Implementation

Prompt It in Claude, Then Publish It on Your Website

From prompt to live page On the left, a Claude conversation with three messages: a request for a landing page for a Dubai workshop, a note asking for a shorter headline and speakers higher up, and "Love it. Put it on the site." An arrow leads to the right, where the finished page sits on the website with a banner, three speaker cards and a sign-up button, marked Draft, ready to publish. YOU, IN CLAUDE A landing page for our Dubai workshop. Three speakers, sign-up. Shorter headline. Speakers higher up. Less text on mobile. Love it. Put it on the site. PROMPT · SHAPE · BUILD · PUBLISH You do the talking. Your website does the building. yoursite.com/dubai-workshop AI for Operators A one-day workshop in Dubai Save a seat Register Draft · ready
Describe the page, give Claude notes until it's right, then have it build that page on your real website. No developer queue, no copy and paste.
Max Doyle
Max Doyle
12 Oct 2026
•
7
min read
•
Updated

Claude Design turns a plain description into a page that looks finished. You give it notes the way you would a designer, "shorter headline", "move the testimonials up", "too much text on mobile", until it's exactly what you pictured. Connect it to your website and the same conversation builds that page on your actual site, from your site's own building blocks, so it works like every other page you have. You keep one job: pressing publish.

Why has this been so hard until now?

Because the design and the website lived in different places. AI tools got very good at making pages you could look at. Getting one onto your real site still meant a developer, a ticket and a wait.

The shortcut people tried was exporting the design as code and pasting it in. That works for about a week. Then nobody can edit it, it doesn't quite match the rest of the site, and the next update breaks it.

What changed in September is that Claude Design moved into ordinary Claude conversations, where Anthropic says it can use "your context, files, and connectors". A connector is a secure link between Claude and another tool you use. Connect your website's content system and Claude can work on the site, not just next to it.

The design and the website finally live in the same conversation.

How does a Claude Design become a page on your website?

In four steps, and you only do the talking. We call it the Prompt-to-Page Loop.

Intersect's Prompt-to-Page Loop
StepYou sayClaude doesYou get
1Prompt"A landing page for our Dubai workshop. Three speakers, a sign-up form, our brand."Designs the whole page in your colours, fonts and toneA finished-looking page, not a wireframe
2Shape"Shorter headline. Speakers above the fold. Less text on mobile."Changes exactly that and leaves the rest aloneThe next version, while you're still looking at the last one
3Build"Love it. Put it on the site."Builds it from your site's own sections and fills in the text, images, page title and search descriptionA draft page on your real website
4PublishNothing. You check it and click publish.Nothing. This step is yours.A live page that works like every other page on your site
Intersect's Prompt-to-Page Loop: you do the talking, Claude does the designing and building, and a person publishes.

Step two is the part that feels like magic. You're not writing a brief and waiting a week for round one. You're reacting, the way you would with a designer sitting next to you, and you can go round as many times as you like.

Step three is the part that used to need a developer. Now it's one sentence.

Why is the page stable if AI built it?

Because Claude doesn't make up new code for it. It assembles the page from the building blocks your website already has: the same banners, feature rows, testimonials and forms your developer built and tested. Your site turns those blocks into real HTML exactly as it does for every other page.

That's the difference from pasting in an export. The page loads like the rest of your site, works on phones like the rest of your site, and survives the next redesign, because it's made of the same parts. Your team can still change any word of it by hand afterwards.

The content system we use, and set up for clients, is Sanity. It stores every part of a page as a labelled field, which is exactly what Claude needs to fill a page in properly, and Sanity publishes an official connector for Claude (Sanity).

Claude designs it. Your website builds it. A person presses publish.

What else can you run this way?

Anything on your site that's content rather than code. In practice:

  • Blog posts, written, designed and laid out with images, a search title and a category, saved as a draft. This one was.
  • Campaign pages, one per offer, event or market, all built from the same blocks so they stay on-brand.
  • Site-wide tidy-ups: the old phone number on every page, search descriptions that run too long, images with no description. Claude finds every one and drafts the fix.
  • Everyday edits: the homepage headline, a new team member, the footer links. On our site each is its own editable item, so changing one is a sentence in a chat, not a ticket.

How do we use it on our own website?

Our blog is the clearest example. Each post starts as a conversation. Claude drafts it, then builds it as a preview that looks exactly like our live site, because it's far easier to give notes on a designed page than on a document.

We go back and forth on that preview. When it's right, the same chat puts it into Sanity as a draft, images and all, and someone on the team publishes it.

The rules that keep it safe were ours to set, and Claude follows them. It saves drafts and never publishes. When it was tagging our older posts with categories and found one with a colleague's half-finished edit in it, it left that post alone and told us, rather than pushing the unfinished version live.

Where does it fall short?

It can only build with the blocks your site already has. Ask for a pricing calculator your site has never had and Claude will design a lovely one, then tell you it can't put it on the site. A developer builds that block once, and from then on it's yours to use from a chat.

Messy sites give messy results. If every page on your site is one big text box, Claude can only fill one big text box. Sorting that out is usually the first thing we do.

Someone still has to say yes. Set Claude up so it can save drafts but not publish. The point is speed, not volume: pushing out dozens of unreviewed pages is how you end up on the wrong side of Google's spam policy.

And plenty of sites don't need it. A five-page site that changes twice a year is fine as it is.

What to do next

Pick one page you've been putting off: the event page, the new service, the market you're launching into. If your site runs on Sanity, connect it to Claude, load your brand, and build that page as a draft. If it doesn't, or you'd rather it was set up properly the first time, that's work our Development Studio does.

Want this on your website?

We'll connect Claude to your site, load your brand so every design comes out looking like you, check your site has the building blocks Claude needs, and set it up so nothing goes live without a person saying yes. Then we'll build your first pages alongside your team. If you'd rather do it yourselves, everything you need is above. Book a call

FAQs

Do I need to be technical to use it?

No. You describe what you want and react to what you see, in plain English. Someone technical sets up the connection once, and builds any new blocks your site doesn't have yet.

Could it break our website?

It can be set up so it can't. Claude saves its work as drafts, builds only from blocks your site already uses, and a person publishes. Sanity keeps the history of every change, so anything can be rolled back.

What if our site isn't on Sanity?

Other content systems can work the same way if they have a Claude connector. If your site has no content system at all, there's nothing for Claude to build into, and that's the conversation to have first.

Related reading

Sources