How to use Grok AI?
Go to grok.com or install the Grok app on iOS or Android and start typing — basic chat is free, no payment method needed. Grok also lives inside X, in Tesla vehicles, and behind an API at $2 per million input tokens. Its real edge is live X search; use it for what's happening now, and verify anything that matters.
Why — the first-principles explanation
Most "how to use Grok" pages describe buttons. The buttons are obvious. What isn't obvious is what this particular tool is unusually good and unusually bad at — and that comes straight from how it was built.
Grok's structural advantage is proximity to X. Its parent company owns the social network, so it can search live posts in real time rather than relying on a bolted-on web search. That makes it the best mainstream chatbot for "what is happening right now" — a breaking story, the reaction to an announcement, whether an outage is real. It also makes it the worst place to accept an answer uncritically, because live social media is news, jokes, and lies in one stream, and a model repeating it can't reliably tell them apart. So the operating rule writes itself: use Grok to find out what people are saying, then verify before treating any of it as fact.
The second thing to internalize is that prompting is specification, not conversation. The model can't read your mind about format, length, audience, or constraints — and everything you leave unsaid, it guesses. "Write about solar" gets you a guess. "Write 200 words for a homeowner deciding on rooftop solar, plain language, lead with payback period" gets you a result. This isn't Grok-specific; it's just where beginners lose most of their value.
Third, pick the right door. There are five, and they're not interchangeable. grok.com is the full standalone experience. The iOS/Android apps add voice. Inside X, you can reply to a post and ask Grok about it in context — that's the killer feature nobody uses enough. Tesla vehicles got Grok in July 2025. And the API is for building software, priced per token rather than per seat: Grok 4.5 runs $2.00 per million input tokens and $6.00 per million output tokens under 200k context, as of 2026-07.
Finally, know the guardrails are looser here by design, and that has a paper trail — the July 2025 "MechaHitler" episode and the December 2025 deepfake scandal both came from that dial. Anything you type is stored data; xAI's Data Controls let you opt out of model training, and private chat exists for a reason. In August 2025, shared Grok conversations were accidentally indexed by Google. Don't paste secrets into it.
An example that makes it click
Grok is the friend who's always online and always reading the group chat. Ask him what everyone's saying about the thing that just happened, and he's the fastest source you have — genuinely useful, genuinely first.
Ask him for your tax numbers and you should check every one, because he'll answer with the same confidence either way, and half of what he's repeating is somebody's joke. Use him for the pulse. Don't use him as the record.
How to do it
- Pick your door: grok.com in a browser, the Grok app on iOS or Android for voice, or the Grok button inside X. Sign in — free chat needs no payment method.
- Ask one real question to calibrate. Free users get a lighter model with a cap that resets every couple of hours; you'll learn your ceiling in a day of normal use.
- Specify instead of chatting. State the audience, the format, the length, and the constraint in one prompt — 'a 150-word email to a client explaining the delay, apologetic but not groveling' beats 'write an email' every time.
- Use its actual edge: for anything live — breaking news, reaction to an announcement, is-this-service-down — ask Grok to search X. This is where it beats other chatbots, not on general knowledge.
- Use it inside X on a specific post: reply and ask Grok to explain, fact-check, or summarize the thread. It has the context right there.
- Verify anything that matters. Grok drinks from live social media, which is a mix of reporting, jokes, and lies. Ask for sources, then open them.
- For images or video, open Grok Imagine — but note it's paid: xAI removed the last free image access on March 19, 2026.
- Protect yourself: don't paste secrets, use xAI's Data Controls to opt out of model training, and use private chat for anything sensitive.
Key facts
- Grok is accessible at grok.com, in iOS (2024) and Android (2025) apps, integrated in X, in Tesla vehicles (since July 2025), and via API — also distributed through Microsoft Azure and GitHub Copilot.
- Free chat is available with usage caps on a lighter model; the newest model, Grok 4.5, shipped July 8, 2026 and rolls out to paid tiers first.
- xAI API pricing as of 2026-07: grok-4.5 at $2.00/M input and $6.00/M output tokens under 200k context, rising to $4.00/$12.00 at or above 200k tokens.
- Grok Imagine generates images ($0.02–$0.05 each via API) and video ($0.080 per second on v1.5, 1–15 seconds, 480p/720p/1080p); free image access was removed March 19, 2026.
- SuperGrok is $30/month; Grok launched November 4, 2023 as a Premium+ only feature.
- In August 2025, private Grok conversations were exposed to Google indexing via a misconfigured session-sharing feature — anything typed into a chat is stored data.
Open the official product and verify the active account, workspace and data settings before you begin.
▶ The 60-second explainer (script)
How do you use Grok? The buttons are easy — go to grok dot com, or install the app on iOS or Android, sign in, start typing. Basic chat is free, no card needed. It's also inside X, in Teslas since July 2025, and behind a developer API at two dollars per million input tokens. That's the boring part. Here's what actually matters. Grok's parent company owns X, so it reads live posts in real time. That makes it the best mainstream chatbot for what is happening right now — a breaking story, reaction to an announcement, whether a service is actually down. Other chatbots learn the world in one giant batch and then go stale. But that same firehose is why you verify. Live social media is news, jokes, and lies in one stream, and the model can't reliably tell them apart. So: use Grok to find the pulse, not to settle the record. Second thing — prompting is specification, not conversation. Everything you don't say, it guesses. Write about solar gets you a guess. Two hundred words for a homeowner deciding on rooftop solar, plain language, lead with payback period — that gets you a result. Third: use it inside X. Reply to any post and ask Grok about it. It has the context sitting right there, and almost nobody does this. And one warning: anything you type is stored. In August 2025 shared Grok chats got indexed by Google. Use Data Controls, use private chat, don't paste secrets.
What authoritative sources say
People also ask
Do I need an X account to use Grok?
No. grok.com and the mobile apps work standalone with their own sign-in. The X integration is one door among several.
What is Grok actually best at?
Live information. It searches X posts in real time, so it's strongest on breaking events and public reaction. On general reasoning it trades the lead with ChatGPT and Claude every few months.
How do I get better answers?
State the audience, format, length, and constraints in the prompt itself. Most bad output comes from a vague request, not a weak model — everything you leave unsaid, it guesses.
Can I use Grok for images and video?
Yes, through Grok Imagine — but it's paid. xAI removed the last free image access on March 19, 2026. Video runs 1 to 15 seconds with sound generated in the same pass.
Is it safe to paste work documents into Grok?
Treat anything you type as stored. Use Data Controls to opt out of model training and use private chat for sensitive material. In August 2025 shared conversations were accidentally exposed to Google's index.
The same question, asked other ways
- How do you use Grok?