Grok (xAI)

User Guide -- Tips, How-Tos & Getting Started

Zoom

What this guide covers

Grok is xAI's AI assistant with unique real-time access to X (Twitter) data and the live web. Its flagship Grok 4.3 model offers a 1 million token context window. Grok is strongest for tasks involving current events, social media trends, and real-time information -- no other major AI assistant has native access to the full X firehose.

This guide covers Grok's core features: DeepSearch for cited multi-source research, Aurora image and video generation, voice mode, plans and limits, and best practices for getting reliable results.

Real-time X data DeepSearch Think mode Aurora image gen Grok Voice 1M token context Plans & limits Prompting tips
⚠ Reminder: Grok is a productivity assistant, not a decision-maker. Always verify output for accuracy before sharing or acting on it. Grok can and does make factual errors, especially on recent events and political topics. Do not paste sensitive or confidential data into any AI tool.
Real-time data: Grok can search the live web and X posts as they are published -- but real-time results reflect what is currently being said on X, which can include misinformation. Treat real-time summaries as a starting point, not a final source.

What Grok Is

xAI's AI assistant with a defining advantage: native, real-time access to the full X (Twitter) data firehose. No other consumer AI can do this.

✨ Key Strengths

🔹 Real-Time X + Web Data
Grok is the only major AI with native access to the full X firehose. Ask it about what is trending right now, what people are saying about a breaking story, or what a specific community is discussing -- and it queries live X posts to answer.
🔒 Grok 4.3 Flagship Model
The current flagship offers a 1 million token context window, native video input, knowledge cutoff of December 2025, and API pricing roughly 40% lower than its predecessor. Think mode adds visible chain-of-thought reasoning for hard problems.

🔹 Grok is embedded directly in the X platform (web, mobile, sidebar) -- making it the only AI assistant natively inside a social network with 600 million+ monthly active users.

How Grok Differs From Other AI Tools

Key differentiators at a glance.

FeatureGrokChatGPT / Claude / Gemini
Real-time X/Twitter data Yes -- native, full firehose No
Real-time web search Yes Yes (varies by plan)
Native image generation Yes (Aurora / Grok Imagine) Varies by tool
Native video generation Yes (paid tiers) Limited or no
Tone Direct, less hedged, witty Neutral or cautious
Context window (flagship) 1M tokens (Grok 4.3) 128K--200K typical
Social platform integration Native in X app and web No native social integration

Think Mode (Extended Reasoning)

Think mode shows Grok's chain-of-thought reasoning before it gives a final answer -- useful for hard problems, not for quick lookups.

Use Think mode forSkip Think mode for
Multi-step math or logic Simple factual questions
Complex coding tasks Short creative writing
Graduate-level science Casual conversation
Debugging ambiguous problems Speed-critical tasks
Legal or financial reasoning Tasks with no reasoning chain

Think mode is available on X Premium+ and above. Big Brain Mode (highest-effort reasoning) is SuperGrok-only. Both consume separate quotas.


How-To Cards: Getting Results with Grok

Practical starting prompts for common use cases.

01
Track breaking news on X in real time
Grok queries live X posts as they appear -- ideal for fast-moving stories where training data is already outdated.
"Summarize the most recent posts on X about [event]. What are people saying right now and what key facts are emerging?"
02
Analyze public sentiment on a topic
Ask Grok to synthesize what X users collectively think about a brand, policy, product, or news event.
"What is the general sentiment on X toward [topic or brand] over the past 48 hours? Summarize positive, negative, and neutral themes."
03
Identify what is trending in a community
Use Grok's X access to see what a specific niche, industry, or interest group is discussing today.
"What topics are trending on X among [specific community or industry] right now? List the top 5 with a one-sentence summary of each."
04
Use Think mode for a hard math problem
Enable the Think toggle before submitting. Grok will show its reasoning chain and self-correct before giving the final answer.
"[Toggle Think on] Solve this step by step and show your reasoning: [paste problem]. Flag any assumptions you make."
05
Research a topic with DeepSearch
Click the DeepSearch button to trigger multi-step web and X research that returns a cited, multi-source report.
"[DeepSearch] Research the current state of [topic]. What are the most credible recent sources saying? Cite your sources."
06
Monitor competitor activity on X
Use Grok's real-time X access to see what customers and users are saying about a competitor's product or announcement.
"What are X users saying about [competitor name or product] since [date]? Highlight the most common complaints and praise."
07
Draft content grounded in current context
Ask Grok to write something that references what is actually happening right now -- a capability general AI tools lack without live data.
"Using current events on X as context, draft a short [post/article/summary] about [topic]. Keep it factual and cite any specific claims."
08
Debug code with Think mode
Think mode helps Grok plan before writing code, catching structural errors before they appear in the output.
"[Toggle Think on] Debug this code and explain what is wrong: [paste code]. Walk through the logic before suggesting a fix."

