Best ChatGPT Prompts: 50 That Actually Work in 2026

Most "best ChatGPT prompts" lists are written by people who never measured whether the prompts work. The 50 below have been used in real work -- by writers, marketers, engineers, founders -- and they survive because they encode constraints, not just clever framings. Each prompt names the role, the format, and the bracket where you substitute your input. Skip the ones that do not fit your work; copy the five that do; edit them after the third use. Public prompt collections plateau quickly. Yours, built from 50 candidates and refined against your own outputs, will outperform anything on the internet within three months.

Table of contents

• Our selection criteria
• Writing prompts (1-7)
• Research prompts (8-14)
• Coding prompts (15-21)
• Marketing prompts (22-28)
• Analysis prompts (29-35)
• Decision-making prompts (36-42)
• Personal productivity prompts (43-50)
• Frequently asked questions
• The bottom line

Our selection criteria

To make this list a prompt had to do three things. First, produce useful output on the first run, not on the eighth. Many viral prompts work only after extensive babysitting; those got cut. Second, generalise across inputs -- a prompt that works for one specific task is a recipe, not a template. Third, name its constraints explicitly. Length, format, audience, role: the prompts below state these directly, because vagueness is the largest single source of bad output.

Each entry follows the same shape: numbered name, one-line use case, and the prompt itself with [BRACKETS] for substitution. Where the prompt benefits from a sample paste, the bracket is named [PASTE]. For the underlying technique behind each pattern, see our complete prompt engineering guide.

Writing prompts (1-7)

1. Voice match. -- For producing prose in a specific style. Prompt: "Read the sample below for voice and rhythm. Write [N words] on [TOPIC] matching that voice exactly, not the topic. Sample: [PASTE]."

2. Tighten draft. -- For making prose denser. Prompt: "Cut 30% of the words from the draft below without losing any claim. Strengthen verbs. Remove hedges. Output the edit only. Draft: [PASTE]."

3. Headline gauntlet. -- For generating headline candidates. Prompt: "Produce 12 headline candidates for an article about [TOPIC]. Constraints: under 60 characters, no questions, no clickbait, must contain a specific noun. Mark the strongest two with '*' and explain the choice in one sentence each."

4. Outline before draft. -- For long-form pieces. Prompt: "Topic: [TOPIC]. Audience: [AUDIENCE]. Word target: [N]. Step 1: outline only -- 6-9 H2 sections with one-line summary each. Stop. Wait for approval before drafting."

5. Single-paragraph rewrite. -- For a tone shift. Prompt: "Rewrite the paragraph below for [AUDIENCE], using [TONE: e.g. plain, blunt, no jargon]. Same meaning, different register. Paragraph: [PASTE]."

6. Story from data. -- For turning numbers into prose. Prompt: "Below is a dataset. Write a 200-word story explaining what it shows. Open with the most surprising single number. Close with what the data does not tell us. Data: [PASTE]."

7. LinkedIn post draft. -- For a professional-network post. Prompt: "Draft a 120-word LinkedIn post on [TOPIC]. Open with one specific observation, not a generic claim. End with a question that prompts replies. No emojis. No hashtags."

Research prompts (8-14)

8. Topic primer. -- For getting up to speed quickly. Prompt: "I know nothing about [TOPIC]. Give me: (a) the three concepts I must understand to read about it; (b) the two most important debates in the field; (c) the canonical reference (book or paper) to read first. Be specific."

9. Steelman an opposing view. -- For testing your own assumptions. Prompt: "I believe [CLAIM]. Write the strongest case against my belief in 200 words, as if you genuinely held it. Use specific evidence. Do not strawman."

10. Gap finder. -- For finding what you do not know. Prompt: "I have just read the document below. List five questions a domain expert would want answered that the document does not address. Document: [PASTE]."

11. Cross-disciplinary search. -- For finding analogues. Prompt: "I am working on [PROBLEM]. Name three other fields where similar problems have been solved. For each, describe the solution in 50 words and explain whether the analogy actually transfers."

12. Source quality check. -- For evaluating evidence. Prompt: "Below is a claim and a source. Rate the source quality 1-10 and explain. Identify any obvious bias. Suggest two stronger sources for the same claim. Claim: [CLAIM]. Source: [PASTE]."

13. Contrarian summary. -- For finding what consensus misses. Prompt: "Summarise the consensus view on [TOPIC] in 80 words. Then list the three strongest minority positions in 30 words each. Mark which minority position has gained ground in the last 5 years."

14. Pre-meeting brief. -- For preparing for a conversation. Prompt: "I am meeting [PERSON/ROLE] tomorrow about [TOPIC]. Produce a 200-word brief: their likely priorities, three questions worth asking, two pitfalls to avoid."

Coding prompts (15-21)

15. Function from signature. -- For writing a function. Prompt: "Write [LANGUAGE] function: signature [SIG]. Behaviour: [3-5 bullets]. Edge cases: [LIST]. No external dependencies. Output the function only, no commentary."

