This is a scannable reference for prompting Grok 4.3, xAI's flagship (launched April 17, 2026) with a 1M-token context window and a December 2025 knowledge cutoff. Bookmark it and copy straight from the tables and prompt blocks below.
Grok runs at grok.com, inside the X app, and via API. Its edge is built-in real-time X (Twitter) search plus modes like DeepSearch, Think, and Grok Imagine. If you want fully written prompts instead of a reference, jump to our best Grok prompts roundup.
1. The prompt formula
Every strong Grok text prompt follows one shape: [Role] + [Task] + [Context] + [Constraints] + [Output format]. You do not need all five every time, but naming the output format is the single biggest quality lever.
| Slot | What it sets | Example |
|---|---|---|
| Role | The expertise Grok adopts | "You are a senior data analyst." |
| Task | The one action to perform | "Compare these three pricing plans." |
| Context | Background, inputs, audience | "For a bootstrapped SaaS founder." |
| Constraints | Limits, tone, length, must/avoid | "Under 200 words, no jargon." |
| Output format | Exact shape of the answer | "Return a markdown table." |
You are a senior financial analyst. Compare the three SaaS pricing plans I paste below for a bootstrapped founder with ~50 customers. Constraints: assume 5% monthly churn, ignore enterprise tiers, keep it under 250 words. Output format: a markdown table with columns Plan, Monthly cost, Best for, One risk, followed by a single recommendation line.Best for: any analysis, writing, or planning task where you can picture the ideal answer's shape before you ask.
2. Grok modes table
Pick the mode to the task. The default handles most chat, but DeepSearch, Think, and Imagine each unlock a different capability.
| Mode | What it does | When to use | How to invoke |
|---|---|---|---|
| DeepSearch | Pulls and cites live sources (web + X) instead of answering from memory | Current facts, research, anything after the Dec 2025 cutoff | Toggle DeepSearch in the composer, or say "research and cite sources" |
| Think | Spends extra compute on step-by-step reasoning before replying | Hard math, logic, debugging, multi-step planning | Toggle Think, or add "think step by step before answering" |
| Grok Imagine | Generates images (Aurora) and ~10s 720p video clips with audio | Visuals, mockups, short clips, storyboards | Open Grok Imagine, or describe the image/video directly |
| Agent Mode (beta) | Batch-edits images and stitches 6-second clips into longer films; runs multi-step tool workflows | Longer video edits, repetitive multi-file jobs | Enable Agent Mode (beta) in the app |
| Heavy | Scales to up to 16 parallel agents for the hardest tasks | Deep research sweeps, complex codebases | SuperGrok Heavy tier (~$300/mo) |
Best for: deciding in one glance whether a prompt needs live data (DeepSearch), reasoning depth (Think), or raw parallel horsepower (Heavy).
3. Real-time X triggers
Grok's real-time advantage lives on X. Open a prompt with one of these trigger phrases and it fetches live posts instead of relying on the knowledge cutoff. Full walkthrough in how to prompt Grok for real-time answers.
| Trigger phrase | What it does |
|---|---|
| "Check X for…" | Forces a live X search on the topic that follows |
| "What are people saying on X about…" | Pulls sentiment and takes from recent posts |
| "posts from the last 24h about…" | Adds a tight time window so old posts are ignored |
| "Summarize @handle's recent posts on…" | Scopes the search to one account |
| "Trending on X in [topic] right now" | Surfaces what is spiking, not evergreen results |
| "cite the posts you used" | Returns links so you can verify each claim |
What are people saying on X about the latest iOS update in the last 24h? Group takes into praise, complaints, and bugs. Cite the posts you used with links, and flag anything that looks like a rumor vs. confirmed.4. Output-format modifiers
Add one line to control how Grok returns the answer. These modifiers stack cleanly on top of any prompt from the sections above.
| Modifier | Add this line |
|---|---|
| Table | "Return a markdown table with columns X, Y, Z." |
| Bullet list | "Answer as a bulleted list, one idea per bullet." |
| JSON | "Output valid JSON only, matching this schema: {…}. No prose." |
| Step-by-step | "Give numbered steps I can follow in order." |
| Cited | "Cite every claim with a link. If you can't cite it, say so." |
| Tone control | "Write it plain and direct" / "keep it playful" / "formal, no contractions." |
| Length | "Under 150 words" / "exactly 5 sentences" / "one paragraph." |
Best for: reusing a good prompt across contexts. Keep the core prompt, swap the last line to change the shape of the output.