Tips for Best Results

How to get more reliable and useful output from Grok.

Lead with an action verb
Start with "Analyze," "Summarize," "Draft," "Explain," or "Compare" -- this anchors Grok to a concrete output type and reduces vague responses.
Specify your audience and format
Tell Grok who it is writing for and what format you want: "write this for a non-technical audience as a bullet list under 200 words."
Use X search for live data tasks
For anything involving current events, trending topics, or real-time sentiment, ask Grok to search X. This is the feature that sets it apart from every other AI tool.
Reserve Think mode strategically
Think mode burns quota and adds latency. Use it for problems where the reasoning path actually matters -- hard math, debugging, or multi-step logic -- not for quick lookups.
Iterate in one conversation
Keep adding to the same conversation thread for richer, more context-aware output. Start a new chat only when the topic changes entirely.
Verify real-time claims
Grok's real-time X answers reflect what is being posted on X right now -- which can include rumors and misinformation. Cross-check important facts with primary sources before using them.

DeepSearch

Grok's multi-step research mode  ·  Available at grok.com and in the X app

What is DeepSearch?

  • DeepSearch is Grok's dedicated research mode -- not the default chat mode. You activate it explicitly by clicking the DeepSearch button.
  • When triggered, Grok performs multiple live web queries and X searches, reads and synthesizes the results in real time, resolves conflicting information, and returns a cited, multi-source report.
  • The search process is visible -- you can watch Grok's iterations unfold as it plans searches, evaluates what it finds, and refines its approach.
  • Sources are linked inline in the response, not collected at the bottom -- making it easy to verify any specific claim.
  • Introduced with Grok 3 in February 2025; available on paid tiers.

How to trigger DeepSearch

  1. Open Grok at grok.com or via the X sidebar or app
  2. Type your research question in the input box
  3. Click the DeepSearch button (appears below the text input) before submitting
  4. Submit -- Grok will show its live search iterations as it works
  5. Review the cited report; click any inline source link to verify a claim
  6. Follow up in the same conversation to refine or drill down on any finding

DeepSearch vs. Standard Mode

DimensionStandard ChatDeepSearch
Data source Training data + optional web search Live web + X firehose, always
Latency Seconds 15--60+ seconds
Output style Conversational Report with inline citations
Quota cost Standard message quota Separate DeepSearch quota
Conflict handling May pick one version silently Notes discrepancies explicitly

DeepSearch quota by tier

TierDaily DeepSearches
FreeVery limited or none
X PremiumLimited (~10--20/day)
X Premium+~50/day
SuperGrokHigh / functionally unlimited
SuperGrok HeavyUnlimited

When to Use DeepSearch

  • Researching current events, recent product launches, or breaking news
  • Verifying claims against multiple independent sources
  • Analyzing what people are saying about a topic on X right now
  • Building a research brief where cited sources matter
  • Competitive intelligence requiring up-to-date market context
  • Fact-checking circulating claims before sharing them
  • Any topic where training data is likely outdated

When NOT to use DeepSearch

  • Coding help, debugging, or code generation
  • Creative writing, brainstorming, or drafting
  • Math problems or quantitative analysis
  • Document editing or summarizing pasted content
  • Any task where the answer does not depend on current external data

Standard mode is faster and does not consume your DeepSearch quota. Reserve DeepSearch for tasks that genuinely need multi-source, real-time synthesis.