16. Test-driven implementation. -- For getting code that passes a test. Prompt: "Implement code that makes this test pass. Do not add features beyond what the test requires. Test: [PASTE]."

17. Refactor with constraint. -- For cleaning existing code. Prompt: "Refactor the function below. Keep public signature. Reduce complexity. Add input validation. Add one comment per non-obvious branch. Function: [PASTE]."

18. Bug triage. -- For diagnosis. Prompt: "Code is failing with: [ERROR]. Identify root cause in one sentence. List three contributing factors ranked by likelihood. Suggest the smallest fix. Code: [PASTE]."

19. Code review request. -- For honest feedback. Prompt: "Review this code as a senior engineer giving honest feedback. Format: (a) what is good (max 3); (b) what would block in code review (max 5, by severity); (c) the single most important improvement. No filler praise. Code: [PASTE]."

20. Migration plan. -- For moving between technologies. Prompt: "We are migrating from [A] to [B]. Code below is in [A]. Produce: (a) equivalent in [B]; (b) the two behaviours that change in subtle ways; (c) the test that would catch a regression. Code: [PASTE]."

21. Performance hypothesis. -- For optimisation triage. Prompt: "Function below is too slow. Without running it, list the three most likely performance bottlenecks ranked by impact. For each, name the smallest measurement that would confirm or refute. Code: [PASTE]."

Marketing prompts (22-28)

22. Cold email. -- For B2B outreach. Prompt: "Write a cold email from [SENDER ROLE] to [RECIPIENT ROLE] at [COMPANY TYPE]. Open with a specific observation. Body 60-80 words, one concrete value claim, no adjectives. CTA: 15-minute call. Total under 120 words. No greeting. No sign-off."

23. Landing page hero. -- For the top of a page. Prompt: "Write hero copy for a landing page selling [PRODUCT] to [AUDIENCE]. Format: H1 (under 8 words, must contain a specific noun), subhead (under 20 words, names the outcome and the friction it removes), CTA button (under 4 words). No buzzwords."

24. Ad variant generator. -- For paid ad testing. Prompt: "Generate 8 ad variants for [PRODUCT]. Constraints: under 25 words each, vary the angle (price, speed, social proof, fear of loss, fear of missing out, identity, ease, expert endorsement). Mark which two are most differentiated."

25. Customer voice extraction. -- For finding your buyers' words. Prompt: "Below are 10 customer reviews. Extract: (a) the top three phrases customers use that we do not use in our marketing; (b) the most common emotional complaint in their words; (c) the unexpected benefit they cite. Reviews: [PASTE]."

26. Newsletter intro. -- For weekly emails. Prompt: "Write a 90-word newsletter intro for an issue covering [TOPICS]. Open with a specific observation from this week, not a generic claim. End with a one-line tease for the longest article. No exclamation marks."

27. Social-proof rewrite. -- For testimonial polish. Prompt: "Below is a customer quote. Tighten to under 25 words while preserving voice and the specific outcome cited. Cut hedges. Cut filler. Quote: [PASTE]."

28. Positioning statement. -- For a one-line description. Prompt: "Generate 6 positioning statements for [PRODUCT] using the format 'For [BUYER] who [PAIN], [PRODUCT] is the [CATEGORY] that [BENEFIT], unlike [ALTERNATIVE].' Mark the strongest and explain in one line."

Analysis prompts (29-35)

29. Document summary. Prompt: "Read the document below. Output: (a) 50-word executive summary; (b) three most important specific findings, each with quote and section; (c) one question the document does not answer. Document: [PASTE]."

30. Comparison matrix. Prompt: "Compare [A] and [B] on cost, speed, accuracy, lock-in, learning curve. Output as a table with verdict per dimension. End with one paragraph naming when each is the right choice."

31. Risk register. Prompt: "List the five highest-risk elements in the [PLAN/CONTRACT] below, ranked by severity. For each: what could go wrong, evidence in the document, mitigation. Document: [PASTE]."

32. Trend extraction. Prompt: "From the [N] items below: (a) three patterns in 50%+ of items; (b) two outliers worth attention; (c) one prediction for next 12 months with the assumption it depends on. Items: [PASTE]."

33. Root-cause five-whys. Prompt: "Problem: [PROBLEM]. Apply the five whys. After the fifth why, name the actual root cause and a corrective action that addresses it, not the symptom."

34. Hidden assumption finder. Prompt: "Below is an argument. List every load-bearing assumption it relies on, ranked by how easily each could be wrong. Mark the assumption whose failure would most damage the conclusion. Argument: [PASTE]."

35. Stakeholder map. Prompt: "Project: [PROJECT]. List every stakeholder. For each: their interest, their power, the one move that would shift them from blocker to ally. Format as table."

Decision-making prompts (36-42)

36. Pros, cons, recommendation. Prompt: "Decision: [DECISION]. Output: (a) three strongest arguments for; (b) three against; (c) the single counterargument that would change the recommendation; (d) recommendation in one sentence using 'recommend X because Y, unless Z'."

