Saurabh Singh   •   11 months ago

Right ways to prompt Bolt.new or any LLM:

After a lot of trial and error, I’ve realized prompting is everything. These techniques helped me get better, faster, and cheaper results:

- Be specific. Don’t just say "Make it better" — tell the AI exactly what you want changed and where.
- Mention file names or functions. For example: "Inside login.tsx, add client-side validation for empty email/password."
- Break large tasks into smaller ones. Handle one feature or fix per message.
- Avoid repeating previous messages — the AI already remembers the recent context (unless you’ve reset it).
- When possible, ask before building. For example: “Can you suggest the best way to add role-based access? Don’t implement yet.”
- If the AI struggles, pause and do some research yourself — then guide it with updated info.
- End a long session by asking the AI to summarize everything — then copy/save that for future use when duplicating the project.

Why I’m sharing this:

I spent a lot of time (and tokens) learning how to work properly with LLMs. I hope this helps new users work smarter, save tokens, and avoid the frustration I went through. These lessons apply not just to Bolt, but to all LLM-powered tools.

If you’ve found other helpful tricks, I’d love to hear them. Let’s build better, not bigger. :)