Tips for Scoping DeepSearch Queries

Be specific about time range
Add "in the past week," "since [date]," or "as of [month/year]" -- DeepSearch will prioritize recent sources that match your window.
Ask for citations explicitly
Prompt: "cite your sources inline" or "link each claim to a source" -- this ensures the report is structured for verification, not just summarization.
Scope the source type
If you want X posts only, say so: "What is being said on X about this?" For broader research, include "check web sources and X." This helps Grok weight sources correctly.
Follow up in the same conversation
After DeepSearch returns a report, you can drill down in standard chat mode without triggering another DeepSearch -- saving quota while still refining your findings.

Aurora: Image and Video Generation

Aurora is xAI's image and video generation model, integrated into Grok as Grok Imagine. As of January 2026, image generation is restricted to paid subscribers only.

What Aurora Can Generate

CapabilityDetails
Text to image Photorealistic images from prompts; 13 aspect ratios; 1K and 2K resolution outputs
Text to video Up to 15 seconds at 480p or 720p; 7 aspect ratios; 24 fps; includes native synchronized audio
Image to video Convert an existing image into a short video clip
Video editing Edit input videos up to 8.7 seconds; "Extend from Frame" adds 2--10 seconds per call
Image editing Edit or remix generated or uploaded images

Generation Speed

  • Still image: approximately 25--45 seconds
  • Video clip: approximately 90--180 seconds

Quality Assessment

Aurora is rated among the top tier of AI image generators for photorealism and prompt adherence. Its standout workflow: prompt to generated image to published X post, achievable within the same Grok session in under 3 minutes.

Access and Restrictions

Paid tiers only (since January 2026): Image generation was restricted to paid subscribers in January 2026 following content policy violations. Free image generation was removed entirely in March 2026.
TierImage accessVideo access
FreeNoneNone
X PremiumLimitedNone
X Premium+~30/monthNone
SuperGrok LiteBasic (480p)480p only
SuperGrok~100/month~20 clips/month (720p)
SuperGrok Heavy~1,000/month~100 clips/month

Content Policy

  • Explicit content, real-person deepfakes, graphic violence, and content involving minors are all blocked
  • The filter operates on semantic intent -- rephrasing or fictional framing does not bypass it
  • Current policy (2026) is substantially more restrictive than at launch

Tips for Image Prompts

  • Keep prompts under 60--70 words; longer prompts dilute focus
  • Lead with camera or composition language (e.g., "wide-angle shot," "close-up portrait")
  • Specify one primary style plus one or two supporting modifiers
  • Avoid contradictory style terms ("cinematic and minimalist and surreal" conflicts)
  • Iterate by changing one variable at a time for predictable results

Grok Voice Mode

Grok's voice capabilities have expanded significantly since the Grok 4 release in July 2025. Voice mode is available to SuperGrok subscribers.

Voice Mode Features

  • Natural spoken conversation with Grok in the iOS and Android apps
  • AI companion personas with distinct voice personalities (Ani, Valentine, Mika, Rudi) -- available in SuperGrok
  • Voice cloning capabilities added in Grok 4.3 (for developers via the API)
  • Apple CarPlay integration live as of the May 2026 iOS app update -- launch the Grok app manually in CarPlay, then speak your prompt
  • Tesla integration with "Hey Grok" wake word -- can control climate, navigation, and media on compatible Tesla hardware (AMD Ryzen, late 2021+) with Premium Connectivity or Wi-Fi

Platform Voice Availability

PlatformStatus (2026)
Grok iOS / Android app Full voice mode; SuperGrok subscribers
Apple CarPlay Live (May 2026); manual launch required; no wake word
Tesla vehicles "Hey Grok" wake word; vehicle control; compatible hardware required

Other Media Capabilities

🎥 Grok 4.3 Multimodal Inputs

  • Text input: All tiers
  • Image input (vision): All paid tiers
  • Video input: Added in Grok 4.3 (flagship)
  • Audio/voice input: Via voice mode (SuperGrok)

Grok 4.3 is the first model in the family to natively process video alongside text, image, and audio inputs.

API Voice Pricing

CapabilityAPI Price
Real-time voice agent (speech-to-speech)$3.00/hr
Text-to-speech (TTS)$15.00 / 1M characters
Speech-to-text, batch$0.10/hr of audio
Speech-to-text, streaming$0.20/hr of audio

Plans and Pricing

Six consumer tiers as of mid-2026. Free-tier access has been progressively tightened since 2025. Verify current pricing at grok.com before subscribing.

