What it does
This is a meta-skill: instead of using a ready-made GPT you build your own for the job. You set the instructions (the system prompt), attach knowledge files, turn on capabilities — web search, code execution (Code Interpreter), image generation — and, if you want, wire up external Actions (your own API). The result is a reusable assistant you never have to set up again.
How to set it up
- You need a paid plan (creating GPTs is available on ChatGPT Plus/Team/Enterprise).
- Open the editor: chatgpt.com/gpts/editor (or "Explore GPTs → + Create" in the left sidebar).
- The Configure tab: name, instructions, Knowledge (files), the Capabilities checkboxes (Web/Canvas/DALL·E/Code) and, if needed, Actions.
- Save it: private, link-only, or published to the GPT Store.
When to use it
- ✅ A recurring scenario with fixed rules and your own source material.
- ✅ You need an assistant you can hand to the team as a link.
- ❌ A one-off task — a regular chat is enough; not on ChatGPT — build the equivalent as a Gemini Gem.
Example
You build a "Metrics analyst" GPT: instructions say "calculate MoM/YoY, plot charts, give a takeaway", Code Interpreter is on, and the report template sits in Knowledge. Drop in a CSV and you get a finished breakdown in your own format.