You don't need to know how to write a prompt. You need to know what you want — and how to ask for it once, so your whole team can repeat it.
That's a skill. These guides show you how to build one, in plain steps, using the tools you already work in.
If you haven't connected yet, start with Connect Propalt AI to Claude and ChatGPT.
Step 1 — Define the skill
A connection lets you ask questions; a skill makes the answer repeatable. Three things turn one into the other:
- Standing instructions — the one task the skill performs, and the inputs you'll give it (an address, a postcode, a UPRN).
- A fixed output format — the exact shape of the answer (a table with set columns, a short brief, a CSV), so everyone gets the same deliverable.
- A data rule — tell the skill to use Propalt AI for anything factual (sold prices, EPC, tenure, comparables) and to flag missing data rather than estimate.
Step 2 — Save it as a reusable skill
Save the configuration so it loads automatically and anyone can run it:
- Claude — save it as a Project, with the instructions and the Propalt AI connector enabled.
- ChatGPT — save it as a Custom GPT.
Give it a name your team will recognise, and it's ready to share.
Example: a comparable-evidence skill
Standing instructions for a "Comp finder" skill:
Anyone who opens it pastes an address and gets the same structured evidence — no prompt engineering, no reminding it to use real data.
Test it before your team relies on it
- Run
pingfirst — it's a free call that confirms the connection without spending credits. - Use a property you can verify independently, and check the figures match.
- Try the awkward cases (a new-build with no sale history, a flat in a block, a Scottish address) and confirm the skill flags missing data instead of inventing it.
- Watch your credits — each data call draws on your balance;
pingis free.
