Claude Fable 5.1 came out on September 1st. The next day I pointed it at a real client project instead of a demo, gave it one prompt, and walked downstairs while it worked.
It came back with 71 pages.
The client is NZH Builders, a net zero home builder here in San Jose. High performance remodels, envelope work, roofs, whole house jobs. Good business, real crew, and a website that was doing almost nothing for him. I filmed the entire rebuild, waiting included, so you can watch it happen instead of taking my word for it.
Here is what actually happened, including the parts that did not work.
The prompt was the easy part
Everybody wants to know what I typed. That is the wrong question, and it is why most people get slop back when they try this.
What I gave it was short. Redesign this website. Look at the SEO. Learn about this business. Make sure it works for local search. One paragraph, basically.
What made that paragraph work was everything sitting behind it. I already knew what the business does, because I had gone through his existing site page by page. I knew what his customers care about, because I had read his reviews. I knew which cities he actually serves. The prompt was mostly facts I already had, not hopes I was projecting.
When people give an AI builder a blank brief, they get a beautiful generic site for a business that does not exist. The model fills the gaps with invention. Give it your real details and it has nothing to invent.
Thirty six minutes of nothing
Let me kill the fantasy right now. This was not instant.
Seventeen minutes in, no pages. I went and did other work. Ten more minutes, still nothing visible. At thirty six minutes it was still generating imagery, building out the photos so the site would not look like every other contractor page running the same stock library.
Start to finish it was about an hour. Which is still absurd for what came out the other end, but it is an hour, not a magic trick. If you sit and watch the progress bar you will talk yourself into thinking it broke.
What 71 pages actually looked like
When it finished I clicked through the whole thing on camera. A few things stood out.
A page for every city he serves
This is the piece most business owners skip and it is the piece that matters for local search. Los Gatos, Menlo Park, Campbell, Mountain View, and on down the list. Each city got its own page with its own content, not one contact page with fourteen town names crammed into the footer.
That is how a contractor competes for "high performance remodel Campbell" instead of losing every local search to whoever bothered to build the page. Doing that by hand is fourteen separate writing jobs, which is exactly why nobody does it.
A qualifier where the contact form usually goes
The hero does not say "contact us." It asks four questions. What are you working on, which situation is closest to yours, and so on. Pick the roof, pick the envelope, pick the whole house.
That is a better front door than a form nobody fills in. Somebody who answers four questions about their roof is a lead worth calling. Somebody who typed their email into a box might be a competitor.
Real content, not filler
It built a glossary of the terms his crew uses on the job and on assessments, which reads like a dictionary for people who have never heard of a blower door test. It built out the how it works flow, the projects, the blog. It pulled the licensing and the address onto the page where a buyer can see them.
It also kept his own words. I told it to use his real copy rather than write new claims, so the site sounds like his business instead of like a template.
The honest part
I am not going to sit here and tell you it was finished. It was not.
The top section of the homepage was ugly and I said so on camera. Two of the generated house images did not hold up. Several photos need swapping for his real job site shots if he has them. The design needs another pass to clean up.
Call it eighty to ninety percent of the way there. That last ten or twenty percent is the part that takes taste and knowing the client, and no model has that yet. When we build sites with AI, nothing is ever done on the first pass. It gets you moving in the right direction fast, and then a person finishes it.
One more thing worth saying out loud. AI website builders will invent testimonials, awards, and years in business if you let them. Fake social proof on a real contractor's site is the fastest way to lose a client and possibly worse. That instruction goes into every single prompt, every time.
What this means if you own a business around here
The thing that changed is not quality. It is the cost of trying.
A 71 page site with local pages for every city you serve used to be a project you saved up for. It was weeks of writing and thousands of dollars, so most local businesses settled for five pages and hoped Google would figure it out. Now the first full draft costs an hour, which means you can actually see the thing before you commit to it.
That flips the order of the work. Instead of paying to find out whether the idea holds up, you look at a real version of your own site and react to it. The money goes into fixing what is wrong rather than into producing the first draft.
What has not changed is that somebody still has to know what should be on the page. The reason this run produced something usable is that the research happened before the prompt, not inside it.
Watch the whole thing
The full rebuild is on YouTube, unedited waiting and all, including me clicking through the site while it was still building itself in the background.
Watch: Claude Fable 5.1 Rebuilt My Client's Website in One Prompt
If you want to learn how to run this yourself, the exact prompts from this build are free in the Academy. No pitch attached, the prompts are just there.
And if you would rather somebody just do it, that is what we do. On Point Tech Solutions builds websites, funnels, and automation for businesses across San Jose and the Bay Area. Tell us what you are working with and we will tell you honestly whether an AI rebuild is the right call for it.