37. Steelman both sides. Prompt: "I am leaning toward [POSITION]. Write the strongest case AGAINST in 80 words as if you held it. Then the case FOR in 80 words. Then your honest verdict and the single piece of evidence that tipped it."

38. Reversibility check. Prompt: "Decision: [DECISION]. Classify as Type 1 (irreversible) or Type 2 (reversible). For each path, name the cost of reversal. Recommend: should this go through full deliberation or can it be made quickly?"

39. Pre-mortem. Prompt: "Project [PROJECT] launches in [TIMEFRAME] and fails badly. It is now [TIMEFRAME + 6 months]. Write the post-mortem: top three causes of failure, ranked by likelihood. For each, the early warning sign and the cheapest preventive action."

40. Cost of inaction. Prompt: "I am considering not doing [ACTION]. List the three highest costs of inaction, ranked by severity. Identify the cost that compounds (gets worse over time) vs costs that are static."

41. Decision under uncertainty. Prompt: "I must decide [DECISION] but data is incomplete. Identify three pieces of information that would most reduce uncertainty, ranked by impact. For each, name the cheapest way to obtain it. If none can be obtained, give a provisional recommendation."

42. Counterfactual. Prompt: "We did [ACTION] last [TIMEFRAME] and outcome was [OUTCOME]. Argue: (a) the strongest case that the outcome would have been the same without [ACTION]; (b) the strongest case that [ACTION] caused the outcome; (c) the single test that would discriminate."

Personal productivity prompts (43-50)

43. Weekly review. Prompt: "Below is my list of completed tasks for the week. Group them by theme. Identify the one task that produced disproportionate value. Identify two patterns of low-value work to drop next week. Tasks: [PASTE]."

44. Inbox triage. Prompt: "Below are subject lines from my inbox today. Classify each as: (D) decide and reply, (W) waiting on someone, (F) FYI archive, (J) junk. Output as table. Subject lines: [PASTE]."

45. Calendar critique. Prompt: "Below is my schedule for next week. Identify: (a) any meeting with no clear decision required; (b) any block under 30 minutes that fragments deep work; (c) the single change that would most increase output. Schedule: [PASTE]."

46. Note synthesizer. Prompt: "Below are my raw meeting notes. Output: (a) three decisions made; (b) every action item with owner and date; (c) one unresolved question that should not be forgotten. Notes: [PASTE]."

47. Goal stress test. Prompt: "I have set the goal: [GOAL]. List: (a) the three most likely reasons I will not achieve it; (b) the leading indicator that would signal each failure 30 days early; (c) one small daily action that addresses the highest-likelihood failure."

48. Learning extraction. Prompt: "Below is a transcript/article on [TOPIC]. Output: (a) the three claims most worth remembering; (b) one claim I should be skeptical of and why; (c) one specific action I should take this week based on this material. Source: [PASTE]."

49. Resistance audit. Prompt: "I keep avoiding [TASK]. Generate three plausible explanations for why, including one I would not want to admit. For each, suggest the smallest experiment to test it."

50. Quarterly retrospective. Prompt: "Below are my goals from the last quarter and my actual outcomes. Output: (a) what worked, with the one specific habit that drove it; (b) what did not, with the one specific cause; (c) the single goal for next quarter that would compound on this quarter's wins. Data: [PASTE]."

Frequently asked questions

Do these work in ChatGPT free or do I need Plus?

All 50 work in the free tier. Plus and Team get you the more capable model and longer context, which matters for prompts that paste large documents (numbers 8, 10, 25, 32, 43-50). For short prompts, the gap is small.

Should I save these in ChatGPT's custom instructions?

Two or three frequently used prompts can go in custom instructions. Beyond that, the constraints accumulate and start interfering with each other. Better to save your library externally and paste the relevant template per session. ChatGPT's "saved prompts" feature in the 2026 interface is the better surface for this.

Why do you use [BRACKETS] instead of putting in real examples?

Brackets force you to think about your specific input before sending the prompt. Pre-filled examples let you fire off lazy versions that produce mediocre output. The bracket discipline is a feature, not friction.

How do I know when a prompt has stopped working?

When you find yourself editing the output more than the input. If you copy the result and rewrite the second half by hand every time, the prompt needs an additional constraint, not another retry. Add the constraint to the template.

Can I use these for Claude or Gemini?

Yes -- most travel cleanly. The marketing and writing prompts work identically across models. Structured-output ones (none in this list, but many in our 30 templates piece) sometimes need a "raw output only" reinforcement on Gemini.

What is the single most underused prompt above?

Number 39, the pre-mortem. People run pre-mortems in workshops but rarely as solo prompts. Done as a 5-minute exercise before any meaningful project, it reliably surfaces failure modes that would otherwise be discovered the hard way.

The bottom line

The fifty above are a buffet, not a meal plan. Pick five. Use them for two weeks. Edit them after the third use. Drop the ones that do not fit your work and add ones you discover in the process. After three months you will have a personal library of fifteen templates that beat anything in any "best ChatGPT prompts" article -- including this one. Continue with our beginners guide if you are early in the journey, or the system prompts guide if you are building production prompts.

Last updated: May 2026