How to Talk to Claude So It Builds What You Want

The secret to getting great stuff out of Claude is not some magic prompt formula. It's describing what you want clearly and being honest when it's not right yet. That's the whole skill, and you already have most of it. "Prompt engineering" sounds like a job title. In practice it's closer to being a clear communicator with an opinion. Here's how I actually do it, as a guy who builds real things and can't code.
Stop hunting for the perfect prompt
People waste a ton of time looking for the one magic sentence that makes the AI do exactly what they want on the first try. That sentence doesn't exist, and chasing it misses the point. Claude is a conversation, not a vending machine. You don't need the perfect input. You need a decent start and the willingness to refine.
Stop hunting for the perfect prompt. A rough start plus a few rounds of refining wins every time.
So just start. A rough first try plus three rounds of "no, more like this" beats an hour spent crafting one perfect prompt every single time.
The things that actually help
Here's what genuinely improves what you get back, no jargon.
- Be specific about the outcome. "Write me a homepage" gets you something generic. "Write me a homepage for a dog-walking business, friendly tone, with a section for prices and a big 'book now' button" gets you something usable. Detail in, quality out.
- Describe the feeling, not just the facts. You don't need design words. "Make it feel calmer," "make it more premium," "make it warmer" all work shockingly well. Claude translates vibes into specifics.
- Give it context. Who's it for? What's the goal? One sentence of "this is for busy parents who feel guilty" steers the whole thing.
- Show, don't just tell. Paste an example of something you like and say "in this style." Paste the thing that's broken and say "fix this." Concrete beats abstract.
- Ask it to ask you. A genuinely useful trick: "before you build this, ask me any questions that would help." It'll surface stuff you forgot to mention.
When it gets it wrong (it will)
This is the part beginners get discouraged by, and they shouldn't. Getting it wrong is step one of getting it right.
- Say what's off, plainly. "Too long." "Too formal." "The colors are muddy." "That's not what I meant, I wanted X." It doesn't have feelings, so be direct.
- Paste errors straight back. If something technical breaks, copy the error, paste it, say "this broke, fix it." It wrote it, it can fix it. You're the messenger.
- Course-correct early. If it's drifting, stop it and reset, don't let it build three more steps on a wrong foundation.
- Start a fresh chat if it gets stuck in a rut. Sometimes a conversation goes sideways and the cleanest fix is a new one with a better opening description.
The mindset that makes it work
Treat Claude like a wildly capable teammate who takes everything literally and has no ego. You don't have to be polite or clever. You have to be clear, and you have to know what "good" looks like so you can tell when you've got there. That second part, the taste, is the thing you bring that the AI can't. (It's the same reason I can build a whole website without coding.)
Get comfortable with the back-and-forth and you'll get more out of Claude than people who've memorized a hundred "power prompts." The conversation is the skill.
New here? Start with how to use Claude. Want to see this in action on real builds? Watch one a week.
FAQ
How do I write good prompts for Claude?
Be specific about the outcome, describe the feeling you want, give a sentence of context, and refine from there. It is clear communication with an opinion, not a secret formula.
Do I need to learn prompt engineering?
No. Describing what you want clearly and saying when it is off is the whole skill. A rough first try plus a few rounds of refining beats hunting for one perfect prompt every time.
What do I do when Claude gets it wrong?
Say what is off plainly, or paste the error straight back and say fix this. If a chat gets stuck in a rut, start a fresh one with a better opening description.
Keep reading more from the field
Don't miss the next one.
Every build, start to finish, free in your inbox.
Get the notes →