💰 Pricing note: Mobile app purchases (iOS / Android) carry higher prices due to Apple and Google app store fees. Always subscribe via the web to pay the lower rate. Annual SuperGrok ($300/yr) saves $60 versus monthly ($360/yr).
Plan Price Models Message Limits DeepSearch Think Mode Image Gen Voice Mode
Free
grok.com / X
$0 Grok 4 Mini (limited) ~10 per 2-hr window None None None No
X Premium Basic
~$3/mo web
~$3/mo Grok 4 (limited) More than free; basic access Very limited Very limited Limited No
X Premium
$8/mo web
$8/mo Grok 4 ~100/day Limited (~10--20/day) Limited Limited No
X Premium+
$16/mo web
$16/mo Grok 4 / Grok 4.3 (staged) ~200/day ~50/day ~100/day ~30/month No
SuperGrok
$30/mo or $300/yr
$30/mo Grok 4 / Grok 4.3 ~1,000/month High / unlimited Yes + Big Brain ~100/mo (720p video) Yes
SuperGrok Heavy
$300/mo
$300/mo Grok 4 Heavy + Grok 4.3 Soft ~10,000/month Unlimited Unlimited + Extended ~1,000/mo + ~100 video Yes + cloning

What Happens at Limits

  • Message limit reached Grok displays a "limit reached" message and blocks further requests until the reset period. Rolling 2-hour resets apply to some message types; daily or monthly resets apply to others.
  • DeepSearch quota DeepSearch uses a separate quota from standard messages. When exhausted, standard chat still works -- only multi-source research mode is paused.
  • Image / video quota Once the monthly cap is reached, generation features are paused until the next billing cycle. Limits have been tightened progressively since 2025.
  • Upgrade prompt At or near limits, Grok typically shows an upgrade prompt. You can dismiss it and wait for the reset.

Free Tier Trends

Free-tier access has been progressively reduced since 2025. As of mid-2026: image generation requires a paid subscription, DeepSearch is essentially unavailable on free, and message quotas have been tightened. xAI has not announced plans to restore free feature access.

Which Plan is Right for You?

If you...Best fit
Just want to try Grok occasionally Free (grok.com)
Already subscribe to X and want Grok in-app X Premium ($8/mo)
Want full X features plus Grok access X Premium+ ($16/mo)
Primarily want AI (not X social features) SuperGrok ($30/mo) -- best value for AI-heavy users
Need multi-agent reasoning and maximum limits SuperGrok Heavy ($300/mo)
Want the best annual deal SuperGrok annual ($300/yr, saves $60)
SuperGrok vs. X Premium+: If your goal is AI capability rather than X social features, SuperGrok at $30/mo offers higher AI-specific quotas than X Premium+ at $16--40/mo. The math favors SuperGrok for users who primarily use Grok as an AI assistant.

SuperGrok Heavy: Multi-Agent Grok 4 Heavy

SuperGrok Heavy ($300/mo) is the highest-tier consumer AI subscription on the market as of mid-2026. Its primary value driver is Grok 4 Heavy multi-agent mode, which runs up to 16 parallel Grok agents and synthesizes their outputs. This is useful for hard research, extended reasoning chains, and tasks that benefit from multiple independent reasoning paths being compared.

  • Up to 16 parallel Expert agents running simultaneously
  • Priority queue access -- requests are processed before lower tiers
  • Early access to new features before general availability
  • Maximum rate limits across all Grok capabilities
  • Grok Build (agentic coding) bundled at no additional cost

When to Use Grok vs. Other AI Tools

Grok has genuine strengths and genuine weaknesses. Knowing which is which saves time and reduces the risk of acting on bad output.

