What Ships · Claude
Build a website with Claude: what you get, and what to check before launch.
You get a plain walkthrough of what Claude, Claude Code and Claude Design produce, plus the launch checks your site needs. And you get a way to score it. The free assessment reads your public pages in two minutes (it's quick because it only sees what a visitor sees), but it cannot see your logins, your analytics or your field data. Think of it as a look from the street, not a tour of the building.

On this page 11 sections
- 00Key takeaways
- 01How do you build a website with Claude today?
- 02What is the difference between Claude Design, Claude Code and the chat?
- 03What does a Claude-built site ship by default?
- 04What should you check before you publish a Claude-built site?
- 05How do you deploy and keep a Claude-built site?
- 06Where do Claude-built sites score well, and where do they drop?
- 07How do you check your Claude-built site in two minutes?
- 08Questions, answered
- 09Check your own site
- 10About the author
Key takeaways.
- Judge your Claude-built site by what the live URL ships to a browser, not by which Claude surface produced it.
- Open at least two routes for every changed shared template, since one metadata helper can return identical values on every page.
- Parse every JSON-LD block, then confirm each marked-up value is visible on the page, because valid markup guarantees no rich result.
- Run the free assessment after each redeploy: it takes about two minutes, and you get three checks per hour per connection.
How do you build a website with Claude today?
You build your website with Claude by describing the site in a prompt, letting Claude Code write your files, pushing them to GitHub and connecting your host, like Vercel or Cloudflare Pages. One beginner tutorial lists the toolbox: a Claude Pro subscription, a Vercel account and a GitHub account. You also need Node.js (LTS) and a terminal, and you install Claude Code through npm. Build small, and your $20 Pro plan works, Ryan Doser says. But for heavy build tasks, he recommends the $100 Max plan.
You have other routes. Chris Lema's 24-step guide has you build your free site with Claude Cowork, GitHub and Cloudflare Pages, and you need no terminal. Claude Design lets you start with visuals instead. Anthropic says you get designs, prototypes, slides and one-pagers, and that if you are a marketer, you can build landing pages in it. Whatever you pick, a host serves your pages to visitors. And those pages are what a public read scores.
What is the difference between Claude Design, Claude Code and the chat?
You use Claude Design for visual work, Claude Code to write and push the files that become your site, and the chat for plain conversation. Subscribe to Claude Pro, Max, Team or Enterprise, and you can try Claude Design, an Anthropic Labs product powered by Claude Opus 4.7, in research preview. It draws on your subscription limits. On your Enterprise plan, your admin has to switch it on in Organization settings (so if you can't find it, ask there first).
Use all three on your site, and watch the handoff between them. Anthropic says product managers can sketch a feature flow in Claude Design and hand it to Claude Code for implementation. For the steps, start with Claude Academy's 15-minute guide to that handoff. You can export your design decks as PPTX or send them to Canva. But none of that tells you what your deployed page contains. What ends up there depends on your files, your host and your prompt, so your live URL is the only honest witness. Check yours, not the tool. Read what your browser receives.
What does a Claude-built site ship by default?
Your Claude-built site ships whatever your prompt, your framework and your build settings put in the output. You cannot quote one default. Leon Furze's first prompt asked for an Astro site, and his Cloudflare Pages settings use the Astro preset, the build command `npm run build` and the output directory `dist`. Compare your settings with his: do yours match? Pelcro describes a live site with no traditional server, where pages are built ahead of time and served by Cloudflare. Does your site work that way? Whatever the answer, the deploy is the truth, so check what you deploy.
Look at your markup. HashBuilds says Claude writes clean, semantic code but needs guidance on viewport meta tags, charset declarations and title hierarchy, so check your head for all three. It can produce JSON-LD schema when you give it specific business requirements. Did you? The assessment reads your homepage and up to eight sampled pages, and you see your headers, markup, structured data, headings, script weight and image weight. Your prompts and your repository stay out of reach. So what you get is a read of what your site ships, nothing more.
What should you check before you publish a Claude-built site?
Before you publish, check your metadata on more than one page, then your canonicals, response headers, structured data, crawler access and image weight. Changed a shared template? Open at least two of your routes for each one. Screpy warns you that your metadata helper can look correct in isolation while returning the same values on every page, so test each route you touched. In one plugin demo, a newly deployed About URL returned a 404 and had to be repaired. And a page that fails quietly is the worst kind, so after you deploy, load your own new URLs.
