About This Talk

Slides: Your Website Is Not Just Your Website Anymore, Now What? (Google Slides)

WordCamp Portugal 2026 in Porto was my third WordCamp on Portuguese ground. I spoke at WordCamp Lisbon in 2023 about front-end performance, joined the WordCamp Porto 2024 organizing team the following year, and returned to Porto in 2026 to give this keynote. Different role each time, same community, same room of people who actually ship WordPress for a living.

The Porto session picked up the thread from the Lisbon talk three years earlier (what your server actually sends, and what visitors actually parse) and pointed it at the visitor most WordPress sites still ignore: the one without eyes. WordPress already won the open web. The question that mattered for this room is not whether WordPress can serve machines well, because it absolutely can. The question is whether the sites you ship actually do.

The visitor you have not been building for

The opening receipts came from Cloudflare Radar. Non-human traffic is now 51% to 53% of all internet traffic, and it is growing roughly 8x faster than human traffic. The crawler that used to be 90% of bot requests is down to 30%. Everyone else (every model, every agent, every retrieval system) is the other 70%. Agent traffic specifically sits at about 5.5% today.

Mobile traffic was 6.5% in October 2011, when the canonical mobile-first design book came out. The lesson back then was not "go mobile because half the web is mobile". It was "go mobile because the most constrained visitor forces the cleanest decisions, and the rest benefit". The number today is essentially the same, and so is the call.

To make the constraint concrete I demoed an AI assistant trying to buy a specific running shoe in a specific country in a specific size. The product page looks normal in a browser. With JavaScript off, which is what most agents see, the size tiles render as empty rectangles, the buy button is grey, the specs and reviews and delivery info are gone. The agent identifies the product. It cannot finish the task. The retailer is not broken. The retailer is unfinished.

I framed every agent interaction as four verbs: Consume. Analyze. Cite. Use. How far down that ladder any given agent goes is anyone's guess. The prep work is the same regardless: semantic HTML, structured data, accessibility, and an identity that is consistent across every place machines might look you up. WordPress shipped most of these primitives a decade ago. The work is using them on purpose.

What "machine-first" means for a WordPress site

The framework I unpack in full at machinefirstarchitecture.com has four sequential pillars. I walked the room through the WordPress-shaped version of each.

Identity is the one nobody owns. Machines learn who you are from roughly 30 places, and your site is only one of them. The other 29 are review sites, business directories, professional networks, podcast appearances, founder bios on portfolio sites, every directory you forgot you were in. The model synthesises all of it. If five of those sources disagree about your name, your URL, or what you do, the synthesis goes sideways and you lose the citation. The WordPress-specific lever is the sameAs array on your Organization and Person schema. It is the one place on your own site where you tell the machines, in their language, which of those 30 sources are actually you.

Structure is the one WordPress is best at and most plugins quietly undermine. JSON-LD is on more than 40% of pages now, and the pages that use it carry roughly six times more machine-readable detail than they did a decade ago. The defect I see most often on WordPress sites is structure that loads after the script does: prices, sizes, availability, reviews, all rendered by a block or a plugin that the agent will not wait for. The fix is unglamorous. Put the critical data into the initial HTML response, server-side, before any client-side anything runs.

Content is the one where the lazy writing penalty is now visible. Roughly 44% of AI citations come from the first 30% of a page. Adding real, attributed statistics improves citation rate by about 41%. Pages with explicit provenance, claim-level freshness, and self-contained sections survive being extracted; pages that buried the citable claim under three intro paragraphs do not. The single move worth taking back to the office is to rewrite the first 200 words of your five most important pages so the citable sentence is in the first one or two.

Interaction is the new one. Visibility and accessibility are solved problems; what is not solved is whether an agent that arrives with a task and a budget can complete it. WebMCP lets a site declare which actions an agent can take, what inputs they need, and what shape the response will be. The MCP ecosystem went from spec to widespread adoption in roughly a year. Major commerce platforms are exposing MCP servers in public. Agentic-commerce checkouts are running inside chat interfaces today. For WordPress, the near-term ask is small: WooCommerce stores and form-driven plugins are the first places where declaring machine-callable actions will start paying back.

The dependency runs all the way down. If Identity is broken, no amount of Structure rescues it. If Structure is broken, Content cannot be extracted. If Content is broken, Interaction has nothing to act on. None of this is a redesign. Most of it is one schema field, one initial-HTML check, or one afternoon spent reconciling what five sites say about you.

The shift is additive, not a replacement

I closed on the part that gets misread. Nothing about this kills SEO, content marketing, paid, email, or social. Those channels still work. The new lane sits next to them. What changes is that every existing lane now passes through a layer of machine interpretation: search results shaped by AI summaries, ads sitting next to AI recommendations, support tickets filtered by agents before a human ever sees them, content read by a model before it is read by a person. Strengthen the foundation, keep the existing playbook, and start showing up in the new lane.

AI is in the business of picking winners. For WordPress sites, the foundation is mostly there. The decision is whether you use it.

What I told the room to do this week

• Audit your sameAs graph. List the five most authoritative third-party pages that describe you or your business. Make sure your Organization or Person schema names them, and make sure those pages name you back where possible.
• Open your most important template with JavaScript off in DevTools. If price, availability, headings, or primary content disappear, that is what the agent sees too.
• Rewrite the first 200 words of your five most important pages so the citable claim is in the first sentence or two.
• Add or fix one piece of Article, Product, Organization, or Person JSON-LD this week. One change, validated, in production.

Related

• Machine-First Architecture — the canonical framework write-up
• Optimising in the AI Era: We Already Have a Universal Language — the closing keynote at Experimentation Elite Space Academy a week earlier, with the longer worked examples (including the AI-as-distribution story) and the full timeline of the 2025 AI-browser shift