Use Grok when you need

Use caseWhy Grok
Real-time X / social data No other AI has native X firehose access
Breaking news synthesis Grok can summarize X posts as they appear
Social listening / sentiment Queries live posts; returns aggregated sentiment
Trending topic research Real-time trend data from X communities
Multi-source research briefs DeepSearch returns cited reports from live web + X
Long-document analysis 1M token context window handles very long inputs

Consider alternatives when you need

Use caseConsider instead
Document-grounded Q&A (your files) NotebookLM, Claude with document upload
Code generation with broad ecosystem ChatGPT / GitHub Copilot (larger plugin ecosystem)
High-accuracy factual claims Claude (lowest hallucination rate in some tests)
Enterprise Google Workspace integration Gemini Enterprise with M365 connectors
Government or regulated industry use OpenAI (234 federal use cases vs. Grok's 3)

Prompting Tips

Lead with an action verb
"Analyze," "Summarize," "Draft," "Explain," "Compare," or "Generate" -- anchors Grok to a concrete output. Vague openers produce vague results.
Define role, audience, and format
Include "I'm a [role]," "write for a [audience]," and "format as [list / table / JSON / 200 words]." Grok fills gaps with assumptions; give it fewer gaps.
Break complex tasks into steps
Submit multi-part tasks sequentially rather than in one massive prompt. Validate each step before proceeding -- this reduces compounding errors.
Add one more constraint when output misses
If the response is off, add one constraint at a time: audience, format, length, or a "do not include X" instruction. Rewriting the whole prompt is usually unnecessary.
Build a prompt library
Save prompts that work well in a notes doc and reuse them. Prompt engineering compounds -- a good prompt used 20 times is worth far more than a perfect prompt used once.
Use context markers in long chats
Reference earlier responses explicitly: "Regarding the approach you described in your second message..." -- this prevents Grok from losing track of context mid-conversation.

Think Mode: When It Helps vs. Overkill

  • Use Think mode for multi-step math, complex code planning, graduate-level science, legal or financial reasoning requiring logical consistency, and debugging ambiguous problems. The reasoning chain catches errors mid-stream that standard mode misses.
  • Skip Think mode for simple factual lookups, casual conversation, short creative writing, and any task where speed matters more than reasoning depth. Think mode adds latency and consumes a separate quota.
  • Big Brain Mode (SuperGrok only) applies maximum compute to reasoning. Reserve it for the hardest problems -- not every difficult question needs it.
  • Think mode tokens are billed at the same rate as standard completion tokens on the API, so extended reasoning can add cost in high-volume workflows.

Known Limitations

  • Hallucination rate Third-party testing found Grok repeating false claims in roughly 40% of tests on circulating news stories -- higher than ChatGPT (~30%) and Claude (near zero at that test period). Verify important claims before sharing.
  • X-data dependency Real-time answers reflect what is being posted on X, which can amplify misinformation and bias. Grok's source selection skews toward social media over institutional sources.
  • Political content Grok's political content calibration has been tightened in January and March 2026 following studies showing over-reliance on X posts versus mainstream sources for political questions.
  • Knowledge cutoff Without Web Search or X Search tools enabled, the model has no awareness of events after November 2024 (Grok 3/4) or December 2025 (Grok 4.3).
  • Government adoption Grok appears in only 3 of 400+ federal AI use cases documented in 2025, versus 234 for OpenAI-based systems. Trust concerns and lack of FedRAMP compliance are cited as barriers.
  • Ecosystem maturity Third-party integrations and workflow tools are less mature than the ChatGPT ecosystem. Fewer size / cost options for developers fine-tuning performance vs. cost.

Good Habits Checklist

Verify before sharing
Always review numbers, names, dates, and any factual claim before using output externally. Grok can and does produce confident-sounding errors.
Match the mode to the task
Standard chat for writing and coding; X search for real-time social data; DeepSearch for cited research; Think mode for hard reasoning. The wrong mode wastes quota and time.
No sensitive data in prompts
Do not paste PII, confidential business data, or personal health information into Grok. Unlike enterprise tools with data processing agreements, consumer Grok has different data handling terms.
Grok drafts; you finalize
Use Grok to draft and iterate -- not to produce final deliverables without review. You own the quality and accuracy of whatever gets sent or published.
Subscribe on the web, not mobile
iOS and Android app store fees inflate subscription prices. Subscribe via grok.com or x.com in a browser to pay the lower web rate for any paid tier.
Manage your quota intentionally
DeepSearch and Think mode quotas are separate and limited. Use standard chat for tasks that do not need them -- preserve these modes for the tasks where they genuinely add value.

Official xAI documentation, beginner tutorials, and hands-on guides for getting started with Grok.


Learn how to use Grok's DeepSearch mode and xAI's search tools for real-time, web-grounded research.