Custom GPTs and Projects
When it's worth creating a custom GPT, what to put into it, and why projects with memory and files exist.
Custom GPT
A custom GPT is ChatGPT with a predefined system prompt, a set of files, and (optionally) connected APIs. Ideal for recurring scenarios: "my content assistant", "my code reviewer", "my sales bot".
What to put into a custom GPT:
- A system prompt with style and rules.
- Knowledge files: brand book, templates, cheat sheets, FAQ.
- If you have an API — a few Actions for specific tasks.
Projects
Projects are "workspaces" inside ChatGPT. A project has its own memory, its own files, and a shared instruction. Chats inside a project see its files and context. Good for:
- one client or one product;
- one large research effort;
- one book or one course.
When to use which
- A recurring task that different people will use → Custom GPT.
- A long task only for you, with files and history → Project.
- An occasional one-off job → a regular chat.
Create one custom GPT for your recurring task. Give it a system prompt and 2-3 files. Use it this week.
Task grader
Copy and adapt to your context. Text in angle brackets should be replaced.
Help me design a custom GPT. The task it will solve: <…> Who will use it: <…> Which files / templates to put into knowledge: <…> Which rules and style: <…> Give me: 1) a system prompt; 2) a list of recommended files; 3) example test queries.
Prompt sandbox
- They make too general a GPT — it's no better than plain ChatGPT.
- They don't update knowledge files, so they go stale.
- They overload a single GPT with ten different tasks.
- One GPT = one task. Five narrow ones beat one universal one.
- Put "how not to" examples into knowledge — that works just as well as "how to".
- Give the GPT an instruction: "if you don't know — ask, don't make it up".
Recurring tasks, team use.
One-off tasks, experiments.
Quiz — 2 questions
Discussion
No comments yet. Be the first!