AI-First Websites
Machine‑Readable And
Human‑Persuasive Are
The Same Problem.
Human‑Persuasive Are
The Same Problem.
The structure that lets a model parse your site cleanly is the same structure that lets a person understand it in eleven seconds. Almost nobody builds for both, because search people do not design and designers do not think about crawlers.
The human visitor
An AI-Referred Buyer Is A Different Animal
They already know what you do, because a model explained it. They have already been handed a shortlist, so they are comparing rather than discovering.
What they came for is a trust decision. Is this company real, is it any good, and would I be comfortable putting it in front of my boss?
That gets answered visually, quickly, and mostly before anyone reads a full sentence.
The other visitor
4 in 5 websites still give those agents no instructions at all, and plenty serve a page that only renders after JavaScript runs. A page an agent cannot read does not exist to it.
AI agents now request pages at close to the rate humans search, and the share is still climbing. BrightEdge, April 2026
7% vs 5%
AI-referred visitors convert better than Google-referred ones, with roughly twice the time on site. They arrive informed, so the site has less to explain and more to prove. Similarweb, US desktop, Sept 2025
What we build
The New Standard for Company Websites
Structure A Model Can Read
Semantic HTML that means something. Schema that is accurate rather than decorative. Clear entity relationships: what you are, who you serve, what you are comparable to. Content in the page source, not assembled after the fact.
Design That Carries Weight
Typography, hierarchy, restraint, motion where it earns its place. A company that looks its size loses shortlists to competitors who look larger. The least discussed and most decisive part of the work.
Speed And Render Safety
Server-side rendering, Core Web Vitals in the green, and a page that survives being fetched by something that is not a browser.
Conversion For Who You Now Get
Answer-first pages. Proof above the fold. Comparison content that does not dodge the question. Paths built for someone arriving informed rather than curious.
The build standard
What we build to,
measured.
Every site we ship is tested against these before launch and again 30 days after. The thresholds are the real ones, not rounded up for a brochure.
CHECKS · PRE-LAUNCH AND AT 30 DAYS
Renders for agents 9 Checks
- Primary content present in the initial HTML response.
- Page readable with JavaScript execution disabled.
- Nothing critical behind a click: tabs, accordions, modals.
- TTFB under 600ms for non-browser user agents.
- Redirect chains capped at 1 hop.
- No soft 404s returning 200.
- Consistent response whether or not a cookie is set.
- Content identical for agents and humans, no cloaking.
- Server log verification that agents received 200.
Core Web Vitals 7 thresholds
- LCP under 2.5s at p75, field data not lab.
- INP under 200ms at p75.
- CLS under 0.1 at p75.
- TTFB under 800ms at p75.
- Total transferred weight under 1.2MB on a landing page.
- Fonts preloaded, swap set, no invisible text flash.
- Images WebP or AVIF, sized, lazy below the fold.
Structure 8 Checks
- Exactly one H1 per page, no skipped heading levels.
- Real landmarks: header, nav, main, article, footer.
- Schema per template, validated against Rich Results.
- Descriptive anchor text, zero instances of learn more.
- Breadcrumbs on every page below the top level.
- Self-referencing canonical on every page.
- Internal links express relationships, not just navigation.
- URL structure lowercase, hyphenated, no parameters.
Accessibility, WCAG 2.1 AA 8 Checks
- Contrast 4.5:1 body text, 3:1 large text and UI.
- Every interactive element reachable by keyboard.
- Visible focus states, never outline:none.
- Alt text on every meaningful image.
- Labels bound to every form field.
- prefers-reduced-motion respected.
- Page usable at 200% zoom without horizontal scroll.
- Tested with a screen reader, not just an automated scan.
Deliverables
What you actually get
Discovery and design Week 1 To 5
Audit
Your current site against all 33 checks above, with a written pass or fail on each and what it costs to fix.
Architecture
Sitemap, URL structure, entity model, and the template list. Signed off before design begins.
Design
Full page designs for every unique template, desktop and mobile, in Figma. Not a theme with your logo on it.
Review
2 rounds of revisions included at design stage, where changes are cheap.
Prototype
Clickable, so you experience the flow before anyone writes code.
Build and launch Week 6 To 11
Build
Server-side rendered, from your approved designs. Staging environment you can access from day one.
Schema
Implemented per template and validated, not bolted on afterwards.
Content migration
Every existing URL mapped, redirected 1:1, and verified post-launch.
QA
Cross-browser, mobile, keyboard, screen reader, and a crawl diff against staging.
Launch
Monitored for 72 hours, with agent access verified in the server logs, not assumed.
Handover
Figma files, repo access, documentation, and a walkthrough recording for your team.
Yours to keep Permanently
The code
Your repository, your hosting account, no proprietary platform.
The designs
Editable source files, not flattened exports.
The domain and DNS
In your name, under your control, always.
No licence
Nothing stops working, expires, or requires a renewal to keep the site online.
Scope
Not everyone needs a new site
Often the architecture is sound and the problem is structure, speed, and three pages that do not exist yet. That is a four-week engagement, not a four-month one.
We tell you which situation you are in during the audit, including when the honest answer is to leave it alone and spend the money on citations instead.
How We Work
A Clear Path From First Audit to Launch
Sites we build are yours. No proprietary platform, and no licence you have to keep paying to keep the site online.
Common Questions
Frequently Asked Questions
What do you build on?
Usually a modern framework with server-side rendering, or WordPress configured properly. What matters is that the content exists in the page source before JavaScript runs. That rules out most drag-and-drop builders for a site that needs to be read by agents.
Which models do you cover?
ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity and Claude. Google’s AI surfaces matter most for volume, and ChatGPT sends the large majority of trackable standalone AI referral traffic.
Can you control what an AI says about my company?
Not control, no. You can change the inputs it draws from, which in practice changes the output. Anyone claiming control is selling something that does not exist.
What does it cost?
We quote after the free check, because until we have looked at your site we would be guessing. Engagements run on a 90-day starting term and we price the work, not a package.
Start Here
Find out whether your site is the problem
The free report includes whether agents can read your site at all, which is usually the first thing we find.