5. Grok Imagine quick reference
Grok Imagine uses Aurora for images and generates ~10-second 720p video clips with audio. The image formula is subject + environment + lighting + style + camera/technical, in 30–80 words. Lead with the subject and commit to one coherent style. More ideas in Grok Imagine prompts.
| Element | Guidance |
|---|---|
| Image formula | subject → environment → lighting → style → camera/technical |
| Word count | 30–80 words; lead with the subject, no quality-adjective spam |
| Style | One coherent style per image (don't mix "watercolor" + "photoreal" + "3D") |
| Aspect / orientation | State it: "vertical 9:16 for a phone" or "wide 16:9 cinematic" |
| Camera / technical | "85mm portrait, shallow depth of field," "wide-angle," "golden hour" |
| Video | ~10s, 720p, with audio; describe motion + sound. Agent Mode stitches 6s clips into longer films |
A lone red kayak on a glassy alpine lake at dawn, pine forest and snow-capped peaks in the background, soft golden-hour light with low mist on the water, cinematic photorealistic style, shot wide-angle 16:9 with a slight aerial tilt and shallow reflections.6. Copy-paste starter prompts
Six ready-to-run prompts covering the common jobs. Edit the bracketed parts and go. For a coding-specific set, see Grok prompts for coding.
Research (DeepSearch)
You are a research analyst. Using DeepSearch, give me a briefing on [TOPIC] as of today. Cover: current state, the 3 biggest developments in the last 6 months, key players, and open questions. Cite every claim with a link. Output as a bulleted brief under 400 words, then a one-line "bottom line."Best for: getting current, sourced facts instead of cutoff-limited answers.
Coding
You are a senior [LANGUAGE] engineer. Here is my code: [PASTE]. Task: find the bug causing [SYMPTOM], explain the root cause in 2 sentences, then give the corrected code. Think step by step before answering. Output the fix as a single code block, then a short note on how to prevent it.Real-time monitor
Check X for posts about [BRAND/TOPIC] from the last 24h. Summarize the mood in one line, then list the 5 most-discussed points with a representative post link each. Flag anything urgent (outages, backlash, viral claims) at the top. Cite the posts you used.Image
[SUBJECT] in [ENVIRONMENT], lit by [LIGHTING], rendered in a [SINGLE STYLE], shot [CAMERA/LENS], [ASPECT RATIO]. Example: A ceramic coffee mug on a rain-streaked window ledge, lit by cool overcast morning light, minimalist photoreal style, 85mm shallow depth of field, vertical 9:16.Video (Grok Imagine)
Generate a ~10s 720p clip: [SUBJECT] doing [ACTION] in [ENVIRONMENT]. Camera: [slow push-in / pan / static]. Lighting: [MOOD]. Audio: [ambient sound / music style]. Keep one coherent cinematic style. Aspect: 16:9.Business
You are a growth strategist for a bootstrapped [BUSINESS TYPE]. Context: [REVENUE, TEAM SIZE, MAIN CHANNEL]. Task: propose the top 3 experiments to grow [METRIC] next quarter. For each, give the hypothesis, the exact first step, and how I'd measure success. Constraints: low budget, no new hires. Output as a numbered list.Best for: turning a vague goal into concrete, testable next actions.
Frequently Asked Questions
What is the basic Grok prompt formula?
Role + Task + Context + Constraints + Output format. Tell Grok who to be, what to do, the background it needs, any limits, and exactly how to return the answer. Naming the output format is the single biggest quality lever.
How do I trigger Grok's real-time X search?
Start the prompt with a live trigger phrase like "Check X for…" or "What are people saying on X about…", and add a time window such as "posts from the last 24h". This tells Grok to pull live posts instead of answering from its December 2025 knowledge cutoff.
When should I use DeepSearch versus Think mode?
Use DeepSearch when you need current facts and cited live sources rather than answers from memory. Use Think mode for hard reasoning, math, and multi-step logic where Grok should spend extra compute before answering.
What is the Grok Imagine image formula?
subject + environment + lighting + style + camera/technical, in 30 to 80 words. Lead with the subject, commit to one coherent style, and skip quality-adjective spam like "8k ultra masterpiece".
Can Grok generate video?
Yes. Grok Imagine generates roughly 10-second 720p video clips with audio. Agent Mode (beta) can batch-edit images and stitch 6-second clips into longer films.
What is Grok 4 Heavy for?
Grok 4 Heavy is the SuperGrok Heavy tier (about $300/month) that scales to up to 16 parallel agents for the hardest tasks, such as deep research sweeps and complex multi-file coding.
Which Grok model is current in 2026?
Grok 4.3, xAI's flagship launched April 17, 2026. It has a 1M-token context window, a December 2025 knowledge cutoff, native video input, and document generation for PDFs, spreadsheets, and slides.