than most.
