How I Build Landing Pages with AI in Under an Hour

Last month I shipped 3 landing pages. Total time across all three: 4 hours.

Not 4 hours each. 4 hours combined, including the one for a friend’s consulting practice I turned around in an afternoon. A year ago that would have been a week of work. Here’s exactly what changed and what the process looks like now.

The old way was a production problem, not a thinking problem

The bottleneck used to be making the thing, not knowing what to say. Hire a designer and developer: weeks of back-and-forth, $2,000–5,000 minimum. Use a template tool: fast but the result always looked like a template. Build it yourself from scratch: doable, but two days minimum if you knew what you were doing.

Most ideas never got a page. The cost of trying was too high. So you’d sketch something in Notion, share a Google Doc, or just describe the product in a message — and the idea would stay invisible.

The constraint wasn’t thinking. It was production.

What I actually do now, step by step

Step 1: Get the message nailed down before opening any tool.

This is the part people skip, and it’s why their AI-generated pages feel empty. The message isn’t a tagline. It’s the one sentence that a tired person at midnight would read and immediately understand why this exists.

For MeetCal, I spent 20 minutes on this before touching anything. Landed on: “Stop copying your availability into emails. Share a link instead.” Once that was written, everything else followed from it. The hero section, the objection handling, the call to action — all of it was already implied.

Step 2: Tell Claude the story, not the spec.

I open Claude Code and describe what I want in plain language. Not “create a landing page for a scheduling tool.” I tell it: what the product does, who it’s for, what they’re probably skeptical about, what I want them to do when they leave the page. I mention the framework (Astro, in this case) and paste in the one-sentence message from step 1.

I describe the sections I want and any specific copy ideas I already have. The more context I give, the less cleanup I do later. Ten minutes of input saves thirty minutes of iteration.

Step 3: Fix the copy. The design is usually fine.

Claude generates something structurally solid. The layout, the section order, the visual hierarchy — all reasonable. I almost never touch the design.

The copy is where the work is. The model knows how landing pages sound, so it writes landing-page-sounding copy: “Powerful solution for modern teams.” “Streamline your workflow.” Vaguely optimistic, structured in the right places, meaning nothing.

I rewrite the headlines with specific language. I cut anything that sounds like it came from a product brief. This iteration loop — regenerate a section, rewrite the headline, check on mobile — runs for 20–30 minutes per page. That’s the whole job.

Step 4: Push to GitHub, Vercel picks it up.

Thirty seconds. Done.

The thing AI consistently gets wrong

AI-generated copy is always too safe. The specific insight that makes your product different will not come from the model — it has to come from you.

For ShipPad, the insight was that builders need a place to log what they shipped, not another task manager. That framing is what makes the page work. I had to bring it. Claude dressed it up well once I handed it over, but the idea itself came from watching people complain about this exact problem for six months.

This is the mistake most people make. They think AI will figure out their message. It won’t. It knows the genre. It’ll write something that sounds correct but says nothing specific. The model has no opinion on what makes your product true — only you do.

The constraint moved, and that changes everything

The question is no longer “can I build this?” It’s “do I know what to say?”

Production is nearly free now if you’re willing to iterate. An hour of focused work with Claude Code produces something real you can put in front of people. The three pages I shipped — MeetCal, ShipPad, a consulting practice — none took more than 90 minutes of actual work once the message was clear.

That thinking still takes as long as it takes. I spent more time writing the MeetCal sentence than I spent building the first version of the page. AI doesn’t compress that work. It just means that when you’re done, the page can exist the same day.

Most people are not blocked by production anymore. They’re blocked by not having done the thinking. That’s a different problem, and it won’t be solved by a better prompt.