If you're searching for "how to build a custom AI website" or "business custom website working with Claude", you're asking the right question at the right time. The honest answer is shorter than most blog posts about it: you need a real codebase you own, hosted on infrastructure you control, with one canonical brand-context file your AI tools can read. Everything else — the SaaS subscriptions, the AI website builders, the page generators — is fighting against where the actual workflow is going.
This guide walks through what a Claude-ready custom website is, why it matters in 2026, and the exact architecture that makes it work. It's based on Velocity X, the Astro template I (Aiden Wood) ship for clients — but the principles apply whether you build it yourself or hire someone to build it for you.
What Is a Custom AI Website?
A custom AI website is a business website whose codebase, content, and brand identity are structured so that AI coding assistants like Claude, Cursor, and Copilot can read, modify, and extend it without guessing. The site is yours — not rented from a SaaS platform — and every brand-affecting input lives in machine-readable files that AI models can pull into their context window on demand.
The shorthand definition: a custom AI website is a website where you can paste "add a new pricing tier called Enterprise" into Claude and get a working, on-brand pull request in under a minute. If you can't do that today, your website is not AI-ready.
Why This Matters in 2026
Three things changed in the last 18 months that broke the old "use a SaaS website builder" advice.
First, AI coding assistants stopped being autocomplete and became collaborators. Claude Code, Cursor, and Aider can take a written brief, read your codebase, plan an implementation, write the code, run tests, and ship a pull request. The bottleneck stopped being how fast a developer can type and became how clearly the AI can understand your codebase and your brand.
Second, the cost of running a small, well-architected custom site collapsed. Netlify, Cloudflare Pages, and Vercel all offer generous free tiers with global CDNs, automatic CI/CD, and one-click rollbacks. The fixed overhead of hosting a custom site went from "$50/month and a dev-ops headache" to "free and a git push". The economics that justified SaaS website lock-in have evaporated for most small businesses.
Third, the moat moved up the stack. Five years ago, the differentiator was design. Three years ago, it was performance. Today, it's how fast you can iterate. A business with a custom AI website can ship a new landing page, pricing tier, or campaign by the end of an afternoon, working with their AI in plain English. A business locked into a SaaS template can wait three days for a support ticket.
The Architecture That Makes It Work
An AI-ready custom website has four pieces. Each one matters because each one is something Claude needs to do its job.
1. Your Own GitHub Repository
The codebase has to live in a git repository that you own. Not "a copy of your site" — the actual production source of truth, in your GitHub organisation, that deploys to your domain. This is non-negotiable because AI coding assistants work on local files in a git context. Claude Code needs to clone your repo, read every file, plan against the whole tree, and commit changes that flow through normal code review and CI.
If your website lives inside a SaaS dashboard — Webflow, Squarespace, Shopify's theme system, WordPress's plugin sprawl — Claude can't see it. The AI is locked outside the wall. You can describe what you want, but Claude can't read the rendered DOM, can't see the build pipeline, can't run your tests, can't ship a PR for you to review. The repo isn't a nice-to-have; it's the precondition for the entire AI workflow.
For Velocity X, every paid customer gets the GitHub repository transferred to their account. Not a license, not a "managed instance" — the actual codebase, theirs to fork, modify, hand to a different developer, or feed into Claude.
2. A Single Brand Context File (brand.json)
The second piece is the part most people miss. AI assistants are pattern-matchers operating on the text in their context window. If your brand information is scattered across twenty components, three CMS dashboards, and a Google Doc somewhere, Claude has to guess. Guessing produces drift. Drift produces a website where the homepage says "Velocity X is the buy-once route planner CRM" and the pricing page says "Velocity X is the leading marketing platform". Two months of AI-assisted edits and your brand voice is everywhere and nowhere.
The fix is a canonical brand context file. In Velocity X it's brand.json. It contains the business name, tagline, voice rules, colour palette, pricing tiers, audience descriptors, the Stripe URLs, every social link, and the value proposition in three forms (one-liner, paragraph, long). Every component on the site reads from this file. Every AI assistant working on the site reads from this file. When you tell Claude "update the homepage tagline", Claude reads brand.json, follows the voice rules defined there, and produces copy that matches the rest of the site.
This pattern has a name in the AI engineering community: canonical context. It's the same idea as having a single source of truth in your database, except the source of truth is for the model that's working on your business.
3. JSON-Driven Page Templates
Pages on a Claude-ready website are templates, not hand-coded HTML. The template defines the layout — hero, three-column features, testimonial slider, CTA — and the content comes from JSON files. To add a new service page, you drop in a new JSON file with the title, hero image, three feature blocks, and the URL slug. The template renders the page. No new Astro components, no new routes to wire up, no AI-induced regressions because nothing else changed.
This is the architecture that lets you say "add a page for our new commercial cleaning service" and have Claude scaffold the whole thing from your brief. The template is fixed and tested; the AI just fills in the variables. Risk of breaking your site approaches zero, because the AI is operating inside a sandbox of pre-tested patterns.
