1. Home
  2. Services
  3. AI-Optimized Web Design and Development
Custom build

Websites built for two readers: the buyer and the model.

A template can give you a page. It cannot give you a page that a buyer converts on and that ChatGPT can quote accurately. We design and hand-build both at once, because in 2026 they are the same job.

Typical scope8 to 45 pages, or a full store
Timeline4 to 12 weeks
HandoverYour accounts, your hosting
IncludedLabelled data and a written AI rule
The problem

Why the template stopped working

  • Next.js
  • Astro
  • WordPress
  • Shopify Plus
  • Headless
  • Structured data
  • Core Web Vitals
  • WCAG 2.2 AA
  • Server-side rendering

AI crawlers now make up roughly 28% of the request volume Googlebot generates, and two of them alone, GPTBot and ClaudeBot, account for hundreds of millions of requests a month. Those crawlers do not run your JavaScript the way a browser does. If your headline, your prices and your proof are painted in by client-side code, a large part of what you paid for is invisible to the systems your customers now ask first.

At the same time the human side got harder, not easier. Buyers arrive later, already briefed by a model, and they arrive with a shortlist. The page has less time to make its case and a narrower question to answer.

So the brief changed. A site now has to render its substance on the server, describe itself in machine-readable terms, expose routes an agent can complete, and still be the most persuasive thing in the buyer’s tab. That is a build, not a theme.

Deliverables

What you get

Every build ships with the AI layer included. It is not an upsell, because a site without it is only half finished.

  • Conversion model first. We map the decision your visitor is making, then design the page order around it, before anyone draws a screen.
  • Custom design. Drawn for your brand, reviewed with you, not assembled from a marketplace theme.
  • Built by hand. No page-builder bloat, no forty plugins holding it together, and nothing jumping around as the page loads.
  • The words in the page itself. Every claim, price and proof point written into the page when it loads, rather than painted in afterwards by code that AI readers do not run.
  • Structure a machine can follow. One main heading, headings in a sensible order, and lists and tables built as lists and tables rather than styled to look like them.
  • Labelled data that connects. Your company, your services, your products, your people and your answers, each one labelled and linked to the others, so a machine reads one description of a business rather than a pile of unrelated facts.
  • A written rule for AI readers. GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot and Google-Extended are the programs that read your site on behalf of AI. You decide which are let in, and the decision is written down rather than left to chance.
  • Agent-completable routes. Forms, search and checkout that work without a mouse, plus feeds or an API where an agent needs one.
  • A Core Web Vitals budget. Agreed before build, enforced at build, measured after launch.
  • Accessibility to WCAG 2.2 AA. Keyboard operation, contrast, focus order, reduced motion.
  • Analytics and an event model. Named events that map to the conversion model, wired to GA4 and your CRM.
  • A CRO test plan. The first three experiments, written up, ready to run on launch day.
Method

How the build runs

The same sequence every time, so you always know what is happening this week and what you will see next.

  1. Decision map

    We interview you and, where possible, your customers, then write down the decision the page has to win and the objections in the way.

    Week 1
  2. Architecture

    Page structure, topics, headings and the single description of your business, all agreed before a pixel is drawn. This is the part that decides whether AI can describe you.

    Week 1 to 2
  3. Design

    Key templates first, then the rest of the system. Two review rounds are standard.

    Week 2 to 5
  4. Build

    The pages people see, the systems behind them, your content tools and every integration, all delivered into an account you own from day one.

    Week 4 to 10
  5. Instrument and QA

    Every browser, every device, Google’s page speed scores, accessibility and the labelled data. Then we read the site the way ChatGPT’s crawler reads it, to confirm what a model sees and not just what a browser shows.

    Week 9 to 11
  6. Launch and aftercare

    Deployed to your hosting with redirects, monitoring and a documented handover. Then 30 days of fixes at no charge.

    Week 12
Evidence

One we ran like this.

Client names and identifying details are changed. The figures are not.

Northrail Supply storefront with a promoted category and product grid
Ecommerce · AI-optimized rebuild

Northrail Supply

A 4,000 product industrial catalog moved off a legacy system. Trade pricing, bulk ordering and a checkout rebuilt around how buyers reorder, with every product written into the page itself and labelled item by item, so search engines and AI shopping assistants can both read the catalog.

+18%Conversion rate
2.4sLargest paint
9 wksDesign to live
See all work →
Questions

Asked before every engagement.

What does "AI-optimized" mean for a website?

Three concrete things. The words are written into the page when it loads, so a machine that does not run a browser’s code still sees them. Every fact on the page is labelled and linked into one description of the business, rather than scattered as unrelated snippets. And the site states, in writing, which AI readers are allowed in. Everything else marketed under that label is either ordinary technical SEO or decoration.

Will an AI-built website rank and convert better than a template?

Better than a template, in our experience, yes, but not because AI wrote it. The gain comes from a conversion model built before the design, content that renders server-side, and a clean structural layer. We use AI heavily inside the process, for research, draft copy, variant generation and QA, and we do not let it make the final call on a claim, a price or a design.

Do you rebuild on WordPress and Shopify, or only on modern frameworks?

Both. WordPress and Shopify are the right answer for most content and commerce teams, and both can be made fast and fully readable by machines. Next.js or Astro is the right answer when the site is an application, needs a component system, or has a content model your marketing team will change weekly. We pick by what you have to run afterwards, not by what is fashionable.

Does llms.txt do anything?

Not much on its own, and we will say so. As of early 2026 no major AI provider has publicly committed to reading llms.txt in production, and Google has said outright that it does not support it. We ship it because it costs an hour and a handful of providers do fetch it. It is insurance, not a strategy. The work that moves citations is entity clarity, structured data, quotable content and third-party presence.

Who owns the code?

You do, from the first day. We work inside your own account where you have one, and set one up in your name where you do not. There is no license, no lock-in and no hosting you are obliged to buy from us.

Find out where AI sends your customers today.

We run your brand through the answer engines your customers use, then send you the transcript of what they say about you. Free, and yours to keep whether or not we work together.

Related

Usually bought alongside this.

  • GEO, AEO & LLMO

    Generative Engine Optimization, Answer Engine Optimization and Large Language Model Optimization, run as one program across six engines.

  • AI Analytics & Growth

    Tracking you can trust, attribution you can defend, forecasting you can plan against, and share of AI answers on the same dashboard as everything else.

  • AI SEO & Content Engineering

    Technical SEO, topical authority and a content engine where AI does the research and the drafting, and a human owns every claim that ships.