Check your schema twice. eSEOspace recommends you parse every JSON-LD block, sample your templates in Google's Rich Results Test, and confirm your markup matches what your visitors see. It cites Google's rule: "Don't mark up content that is not visible to readers of the page." Screpy tells you a clean parse does not prove your pages are eligible, and valid markup can carry invented or outdated values. Read yours before you ship. And Pelcro flags anything over 500 KB, so weigh your images.
How do you deploy and keep a Claude-built site?
You deploy by connecting your GitHub repository to a host such as Vercel or Cloudflare Pages. Push changes to your repository, and your host redeploys. In one beginner tutorial, Claude Code asks whether to push each of your updates to GitHub, and Vercel redeploys your site automatically. Ryan Doser tells you Cloudflare deploys in under a minute. He keeps repositories private for real client sites (a sensible habit). Keep yours private too.
Your site lives at a pages.dev address until you point your own domain at it, Chris Lema tells you. Once you do, MindStudio says your DNS changes usually propagate within a few minutes to a few hours. Vercel provisions HTTPS on your custom domain automatically. Pelcro opens each of your changes as a pull request with a preview link you can open, and runs checks first. So should your required check fail, Cloudflare is never told to deploy your build.
Where do Claude-built sites score well, and where do they drop?
None of the sources publish scores, so they can't give you a general answer. Your own pillar results are your answer. You get one score out of 100 for your site, built on eight weighted pillars: architecture 15%, performance 15%, security 12%, search 12%, schema 10%, AI readiness 10%, editorial experience 10% and operational health 16%. Behind each pillar sits the evidence, so you can check it before you trust your score. A number without its evidence is just an opinion.
You can see the drops the sources name on your own pages. Your metadata may repeat across pages. Your new route may return a 404, and your images may run over 500 KB. Your sampled public pages show you all three. The read cannot tell you whether a schema value is true for your business, so you judge that. Your score lands in a band: 80 to 100 good standing, 60 to 79 constrained, 40 to 59 at risk, and 0 to 39 legacy condition. Find your band. Then fix what you can see on your site.
How do you check your Claude-built site in two minutes?
Paste your site's address into the free assessment, enter your work email so your report link arrives once, and about two minutes later you see your score out of 100. The assessment reads your homepage plus up to eight sampled pages from your site: your response headers, markup, structured data, headings, script and image weight, and your robots and canonical signals. You get one score. And you get the evidence behind each pillar of it. Before you change anything, you see what lowered it, so you never guess which fix to try first.
I use Claude Code daily and built this assessment with it, so the checks come from the same habit this page recommends: judge the output, not the tool. The assessment does not log in to your site, read your analytics, run field performance data or interview anyone. You get three checks per hour per connection. Fix one pillar, redeploy your site, and run it again. Your next score then shows you what your change did, and that is the only feedback worth trusting.
Questions, answered.
01Can Claude build a whole website by itself?
Claude Code can write the files and push them to GitHub, and in the sources a host such as Vercel or Cloudflare Pages then builds and serves them. You still decide the prompt, the domain and the checks. And metadata, schema and image weight need your review before the site goes live.
02Do you need a paid Claude plan to build a website?
Claude Code needs a paid plan, Ryan Doser says. The $20 Pro plan includes Opus and works for small builds. But he recommends the $100 Max plan, because heavy build tasks use up Pro's allowance fast. Claude Design is in research preview for Pro, Max, Team and Enterprise subscribers.
03Does valid JSON-LD mean you will get rich results?
No. Screpy says a successful parse does not establish rich-result eligibility. A valid object can still hold invented or outdated values, and valid markup does not guarantee a rich result. So sample each template in Google's Rich Results Test. Confirm every marked-up value is visible on the page.
04What can a public read of your site not show?
A public read cannot log in, read your analytics, use field performance data or interview your team. Your prompts and repository are also out of view (those live behind the build, not in it). What it does show is what ships to a browser: headers, markup, structured data, headings, script and image weight, and robots and canonical signals.
Sources
Facts on this page come from the documents below, checked on 2026-10-25. Figures about the assessment itself describe how it works.
- Introducing Claude Design by Anthropic Labs \ Anthropic
- Using Claude Design for prototypes and UX · Claude Academy
- Claude for Creative Work \ Anthropic
- The Beginner's Guide to Building a Website With Claude Code & Vibes
- Watch Me Build a Website With Claude In 30 Minutes
- How to Deploy a Website Built with Claude Design to Vercel | MindStudio
- Building Websites with Claude Code
- The Non-Developer's Guide to Building a Website with Claude | Chris Lema
More from What Ships.
Two minutes from now
