7 Prompts That Feel Illegal to Know (But Aren't)
The gap between a mediocre AI answer and a jaw-dropping one is almost never the model — it's the prompt. These seven prompts are the ones our power users keep coming back to. They work on GLM-4.7, DeepSeek-V4, and Qwen3.5 out of the box. Copy them verbatim and fill in the brackets.
1. The Brutal Editor
Act as a ruthless senior editor. Read the text below and return:
1. The weakest paragraph and WHY it fails
2. Every sentence that can be cut without losing meaning
3. A rewritten version that is 30% shorter and 2x punchier
TEXT: [paste your draft]Most people ask AI to 'improve' their writing and get polite nothing. Asking for what's broken first forces the model to actually evaluate instead of flatter.
2. The 60-Second Expert
I need to sound credible about [topic] in a meeting in one hour.
Give me:
- The 5 concepts I must understand
- The 3 questions a real expert would ask
- The 2 common misconceptions I should never repeat
Use plain language. No fluff.3. The Code Autopsy
Review this code like a staff engineer doing a hostile code review.
Find: bugs, security issues, performance traps, and one thing you'd
refuse to merge. Rank findings by severity. Be specific, cite line numbers.
CODE: [paste code]'Hostile' is the magic word. Without it, models default to gentle suggestions. With it, you get the review your code actually needs.
4. The Decision Matrix
I'm deciding between [option A] and [option B].
Build a decision matrix with criteria I haven't thought of.
Score both options, then argue FOR the loser to stress-test the winner.
End with a clear recommendation and the one condition that would flip it.5. The Explanation Ladder
Explain [concept] five times:
1. To a 5-year-old
2. To a high schooler
3. To a college student
4. To a professional in the field
5. To an expert who thinks it's overratedThis one is a cheat code for learning. If you can follow all five rungs, you actually understand the concept.
6. The Objection Crusher
Here is my [sales page / proposal / pitch].
List the 10 strongest objections a skeptical reader would have.
Then rewrite the pitch so each objection is answered BEFORE it's raised.
PITCH: [paste text]7. The Second-Order Thinker
I'm planning to [decision].
Don't tell me if it's a good idea. Instead:
1. What happens immediately after I do this?
2. What happens in 6 months because of that?
3. What happens in 2 years because of THAT?
Map the full chain of consequences, including the ones nobody mentions.Why these work everywhere
Every prompt here does the same three things: assigns a sharp role, demands a specific output shape, and removes the model's escape hatch of generic advice. That's the entire formula. Save these seven, and you'll never stare at a blank chat box again.
“The model isn't the moat. The prompt is.”