Build a Custom GPT That Knows Your Pricing, Your Crew and Your Voice
Every prompt in this library gets better when the model already knows who you are. A custom GPT — or a Claude Project, or a saved system prompt — is just a container for the context you'd otherwise paste every single time.
It takes about thirty minutes to build and it's the highest-leverage half hour in this whole library.
On this page
What goes in the instructions
The instruction field is where you encode identity and rules. Keep it under about 1,500 words; longer instructions get diluted.
What to upload as knowledge
Uploads are where the specifics live. Keep them short, plain-text where possible, and versioned.
- Your rate card: labor rates, common material costs, markup, minimum charge.
- Two or three of your best past proposals — the ones that closed.
- Your standard exclusions and assumptions list.
- Your warranty terms, verbatim.
- Your service area list, and any areas you refuse.
- Three past review replies you were happy with, for tone calibration.
Test it like you'd test a new hire
Before you trust it, try to break it. Ask it to price a job you never gave it rates for and confirm it asks rather than guesses. Ask it whether a permit is required and confirm it flags rather than answers. Ask it to write something enthusiastic and confirm it stays in your voice.
Anything that fails the test becomes a new line in the HARD RULES section. Two or three rounds of this and it's genuinely useful.
Keep it current
A rate card from eighteen months ago is worse than no rate card, because you'll trust the output. Put a quarterly calendar reminder to re-upload pricing, and re-read the instructions any time you add a service line or a service area.