TLDR: Don't ask AI to do the thing directly. Ask it to interview you first — what are your constraints, what have you not thought of, what would you recommend? Collect those answers into a brief. Use that brief as your real prompt. This works for anything: websites, business plans, marketing campaigns, internal tools, hiring processes. The example below is a website, but the method is universal. A 350-word brain dump became a 1,200-word spec, and the result wasn't even in the same category.
You've got an idea for a website. You open Claude or ChatGPT and type something like:
"I'm starting a surf lesson business in Portugal. Can you build me a website?"
And the AI will do it. It'll give you a homepage, maybe a contact section, some copy about how your services are "tailored to your needs." It works. Technically.
But it's thin. It's missing things you didn't know to ask for — SEO tags, legal pages, a contact strategy, cookie consent, schema markup. You never mentioned them, and the AI didn't want to bother you with questions.
This isn't a website problem. It's a prompting problem. Ask AI to "write me a marketing plan" and you'll get five generic bullet points. Ask it to "draft an employee handbook" and you'll get boilerplate. Ask it to "plan my product launch" and you'll get a timeline that could apply to literally any product. The pattern is always the same: vague input, vague output.
The quality of what you get out is almost entirely determined by the quality of what you put in. But to write a detailed prompt, you need to know what details matter. If you already knew that, you wouldn't need the AI's help. It's like walking into an architect's office and saying "build me a house" — you'll get a house, but not the one you actually wanted.
So what do you do?
Make the AI interview you first
Instead of asking the AI to build the thing directly, you ask it to help you figure out what to ask for. Thinking partner first, builder second.
I do this with clients all the time. Before I touch any automation, I spend the first few sessions just asking questions. What breaks when you're on vacation? Where do you lose money to slowness? The answers shape everything that comes after. You can do the same thing with AI — for free.
Say you've got a rough idea for a corporate surf retreat business called "Salt & Suit." If you dump all of it into an AI and say "build me a website," you'll get one page, some blue colors, and generic copy. No SEO strategy, no legal compliance, no plan for how anyone will find it.
But if you say this instead:
I have a business idea and I want you to eventually build me a website. But not yet. First, help me think through what the website actually needs. Here's my rough idea: [your brain dump]. Ask me the questions I haven't thought of. Tell me if I'm missing something obvious. As we talk, build up a detailed brief that captures all our decisions. That brief becomes the build prompt later.
Now the AI starts asking things like:
- Where will you host this? (It might suggest Astro + Tailwind for static sites with good SEO.)
- How will people contact you? A form? A Calendly link? Each has trade-offs.
- What about legal stuff — privacy policy, GDPR cookie consent?
- Will the site be English only, or Portuguese too?
- Do you have photos? If not, where will the visuals come from?
You answer these. After three or four rounds, you've got a document that's no longer a vague idea — it's a proper brief. And that brief is your prompt.
The template
If you want to try this yourself, adapt this to whatever you're building:
I have an idea for [what you're building] and I eventually want you to help me build it. But not yet.
First, I want you to be my thinking partner. Here's my rough idea: [your brain dump — be as messy as you want].
Before we build anything:
- Ask me questions I haven't thought of. Explain why each one matters, and suggest what you'd recommend if I'm not sure.
- If something in my plan is a bad idea, tell me directly.
- Think about this from the end user's perspective. What would they expect? What would make them trust this?
- After each round of questions, update a running brief that captures all our decisions. This brief becomes the build prompt.
Ask me questions in small batches so I don't get overwhelmed. Don't build anything yet.
You go back and forth a few times. The brief grows. When you're done, you've got a prompt that's five or ten times more detailed than what you started with — not because you spent weeks researching web development, but because you had a conversation.
The worked example: before and after
Here's the full process from the surf retreat scenario. The rough idea on day one, then what came out the other end.
The starting prompt (what's in your head)
Elena had a ~350 word brain dump about her corporate surf retreat business "Salt & Suit" in Portugal. Good energy, clear value proposition — but zero implementation detail. She mentioned the business idea, her background (finance in London, surfing in Ericeira), her business partner Marco (ISA-certified surf instructor), and the tone she wanted (playful but professional).
The refined prompt (what came out the other end)
After several rounds of AI-assisted questioning, Elena's 350 words became 1,200. The extra words aren't fluff — they're decisions about:
- Hosting: Astro + Tailwind on Cloudflare Pages for static SEO
- Contact strategy: Calendly vs forms vs phone number, with trade-offs
- Legal: Privacy policy, terms, GDPR cookie consent, Portuguese/EU jurisdiction
- SEO: Sitemaps, robots.txt, canonical tags, OpenGraph, JSON-LD, topical maps
- Analytics: PostHog for tracking
- Content strategy: Google E-E-A-T compliance, AI-quotable content formatting
- Multilingual: English first, Portuguese later with a disabled language switch
- Self-containment: No external resources unless they add marketing value
- Go-to-market: A full cold-start promotion strategy
- Visual identity: SVG logo combining a surfboard and necktie, ocean blues and coral accents
Same AI. Same person. Wildly different output.
The prompt is the product
People treat prompts as throwaway inputs — type something, get a result, move on. But for anything that's not trivial, the prompt is the product. It's the spec. The blueprint. You wouldn't build a house from a napkin sketch.
The example above is a website, but I use this exact method for everything. Designing an automation workflow for an accounting firm? Interview first, build second. Planning a content strategy? Same thing. Migrating a client's data pipeline? You'd better believe we're spending the first hour on questions, not code. The two-step process works wherever the gap between "what you know to ask for" and "what you actually need" is wide — which is most places.
I spend the first chunk of every client engagement just asking questions and building the brief before anyone touches a keyboard. But you can get 80% of that value on your own, for free, by running this two-step process with any AI chatbot. The AI won't know your industry as well as a consultant would, but it'll catch the 30 things you forgot to think about. That's usually enough.
Originally published at lobsterpack.com.
Top comments (0)