50 Practical Grok Prompts for Every Use Case

Grok is more conversational and more personality-driven than most models, which is a feature right up until you need a straight answer. Ask it something vague and you can get a joke instead of a conclusion. A structured prompt keeps it on task.

The fifty below start with the two things Grok does that other models mostly cannot: pull from live activity on X and the open web, and let you dial its voice up or down deliberately. The remaining three groups cover reasoning, writing and technical work.

Each prompt has bracketed parts to replace. Copy it, swap the brackets, keep the structure. You can run them in Grok directly or in Chat Smith, which lets you send the same prompt to a second model and compare without retyping it.

A request names a task. A prompt names the task and then removes the decisions you do not want Grok making on your behalf. Five parts, every time:

[Role] + [Task] + [Context] + [Constraints] + [Output Format]

• Role - who is answering. One phrase narrows vocabulary and assumptions at once.
• Task - one job, stated narrowly. Two jobs in one prompt produce a compromise on both.
• Context - the background that changes the answer. Leave out the background that does not.
• Constraints - length, tone, what not to do. With Grok this is also where you dial the personality up or down, and it is the part people skip.
• Output format - a table, a list, a word count, one question per turn. Anything unspecified comes back as prose.

A request looks like this:

What do people think about [topic]?

The same question with the formula applied:

Act as an analyst with no stake in the outcome. Search X for posts about [topic] from the last 24 hours. I need this for a decision about [context], so ignore jokes and engagement-farming. Give me the three positions people are actually arguing, in a table with columns for position, roughly how common it appears to be, and the strongest version of that case. No preamble.

Same question, completely different return. Every prompt below applies that formula, which is why they are longer than the one-liners you are used to.

This is the group that justifies using Grok over anything else, and the one most prompt lists ignore. The trick is telling it what to discount as well as what to look for, because raw social data is mostly noise. For the social-media production side rather than the listening side, see our Grok prompts for social media.

1. Topic monitoring

Search X for posts about [topic] from the last 24 hours. Give me the three positions people are actually arguing, with roughly how common each appears to be, and flag anything that looks like coordinated posting rather than organic opinion.

2. Sentiment on a launch

Look at what people are saying about [product or announcement] since it launched. Separate genuine complaints from jokes, and tell me the single most repeated substantive criticism. Do not summarise the positive posts.

3. Why this is trending

[Term] is trending. Tell me what actually happened, when it started, and which accounts are driving it. Distinguish between the original event and the reaction to the reaction.

4. Competitor watch

Check what [company] has posted publicly in the last two weeks. List only things that represent a change from before, and skip anything that is routine marketing.

5. Claim check against live sources

This claim is circulating: [claim]. Find what current sources actually say, tell me whether it holds, and identify the earliest version of it you can locate. If you cannot verify it, say so plainly rather than hedging.

6. Live news brief

Give me a brief on [ongoing story] as of today: what is confirmed, what is reported but unconfirmed, and what is speculation. Label every item with which of the three it is. Under 300 words.

7. Practitioners versus documentation

For [technical question], compare what practitioners are currently recommending against what the official documentation says. Where they disagree, tell me which is more likely to be out of date and why.

8. Tracking a term's meaning

The term [term] seems to be used in several different ways. Search current usage and give me the two or three distinct meanings in circulation, with which communities use which.

9. Detail verification

Find the current confirmed details for [event]: dates, format, and anything that has changed from earlier announcements. Say where each detail comes from, and flag anything you could only find on a single source.

10. Reaction mapping before a decision

I am considering [decision]. Find people who have recently done this and posted about it. Give me the three problems they hit that I would probably not have anticipated.

Treat live-search output as a lead, not a finding. Always ask where a detail came from, and check anything that appears on only one source before you act on it. Social data is fast and frequently wrong.

Grok's voice is its most distinctive feature and the most common complaint about it. Both are solved the same way: treat tone as a constraint you state explicitly rather than something you hope for.

11. Dial the personality down

Answer the following with no jokes, no asides and no personality. Plain, neutral, professional register only. If you feel the urge to add a quip, leave it out. Question: [question]

12. Dial it up, with a limit

Explain [topic] with genuine humour, but every joke must also carry information. Nothing purely decorative. Under 250 words.

