The “1,000 ChatGPT prompts” PDFs going around are mostly useless. They’re padded with “Act as a senior CEO with 30 years of experience
” preambles that GPT-4o ignores anyway.

Here are 20 prompts that actually move work forward. Each one is short. Each one solves a specific problem. We use these weekly.

How modern prompts have changed

GPT-4o (and Claude Sonnet 4) is good enough now that specificity beats roleplay. The 2023-era “Act as X” prefix doesn’t help. What helps:

  1. State the goal in one sentence.
  2. Give 1-2 constraints (length, tone, format).
  3. Provide an anti-pattern (what NOT to do).
  4. Show one example when format matters.

Every prompt below follows that pattern.

Writing & content (5)

1. The “kill the fluff” rewrite

Rewrite this paragraph to be 30% shorter. Cut filler ("in today's world",
"it's important to note", "when it comes to"). Keep every concrete claim.
Don't make it sound like ad copy.

[paste paragraph]

Use when: Polishing your own writing or AI-generated drafts.

2. The honest book/article summary

Summarize this article in 5 bullets. Each bullet must contain a specific
claim or fact, not a generic statement. If a bullet starts with "The article
discusses..." or "It explains...", rewrite it.

[paste article or URL]

Use when: You’d rather read 5 bullets than 5,000 words.

3. Headline alternatives that aren’t clickbait

Give me 5 headline options for an article titled "[CURRENT TITLE]".
Constraints: max 60 characters, no question marks, no "ultimate" or "complete",
no power words ("shocking", "secret"). Each headline should imply a specific
takeaway, not curiosity bait.

Use when: Publishing anything.

4. Email reply at the right tone

Write a reply to this email. Tone: professional but warm, max 4 sentences.
Don't start with "I hope this email finds you well." Don't end with
"Looking forward to hearing from you." Get to the point.

Original: [paste email]
My response goal: [one sentence]

5. The “what am I missing” sanity check

Read this draft and tell me three things:
1. What's the strongest claim?
2. What claim is weakest or could be challenged?
3. What's one thing a smart reader would expect to see that's missing?

[paste draft]

Use when: Self-editing.

Research & analysis (4)

6. Compare X and Y honestly

Compare [TOOL A] and [TOOL B] for [SPECIFIC USE CASE].
Format: a table with rows for Price, Best for, Weakness, Free tier.
Don't hedge with "depends on your needs." Pick one as the better choice
and say why in one sentence after the table.

7. Steel-man the opposite view

I believe [STATEMENT]. Argue the opposite as well as you can.
Don't strawman my position. Give me the strongest 3 reasons someone smart
might disagree with me, plus one piece of evidence I should look up.

Use when: Avoiding confirmation bias.

8. The “is this a real thing” check

Quick check: is "[CLAIM OR STAT]" actually true, or is it a frequently
repeated number that nobody verified? If it's likely true, give a source.
If you can't verify it, say so plainly.

Use when: Before citing any “studies show” stat.

9. Industry jargon translator

Translate this paragraph from [INDUSTRY] jargon into plain English. Don't
oversimplify — keep technical terms when they matter, but explain them in
one phrase the first time. Target reader: smart generalist.

[paste]

Coding (4)

10. The “explain this code” prompt

Explain this code in plain English. Skip "this function does..." opening —
go straight to what it accomplishes and why. Flag any line that looks
suspicious, buggy, or unusual.

[paste code]

11. Refactor with constraints

Refactor this function. Constraints:
- Don't change the public API
- Don't add new dependencies
- Reduce length by at least 30%
- Add inline comments only where logic is non-obvious

[paste function]

12. Find the bug

This code throws [ERROR]. Walk through it step by step and identify the
most likely cause. Don't list 5 possibilities — give me your single best
guess and tell me how to verify it.

Code: [paste]
Error: [paste]

13. Test cases I haven’t thought of

Here's a function. Generate test cases focused on edge cases and inputs
I'm likely to forget — empty inputs, boundary values, unicode, very large
inputs, malformed data, race conditions if relevant.

[paste function]

Productivity (3)

14. Decompose a vague task

I have a vague goal: "[GOAL]". Break it into 5 concrete next actions I
could start in the next 30 minutes. Each action should be a verb-first
sentence under 10 words. Skip "research" actions unless they're truly
necessary.

Use when: Procrastinating because the task feels too big.

15. The “what would a pro do” prompt

Imagine someone who has done [SPECIFIC TASK] hundreds of times. What's
the one shortcut, tool, or counterintuitive habit they'd use that a
beginner would miss? One specific answer, not a list.

16. Meeting prep in 60 seconds

I have a meeting with [WHO] in [HOW LONG]. The goal is [ONE SENTENCE].
Give me:
- One question to lead with
- One pushback I should be ready for
- One thing to confirm before I leave the meeting

Decision-making (4)

17. The “should I” pre-mortem

I'm about to [DECISION]. Walk me through the most likely failure mode
in 3 sentences. Then tell me one cheap way to test the assumption that
would matter most if it's wrong.

18. Trade-off forced rank

I'm choosing between [A] and [B] for [GOAL]. Don't say "it depends" or
list pros and cons. Pick one. Tell me why in 3 sentences. Then tell me
the single condition under which I should pick the other one instead.

19. The “what’s the worst that could happen” check

I'm planning to [ACTION]. What's the realistic worst case, the recovery
cost if it happens, and the probability I should mentally assign to it?
Be specific — not "things could go wrong."

20. The “second-order effects” prompt

If I [ACTION], what happens? Don't tell me the obvious first-order effect.
Give me 2 likely second-order effects (consequences of the consequences)
that I might miss in my planning.

What we deliberately left off

  • “Act as a [role]” prompts. Modern models don’t need them, and they bloat the context window.
  • “Take a deep breath and think step by step.” Already built into the model. Wastes tokens.
  • “You are an expert.” Doesn’t change output quality. We tested.
  • Multi-thousand-character “ultimate” prompts. They get truncated mentally and produce worse output than focused 50-word prompts.

The pattern: modern LLMs respond best to specificity, not flattery or ceremony. Treat them like a competent colleague, not a method actor.

How to use these

Save the prompts you actually use as ChatGPT custom GPTs or Claude projects so you don’t paste the same template 50 times. The 5-10 prompts you use weekly should be one click away.

For a deeper dive into when to use ChatGPT vs Claude vs Gemini for which type of prompt, see our head-to-head comparison.

And if you want the full PDF of these prompts plus the rest of our AI workflow, grab the free guide.