4. CI/CD Pipeline with Branch Previews
The fourth piece is automatic deployment. Every push to a branch in your GitHub repo builds and deploys to a unique preview URL. Every merge to main ships to production. Rollbacks are one click. This isn't a nice-to-have either — it's what lets you trust the AI workflow.
Here's why: when you ask Claude to "redesign the pricing page", you don't want to wait until it's deployed to production to see if it looks right. You want the AI to push to a branch, the CI to build a preview, and you to click a link and see the new pricing page on a real URL with real data. If it's wrong, you reject the PR. If it's right, you merge it. The whole loop takes under three minutes.
Netlify, Vercel, and Cloudflare Pages all do this natively. Connect your GitHub repo, pick a branch, done. No infrastructure team required.
The Anti-Pattern: AI Website Builders
You'll see a lot of "AI website builders" advertised in 2026. The pitch is seductive — type a description, get a website. The reality is that most of them are SaaS lock-in dressed in AI clothing. The generated website lives inside their dashboard, not in your repo. You can't take Claude Code and modify it. You can't add a custom integration that's specific to your business. You can't even see the source HTML in a meaningful way.
The test for whether an "AI website builder" is actually AI-ready: can you clone the source code to your laptop and ask Claude Code to add a new feature? If the answer is "no, the code is on our platform", it's not an AI website. It's a website built by an AI, hosted on a closed platform, that you can't iterate on with AI.
The difference is the same as the difference between a Lego set you assemble (you can take it apart and rebuild it) and a plastic model glued together at the factory (you got what you got).
How to Get Started Today
Three paths, in increasing order of effort.
Path 1: Buy a buy-once template. Velocity X is one option in this category — you get the Astro codebase, brand.json, JSON-driven page templates, Netlify CI/CD config, and the GitHub repo transferred to your account, for a single payment. From there, you (or your AI) can iterate however you want. No subscription, no lock-in. See pricing.
Path 2: Hire a developer who works the AI-native way. Look for someone who works in Claude Code or Cursor, ships to Netlify or Vercel, and structures their codebases around a single brand context file. They'll deliver something you can keep iterating on with AI after they're done. Ask them specifically: "Will I get the GitHub repo, and can I run Claude Code on it?" If they hedge, find someone else.
Path 3: Build it yourself with Claude Code. If you're technical, the stack is Astro + React + Tailwind + Supabase + Netlify. Start with an Astro starter, define your brand.json, write JSON files for your pages, deploy to Netlify. Spend a weekend with Claude Code and you'll have a custom AI website. It's not magic — it's a stack that's been mature for two years and a workflow that's been mature for one.
Frequently Asked Questions
Can I use Claude to build a website for my business?
Yes. Claude can scaffold and iterate on a complete business website provided it has access to a real codebase. The recommended setup is Claude Code running locally on an Astro or Next.js project with a brand context file (such as brand.json). Tell Claude what you want in plain English and review the resulting pull requests. For non-technical users, hire a developer who works in this stack and have them set you up — then you can keep using Claude yourself for content edits.
What is the best framework for an AI-ready website?
Astro is the strongest choice for content-driven business websites in 2026. It compiles to static HTML by default (fast and cheap to host), supports React/Vue/Svelte components when you need interactivity, has excellent SEO defaults, and its component model is easy for AI to reason about. Next.js is a fine alternative if your site is more application than content. Avoid WordPress, Webflow, and Squarespace for AI-native workflows.
How much should a custom AI website cost?
A buy-once template like Velocity X starts at $4,995 AUD for the small package (32 pages, dashboard included, GitHub repo transferred). Custom builds with a developer typically run $5,000 to $50,000 AUD depending on scope. Ongoing iteration with Claude or Cursor is effectively free once the codebase exists — you're paying for AI tokens, not developer hours.
Will Claude break my website?
Not if your project has automatic deploy previews. Every change Claude makes goes through a branch and a preview URL before it lands in production. You review the visual result before merging. If something looks wrong, you reject the PR. The risk model is the same as code review with any human collaborator.
Do I need to know how to code?
For day-to-day content edits and adding pages — no. With a JSON-driven template and Claude Code set up, you can describe changes in plain English. For the initial setup (buying a template, transferring the repo, connecting to Netlify, configuring environment variables) — either you spend a weekend learning the basics, or you hire someone to do the one-time setup for you.
Why not just use Webflow with their AI features?
Webflow's AI features operate inside their dashboard. The site can't be cloned to your laptop and edited with Claude Code. You're paying monthly to rent a website you can't fully control. For some non-technical users this trade-off is fine, but it disqualifies the setup from being "AI-ready" in any deep sense. If Webflow ever raises prices, deprecates features, or shuts down, you have limited recourse. A custom AI website on your own GitHub repo is forever.
The Bottom Line
A custom AI website in 2026 is a business website that you own, built on an open codebase, with a canonical brand context file your AI tools can read. The technology is mature. The cost is lower than ever. The workflow is faster than any SaaS website builder. The only thing standing between most businesses and an AI-ready custom website is the inertia of believing the SaaS pitch from five years ago is still the right answer.
It isn't. Buy the template, or hire the developer, or build it yourself. The businesses that get this right in 2026 will spend the rest of the decade out-shipping the ones that don't.