Your Website Was Never a Brochure. You Just Built It for the Wrong Reader.

You did everything 2022 told you to do.

You migrated off the old brochure site. You added a funnel. You wired up a CRM, dropped in the trust badges, collected the testimonials, ran the SEO audit, maybe paid real money for a Webflow rebuild that looks genuinely good. You followed the playbook line by line.

And inquiries are still flat.

The instinct is to assume you executed badly. You didn’t. You executed a playbook written for a reader who is leaving the building.

I. The tell is in who each era was built for

Stop thinking about websites as technology stacks and start thinking about them as briefs. Every era of the web had an implied answer to one question — who is this page for? — and the whole industry organised itself around that answer without ever saying it out loud.

2010 — built for a human to read. Static HTML and CSS, or a CMS to manage it: Joomla, Drupal, Magento, WordPress. The site was a digital pamphlet. You arrived, you read, you maybe found a phone number. The job was to inform a person.

2015–2025 — built for a human to convert. This is the era most sites are still trapped in. Custom CMS builds, Webflow, the page-builder explosion, then the AI site generators. The craft got extraordinary — funnels, heat-mapped CTAs, social proof above the fold, trust-building sequences, technical SEO. But notice what every one of those techniques optimises: a human’s journey down a page. The scroll. The hesitation. The click. We spent a decade getting world-class at persuading a person who was reading.

2026 — built for an agent to act. And here the brief changes completely. The reader you spent fifteen years learning to persuade is being quietly replaced by something that doesn’t read, doesn’t feel your trust signals, and doesn’t travel your funnel.

II. The human reader already left. This part isn’t contrarian anymore.

I’ll move fast here because this is now consensus, not insight.

The discovery layer collapsed. Zero-click searches — where someone gets their answer without ever clicking through to a site — rose from 56% to 69% after Google’s AI Overviews rolled out. When an AI Overview appears, that figure hits 83%. In Google’s newer AI Mode, roughly 93% of searches end with no click to anyone’s website at all. Business Insider lost 55% of its search traffic between 2022 and 2025. Gartner called a 25% decline in traditional search traffic by 2026, and as of early 2026 it’s playing out.

If your strategy is “rank, get clicked, convert on the page,” the get clicked step is evaporating. You can keep perfecting the funnel. Fewer and fewer humans are walking into the top of it.

But this is the easy version of the argument, and it’s where most pieces stop — “so do AEO and GEO instead, get recommended by the AI instead of ranked by Google.” That advice isn’t wrong. It’s just still treating the agent as a fancier search engine that you want to be found by.

It misses what the agent actually does next.

III. The break: the agent doesn’t want to find you. It wants to transact.

Being recommended is the old goal wearing new clothes. The real change is that the machine on the other end doesn’t stop at “here’s a good option.” It acts.

This is no longer a forecast. It’s shipping.

OpenAI and Stripe released the Agentic Commerce Protocol (ACP) — an open standard for agent-initiated checkout — on 29 September 2025, launching it alongside ChatGPT’s Instant Checkout. Google followed with its own Universal Commerce Protocol (UCP), launched January 2026 with Shopify, Target and Wayfair, and backed by Visa and Mastercard. Underneath both sits the same idea as Anthropic’s Model Context Protocol: a standard way for an agent to read a catalogue, check inventory, pull a price, and complete a transaction.

Here is the part that should reframe everything for you. In that checkout flow, the user never leaves the conversation. There is no click-through to your website. The agent discovers, compares, selects, and pays — and you, the merchant, remain merchant of record while never being visited in any sense you’d recognise.

And here’s the detail that makes the point sharper, not weaker. The consumer-facing feature — Instant Checkout — was retired in March 2026, after only about a dozen merchants ever shipped against it. The shiny front-end experiment was premature. But the protocol underneath kept advancing: PayPal joined as a payment provider, Stripe shipped a full Agentic Commerce Suite, and the spec is still maintained and versioned on GitHub through 2026. That’s the tell. The storefront experiment stumbled; the machine-readable plumbing did not. The infrastructure gets laid whether or not any single front-end survives. McKinsey, in its October 2025 research, puts the agentic-commerce opportunity at $3–5 trillion in mediated global spend by 2030.