13. Blunt assessment

Give me your honest assessment of [plan or draft] with no diplomatic softening, no compliment sandwich and no encouragement. I want the version you would give a peer, not the version you would give a client.

14. Register matching

Rewrite the message below so it sounds like a competent person in a hurry: short sentences, no filler, no pleasantries beyond one line. Keep every piece of information intact.

15. Take a position

Explain [concept] as one side of an argument rather than a neutral summary. Commit to a position, defend it, and acknowledge the strongest counter-point once at the end.

16. Consistent voice across a set

Here is one piece written in the voice I want. Write four more on the topics below in that same voice. If you drift, note where you drifted and what about the topic pushed you.

17. Deliberate simplicity

Explain [topic] using only words a twelve-year-old would know, and no analogies to things a child would not have encountered. If a technical term is unavoidable, define it in the same sentence.

18. Strip the hedging

Answer the question below without using the phrases it depends, there are many factors, or it is important to note. Commit to an answer, then state your confidence in it as a separate line.

19. Persona with a viewpoint

Respond as someone who has been doing [job] for twenty years and is tired of hearing bad advice about it. Warm towards me, impatient with the received wisdom.

20. Tone comparison

Write the same 100-word announcement twice: once warm and human, once clipped and factual. Do not tell me which is better. Tell me what each one signals about the sender.

Ban specific phrases rather than asking for a tone. No jokes works; be professional does not, because professional is a label and no jokes is a rule.

Asking for a conclusion gets you a conclusion. Asking for the steps, the assumptions and the confidence level gets you something you can check. This group matters most when the answer has consequences.

21. Show the working

Work through [problem] step by step, stating each assumption as you use it. At the end, say which step you are least confident about and what would change the answer.

22. Break it down before solving

Break [complex problem] into its independent sub-problems and tell me which to tackle first and why. Do not attempt a solution in this response.

23. Pre-mortem

Assume [plan] failed badly six months from now. Write the post-mortem: the three most likely causes, the early warning signs each would have produced, and what to monitor now.

24. Both sides, properly

Build the strongest case for [position I disagree with], using only arguments its actual proponents make. Then give the two best objections. Do not tell me which you find more convincing.

25. Estimate with visible inputs

Estimate [quantity] and show your reasoning as a chain of stated inputs and multipliers. Say which input the answer is most sensitive to, and give a range rather than a single number.

26. Constraint satisfaction

I need to satisfy all of these constraints at once: [list]. Tell me whether they are jointly satisfiable. If they are not, identify which pair conflicts and what the cheapest constraint to relax would be.

27. Assumption audit

Here is my plan. List every assumption it rests on, ordered from most to least load-bearing, and say what breaks if each of the top three turns out to be wrong.

28. Second-order effects

If [change] happens, list the first-order consequences, then the second-order consequences those would produce. Stop at second order, and mark any that need specific conditions to occur.

29. Decision on my weighting

I care most about [A], second about [B], and not at all about [C]. Compare [options] on that weighting specifically, and tell me whether my weighting rules any option out immediately.

30. Find the flaw in my reasoning

I concluded [X] from [evidence]. Tell me whether that inference holds, what alternative conclusion the same evidence would support, and what additional evidence would distinguish between them.

Ask for the confidence level on a separate line. A model that has to label a claim low-confidence tends to hedge in the right places instead of hedging everywhere at once.

These assume you are writing the thing yourself and want Grok doing what it is good at: diagnosis, compression and arguing against you. Pair them with an AI writing tool for the mechanical passes afterwards.

31. Commissioning outline

Act as an editor commissioning a 1,500-word piece on [topic] for [specific reader]. Give me a section outline with one line on what each section must accomplish and a note on what to leave out. Do not write the piece.

32. Hostile read

Read the draft below as an editor who wants to cut it. Name the three weakest paragraphs, say why each is weak, and rank them by how much fixing would improve the whole. No rewriting and no compliments.

