Step‑by‑Step: Create a GPT
-
1
Open the GPTs editor
In ChatGPT, go to Explore GPTs → Create. The editor has two tabs: Build and Configure.
-
2
Name & describe
Pick a clear, searchable name (e.g., AI SEO Strategist) and a one‑line description users instantly understand.
-
3
Write Great Instructions
Define role, goals, boundaries, refusal rules, and style. Favor specific bullet rules over prose. Keep it under ~400–700 words.
-
4
Add Knowledge (optional)
Upload clean, text‑heavy files you want the GPT to use. Use headings and Q&A formatting for high retrieval accuracy.
For a deeper dive, see /gpt-knowledge/.
-
5
Enable tools or connect Actions
Turn on built‑in tools (browsing, image, code) or add Actions via an OpenAPI schema to call external APIs from conversations.
-
6
Test ruthlessly
Run 3–5 real prompts. If the GPT wanders, tighten the Instructions, add negative directives, or restructure Knowledge.
-
7
Publish
Choose visibility (private / link / public). Add a privacy policy if you use Actions. Share a clean URL and a one‑tweet pitch.
Write Great Instructions
Copy‑paste this baseline, then tweak:
You are a precise, helpful assistant. Scope: [TOPIC]. Prioritize factual, concise answers in bullets. Use uploaded Knowledge when relevant; if not relevant, say so. Style: developer‑friendly, playful one‑liners; avoid fluff. Hard rules: - Don’t invent tools or data sources. - If info may have changed since today, say "as of today" and show how to verify. - Ask one clarifying question only when essential.
Add Knowledge (limits, formats, tips)
Prefer Markdown, TXT, or clean PDFs. Use short sections, FAQs, and explicit labels like Q: and A:. Avoid images‑only PDFs.
- Use semantic headings (H1→H3), lists, and tables where it helps.
- Split giant docs into themed files; title them clearly.
- Embed source identifiers so the GPT can quote sections.
Heads-up: Limits change. Before publishing, confirm current file count/size/token limits and update this page. For a deeper dive (formats, naming, tests), see the full guide: /gpt-knowledge/.
Tools vs Actions
Built‑in Tools
- Web browsing for fresh info
- Image generation / analysis
- Code interpreter for data tasks
Actions (custom APIs)
Connect your API via OpenAPI. You’ll need domain verification and a privacy policy URL.
openapi: 3.1.0
info: { title: Example Weather Action, version: '1.0.0' }
servers:
- url: https://api.example.com
paths:
/weather:
get:
operationId: getWeather
parameters:
- in: query
name: city
schema: { type: string }
responses:
'200':
description: OK
Test, Iterate, Publish
- Create a “golden prompts” list and keep it in your Notes. Run them after each change.
- If Actions fail: check auth headers, domain verification, and that your privacy URL is live.
- Write a 140‑char share blurb and a one‑paragraph product description for your listing.
Custom GPT vs Assistants API
| Feature | Custom GPT | Assistants API |
|---|---|---|
| Setup | No‑code in ChatGPT | Code in your app |
| Distribution | Used inside ChatGPT | Your product/UI |
| APIs/Integrations | Actions (OpenAPI) | Full SDK flexibility |
| Brand control | Moderate | High |
| Best for | Creators, support, small tools | Apps, platforms, SaaS |
What’s new
- New: GPT Instructions + OpenAPI Recipes
- Templates: Knowledge + Prompt Library
- Changelog tracks updates and how to adjust your GPT.
FAQ
Where do I create a GPT?
In ChatGPT → Explore GPTs → Create. That opens the GPTs editor.
Can I connect my API?
Yes, via Actions using an OpenAPI schema. You’ll verify your domain and provide a privacy URL.
What are Knowledge limits?
They change over time. Confirm current file count, size, and token limits before uploading large docs.
Is there an API for my Custom GPT?
Not directly. If you need programmatic control, build with the Assistants API instead.
Templates & Downloads
Hook downloads to your email provider (Formspree, ConvertKit, Beehiiv, etc.).