Build a website from a prompt: how to write the sentence that actually works
The same request produces wildly different results. Here are the four ingredients of a good sentence, three before-and-after examples, and how three rounds of conversation turn the result into something usable.
Why the same request gives different results
Quality does not come from the tone of your sentence; it comes from **how much executable information it carries**. "Build me a site" and "build a pet-supply store selling dog food and leashes, with add-to-cart and an order status page" differ by an order of magnitude in what the model can work with.
In other words, you are not making a wish, you are writing a requirement. The more specific it is, the closer version one lands, and the fewer rounds you spend fixing it.
The four ingredients of one good sentence
Include all four and the sentence passes. Order does not matter, coverage does:
- What you sell or do, specifically (pet supplies, not "products")
- Who it is for — wholesale buyers and individual shoppers need different pages entirely
- The two to four actions that must work: add to cart, order, book, submit, check status
- Style and constraints: one line about reference, mood, mobile-first, and what to avoid
Three before-and-after examples
The right-hand version saves you a round or two of rework:
- Pet store: "make a pet website" → "a pet-supply store selling dog food, cat food and leashes, where shoppers filter, add to cart, and see order status in their account, and I can edit prices and stock in a back office"
- Restaurant: "make a restaurant site" → "a Sichuan restaurant site, mobile first: menu and opening hours on the homepage, table booking with date, party size and phone number, and the address opens in maps"
- Portfolio: "make me a portfolio" → "my design portfolio: one scrolling page, a one-line intro, project cards with cover, two lines of copy and an external link, email at the bottom, black white and grey, nothing fancy"
The four ways people get it wrong
If the result disappoints, it is almost always one of these:
- Three goals in one sentence ("store + blog + members area") — build the one that matters, add the rest by conversation
- Adjectives with no actions ("premium, modern, international") — the model cannot generate a feature from a mood
- Never saying it needs a backend — if data must persist or money must move, say so; a static site cannot do it
- No boundaries — no language, no currency, no mobile-first instruction means a guess, and a wrong guess means starting over
Three rounds of conversation after the first generation
Version one is a draft; do not throw it away. Edit in this order:
- Round one, structure: "put the featured products above the fold and reduce the nav to Products / About / Contact"
- Round two, voice: "rewrite the copy for individual shoppers, not wholesale buyers"
- Round three, features: "add filtering by pet type, without reloading the whole page"
Make the hero a featured product, one short line, and a buy button
Add a sticky buy bar at the bottom of the product page
Make checkout one column: delivery, payment, then total and pay
What counts as usable
Do not judge the design first. Walk these four checks: buttons respond, entered data survives a reload, the phone layout holds, and payment reaches a real checkout. Once those pass, talk about looks.
Then it meets the problem every new site meets: search engines do not know it exists yet. The checklist for that is below.