20 Rules of Using Claude
Commands, prompts, and the eleven-minute fix behind all of it
A founder I work with called me last month, half annoyed, half ready to cancel her subscription.
She’d spent three weeks using Claude to write her company’s emails and posts. Everything it gave back sounded, in her words, like a brochure nobody asked for. So I asked her to show me. She typed seven words: write a LinkedIn post about our new feature. Claude handed her four tidy paragraphs about thriving in a fast-paced world. She wasn’t wrong. It was lifeless.
We fixed it in eleven minutes.
No new model. No secret prompt. Eleven minutes of setup, which I timed because she didn’t believe me. By the end she typed the exact same lazy request, and the post came back sharp, opinionated, in the voice she uses with her own customers. She read it twice and went quiet, then said: that’s the thing I’ve been chasing for a month.
Why the same question gives two different answers
Here’s what changed in those eleven minutes, and it’s the whole article in one idea.
The first time, Claude knew nothing about her. No company, no audience, no opinions, no sense of how she talks. So it reached for the safest, blandest version of everything. The second time, it had her context loaded before she typed a word. Same model. Same question. A completely different answer, because the thing feeding the model was different.
That’s the lever. Not clever wording. Context.
Below are the 20 rules I now give every client and set up on their account the same afternoon. Each one comes with the exact thing to type or click. Steal all of them. You can do the first four before your coffee goes cold.