A fair objection, and I’ll be honest about it: this is moving fastest in retail, and even there the first consumer experiments are stumbling. For B2B services — someone’s agent autonomously shortlisting and engaging a software studio, say — we’re still in pilots and order-workflow automation, not autonomous hiring. But the direction is not ambiguous. The plumbing being laid for retail is the same plumbing every category inherits next. If you sell to businesses, you have a short, visible runway before the same mechanic arrives at your door. The question isn’t whether. It’s whether your site can be read and acted on when it does.

IV. The contradiction worth staring at

Here’s the tension that tells you who to believe.

Google’s own 2026 AI-search guidance says, essentially, nothing has changed — AEO and GEO are “still just SEO,” and you don’t need machine-readable files or special markup to show up. At the very same moment, the entire agentic-commerce stack — ACP, UCP, MCP — is being built on the premise that machine-readability is everything.

Both cannot be true. And the incentives explain the gap. The incumbent that owns the old discovery monopoly has every reason to tell you the ground isn’t moving. The companies building the agent layer are quietly re-plumbing the web for machines and telling you exactly what those machines need: clean, server-rendered, structured, accessible content — not div-soup behind a JavaScript wall, not copy locked inside a proprietary page builder’s runtime.

V. So what is a website for, now?

It is the part of your product that agents touch. That’s the entire spec.

Which means the job is no longer “design a beautiful journey for a person.” The job is “be legible and actionable to a machine.” Server-rendered, not hidden behind client-side JavaScript. Structured and consistent, so a model extracts the same facts every time. Fast and clean, because Core Web Vitals stopped being a vanity metric the moment a crawler’s patience became your distribution.

And here’s the quiet consequence almost nobody has internalised: the only way to build something that legible is to specify it precisely. You cannot hand-wave your way to a machine-readable site. Ambiguity that a human visitor would forgive — a vague heading, an inconsistent label, a fact stated three different ways across three pages — is exactly what breaks an agent. Precision stopped being craftsmanship and became function.

VI. Two things we built that make this concrete

I don’t want to leave act four as assertion, so here are two real artefacts.

specshop.dev was built on a Tuesday, in a couple of hours. It says so on the site. The point is not the speed — speed is the obvious read and the boring one. The point is what the speed frees. When a site takes hours instead of weeks, it stops being a capital-D Deliverable you agonise over and protect, and becomes a render of the spec. The thinking moved upstream. The website is now downstream of the spec, the way a compiled binary is downstream of source. You don’t fuss over the binary. You get the source right.

We migrated a 2020 WordPress site — over 350 pages — in three days. Off the bloat, no more lag, perfect PageSpeed scores. And in this thesis, a perfect Core Web Vitals score is not a brag about polish. It’s the visible symptom of legibility. The old 2020 build couldn’t be cleanly read or acted on by anything. The rebuilt one can. We didn’t “make it faster.” We made it machine-readable, and the speed score is just the part of that you can see on a dashboard.

Note what changed and what didn’t. WordPress is back — but not the Elementor-and-forty-plugins WordPress that gave the platform its bad-SEO reputation. WordPress 7 shipped with an Abilities API; headless WordPress with a clean modern frontend is now the default enterprise pattern. Built from the ground up with AI tooling, in hours, it’s one of the most agent-legible things you can ship. The tool came full circle. What changed is who we built it for.

VII. The thesis, stated flat

Your website was never really a brochure, and it was never really a funnel. It was always the surface of your product — the part the outside world touches.

For fifteen years we optimised that surface, beautifully, for the wrong reader. The cost of that era was never measured in dollars or in build time. The cost was building exquisitely for an audience that was on its way out the door.

The reader changed. The brief didn’t. Update the brief.