Most firms talking about “AI-ready websites” are asserting expertise. We would rather show it. Run Google Chrome’s Lighthouse audit against this site and you will see the same thing we see: a perfect or near-perfect score on every category Google measures, and a clean pass on Agentic Browsing, the new audit that checks whether an AI agent can actually read and operate a site.
Then we do the thing almost nobody selling this will do: tell you exactly what that score is worth, and what it is not. Passing the audit is the price of entry, not the prize.
The proof, and how to verify it yourself
This is not a claim you have to take on trust. Open Google’s PageSpeed Insights, enter our domain, and read the result. As of the latest run:
| Category | Desktop | Mobile |
|---|---|---|
| Performance | 100 | 98 |
| Accessibility | 100 | 100 |
| Best Practices | 100 | 100 |
| SEO | 100 | 100 |
| Agentic Browsing | 3 / 3 | 3 / 3 |
| Cumulative Layout Shift | 0 | 0 |
| Largest Contentful Paint | 0.5 s | 2.2 s |
Verified June 2026. Re-run it yourself, the numbers are public.
What each pass means once you think about machines, not humans
Accessibility 100 is the score that matters most for AI. An AI agent does not see a page the way a person does. It walks the accessibility tree, the structured, labelled model of the page that screen readers also use. That tree is the machine-eye view of a site. Ours scores 100: every interactive element has a programmatic name, every relationship is valid, nothing important is hidden from the agent reading it. Accessibility stopped being only an ethical obligation; it is now the primary data model AI agents rely on.
A Cumulative Layout Shift of zero means agents can trust what they see. When an agent identifies an element and then acts on it, a shifting layout moves the target. Zero shift means the page an agent reads is the page it can reliably interact with.
A clean, fast fetch lowers the cost of reading us. Agents fetch the live web at runtime, and markdown or stable HTML is roughly five times cheaper to parse than bloated pages. Sub-second desktop load and zero blocking time mean any agent already crawling us spends less to read us, and the agents already crawling the web are real: ClaudeBot, ChatGPT-User, PerplexityBot and Google-Extended fetch content and well-known files daily.
The 3/3 Agentic Browsing pass covers the deterministic checks that apply to a content site: a machine-readable summary at the domain root (our llms.txt, generated by a plugin we built and published on WordPress.org), a sound accessibility tree, and visual stability. The audit marks tool-exposure protocols like WebMCP as not-applicable for a site that does not expose tools, so we do not claim a WebMCP pass we have not earned. We claim exactly what the audit shows. The fuller breakdown of what the audit checks is in our Agentic Browsing audit guide.
What passing the audit actually buys you, and what it does not
Here is the part the green-tick vendors leave out. Being agent-ready is necessary. It is not sufficient. It does not, on its own, get you cited or recommended by an AI. To see why, separate the three surfaces where content meets a large language model, because almost every confused argument about AI visibility comes from collapsing them.
Surface one, the training data. How a model forms its standing knowledge of your brand, before any prompt. Labs filter vast web crawls through quality classifiers that upweight content the trusted web already references. No file on your own site touches this. The lever here is corroboration, being referenced by sources the filter trusts. We cover the mechanism in how AI learns your brand.
Surface two, runtime retrieval. What an AI fetches live when it answers a question. This is the only surface agent-readiness touches, and it touches it well. Clean structure, a curated llms.txt and cheap parsing make you legible to the agents that read these signals at the moment of answering.
Surface three, Google ranking and AI Overviews. Google has stated it ignores llms.txt for Search, including its AI features, the file will not help or hurt rankings. Google ranks on its own systems.
So when a tool promises an llms.txt will get you cited, hold it against those three surfaces: nothing for surface one, helpful for surface two, ignored on surface three. That is not a reason to skip the work, it is a reason to know precisely what it is for. We ship the tooling and the honesty.
Google’s own guidance contradicts itself, unless you know where to look
If the recent guidance has felt confusing, you are reading it correctly. Google Search says you do not need machine-readable files. Google Chrome shipped an audit that checks for one. Google Cloud released a markdown-bundle standard for handing content to agents. Three teams, three signals.
They only conflict if you treat “AI” as one thing. Each statement is correct about a different floor: Search is talking about ranking, Chrome about agent operation, Cloud about content portability. The audit measures the top floor, agent execution, and it cannot see the floors below it, which is where citation and recommendation are actually decided. The full resolution sits in our Four-Floor Model, and the canonical framework lives at citate.io.
The work the audit cannot measure, where citations are actually won
Lighthouse measures whether an agent can read you. It cannot measure whether an AI will choose you, and it certainly cannot measure whether you are recommended, which is a different outcome from being cited. Independent analysis of “best” results in AI Overviews found that a brand’s own self-promotional listicle frequently gets cited as a source while the recommendation goes to the established competitors named inside it. Citation and recommendation have decoupled, and recommendation is anchored to genuine authority, who the rest of the web talks about, not to how many times you have called yourself the best.
That is why we frame the goal the way we do: strong brands rank, get cited, get recommended, and dominate. Four separate states, won in order, and the commercial value sits in the last two. The audit gets you to the starting line. Corroboration, third-party references, consistent entity data, named-voice proof, the off-page signals a machine can verify, is what moves you from cited to recommended. It is invisible to Lighthouse and central to what we do. See the best-of problem and editorial selection.
Why this matters now, not in 2027
The reason any of this is urgent is the click. In the UK, the majority of Google searches now end without a click on anything, and the share that produces a website visit keeps falling. Visibility inside an AI answer is no longer the same thing as traffic to your site. The measure of success is no longer only “do we rank and get the click”, it is “are we present in the answer, cited and recommended, and is the authority underneath that presence real enough to survive the next algorithm shift.” Agent-readiness is the price of being eligible. Corroboration is what wins once you are in the room.
What we do
We make your site genuinely agent-ready, the demonstrable floor you can audit, exactly as we have done with our own. Then we do the harder, durable work the audit cannot see: building the entity foundation and off-page corroboration that move you from cited to recommended, measured against a published standard (CITATE) rather than our opinion. We build our own tooling, including the LLMs.txt Curator plugin on WordPress.org, so we are executing with instruments we understand to the line of code, not reselling someone else’s checklist.