33. Cut to length

Reduce the text below to exactly 400 words without losing any argument. Then list what you cut and which cut you were least comfortable making.

34. Two openings

Write two opening paragraphs for a piece on [topic]: one opening on a concrete detail, one on the stakes. Both under 80 words. Then say which suits a reader arriving from search.

35. Voice match

Here is 300 words of my writing. Analyse it for sentence-length variance, vocabulary level and use of contractions, then write 200 new words on [topic] matching those features. List what you matched.

36. Headline set with rationale

Give me twelve headlines for [article], grouped into three approaches: curiosity, benefit, and specificity. Under each group, say who it works on and who it puts off.

37. Jargon audit

Read the copy below and list every phrase a reader outside the industry would not understand, or would read differently than intended. Suggest a plainer alternative for each. Do not rewrite the whole thing.

38. Structure diagnosis

Map the argument in the draft below as a numbered list of claims, marking each as supported, asserted, or assumed. Do not comment on style. I want the skeleton, including where it does not hold weight.

39. Counter-argument pass

Here is my thesis. Give me the three strongest objections a well-informed sceptic would raise, in their strongest form. For each, say whether my draft addresses it and where.

40. Repurposing with constraints

Turn the article below into a 200-word summary, a five-bullet takeaway list, and a 60-word newsletter intro. Each must stand alone, and you may not reuse sentences across the three.

Ask for the diagnosis before the rewrite. A model that rewrites first hands you something fluent, and you lose the chance to see what was actually wrong.

The unglamorous group, and probably the one you will use most. The move that makes all of these work is asking for a decision or a structure rather than for text.

41. One-page decision brief

Turn the material below into a one-page decision brief: the decision to be made, the two or three real options, what each costs, what we would need to believe for each to be right, and a recommendation. Under 400 words.

42. Code review with priorities

Review the code below and list issues in three groups: will break in production, will cause maintenance pain, and stylistic. Fix nothing yet. Tell me which single change gives the most benefit.

43. Debugging by elimination

Here is the error and the relevant code. List the possible causes in order of likelihood, and for each give the one test that would rule it in or out. Do not propose a fix until I tell you which test passed.

44. Explaining a technical decision upward

Explain why we chose [technical approach] to a non-technical stakeholder who is worried about cost and timeline. One page, no jargon, and be honest about the trade-off we accepted.

45. Requirements from a vague ask

A stakeholder has asked for [vague request]. Write the eight questions I need answered before this can be scoped, ordered so that answering the early ones makes the later ones easier.

46. Data anomaly check

Review the data below and identify anomalies: outliers, impossible values, suspicious duplicates, and any column whose distribution suggests a data-entry problem rather than a real pattern. Explain your reasoning for each.

47. Process documentation

I will describe how I do a recurring task. Write it up as a numbered procedure someone else could follow without asking me anything, and flag every point where I made a judgement call the document needs to explain rather than just state.

48. Pricing sanity check

Here is my pricing for [product] and my cost structure. Tell me what assumptions the pricing implies about customer behaviour, and which of those assumptions is most likely to be wrong.

49. Notes to decisions

From the notes below, extract only decisions made and actions assigned, with the owner and any stated deadline. List separately anything discussed and left unresolved. Ignore everything else.

50. Risk register

For [project], list the eight most likely risks in a table with columns for risk, likelihood, impact, earliest warning sign, and cheapest mitigation. Rank by likelihood multiplied by impact.

Save the ones that worked. A prompt that produced a good decision brief once will do it again next quarter, and rewriting it from memory every time is the most common waste in this category.

Chat Smith carries several Grok tiers - Grok 4.5 and Grok 4 Fast among them - alongside models from other providers, and our Grok models overview sets out which tier suits which job. The reasoning prompts above are the best place to see the difference between a fast tier and a heavier one on the same input.

Worth knowing what does not transfer: the live-search prompts in the first section depend on Grok's own access to current activity, so running them against a model without that access will produce confident guesses rather than findings. Everything else in this article works on any capable model. For visual work, our Grok image prompts cover that separately, and you can work straight from an AI image generator if you would rather skip the chat step.