PromptSharp
Issue #002 · Thu, Jun 25, 2026
Sharpen how you instruct AI — from your first prompt to full agents. |
Chain-of-thought: make it think before it answers
For anything with steps, math, or logic, 'show your reasoning first' is the cheapest accuracy boost there is. |
Models that answer instantly often answer wrong on multi-step problems because the answer is generated before the reasoning exists. Asking the model to reason step by step before giving the final answer forces it to allocate compute to the actual work. The key detail most people miss: put the reasoning before the answer, and separate them clearly, so the final answer is conditioned on the worked steps — not the other way around. |
▸ Copy-paste prompt
Solve this carefully.
First, under a heading 'Reasoning:', work through the problem step by step. List every assumption and show each calculation.
Then, under a heading 'Answer:', give ONLY the final answer in one line.
Problem:
A store sells pens at 3 for $4.20. A customer buys 17 pens and pays with a $50 bill. The store rounds the total UP to the nearest 5 cents. How much change is returned? |
▸ Level it up
New to this: Add 'Think step by step, then give the answer' to any math, logic, or multi-part question. That one line fixes most wrong answers.
Vibe coder: Force structure: 'Reasoning:' then 'Answer:'. When it's wrong, you can read the steps and see exactly where it slipped — then correct that step.
Advanced: Use structured CoT for verifiability but hide it from end users (parse out the 'Answer:' block). For higher stakes, sample multiple reasoning paths and take the majority answer (self-consistency) — accuracy rises with the vote, not the eloquence. |
Today's takeaway: Reasoning before the answer, not after. Order is the whole trick. |
|
|
|
Know someone who'd sharpen their prompts with this?
Forward it. Refer 3 friends and we'll send you our private prompt-pack.
Refer & get the prompt-pack → |
|
|
|
|