Prompt Builder
Pick what you’re making, answer a few questions, and get a prompt that follows current best practice, with a live check and links to learn why. Free, in your browser.
Version 2.0 · · What’s new
Eight kinds of prompt (chat, thinking model, deep research, agent brief, standing instructions, Claude Skill, image, video), ready-made starting points, a live quality check, an improve mode, and prompts saved in your browser.
What it has
Notes
Or ask your assistant to improve it
Copy this and paste it into ChatGPT, Claude or Gemini: it will ask you questions, then rewrite your prompt.
Everything stays in your browser: nothing you type is sent anywhere, and there’s no AI behind this page.
Eight kinds of prompt
Each kind asks for what that kind of AI needs. A thinking model wants a goal and success criteria; an agent needs to know when it’s done and what to ask before acting; an image model needs framing and light.
- Ask an assistant. An everyday task in ChatGPT, Claude, Gemini or any assistant. What is the simple, improved, refined prompt method?
- Solve a hard problem. For a thinking model: a decision, an analysis, a plan or a tricky question. How do you prompt a reasoning model?
- Brief a deep research. For deep research modes that browse many sources and write a report. How do you use ChatGPT deep research for a report?
- Brief an agent. A task for a coding, browser or office agent that works on its own. How do you write a good spec for a coding agent?
- Set standing instructions. Custom instructions, a project, a custom GPT or Gem, or an app’s system prompt. How do you write custom instructions that actually help?
- Package a Claude Skill. A reusable SKILL.md file for a task you repeat (Claude, Codex and other agents read it). How do you package a repeatable workflow as a Claude Skill?
- Describe an image. For ChatGPT Images, Gemini, Firefly, Midjourney and other image tools. How do you make images with ChatGPT Images 2.0?
- Describe a video clip. For Veo, Sora-style and other text-to-video tools. How do you make a video with sound using Veo 3?
Everything stays in your browser: nothing you type is sent anywhere, and there’s no AI behind this page.