An agent-ready website: built for the visitor that is software.
In 2026 the thing filling out your quote form may be an AI agent working for a customer. Chrome now ships a Lighthouse audit for it. We build sites that pass, and the surprising part is how much of the work is accessibility and stable layout rather than anything exotic.
<!-- an agent can act on this --> <form aria-label="Request a quote"> <label for="svc">Service needed</label> <select id="svc" name="service" required>...</select> <button type="submit">Send quote request</button> </form>
An agent-ready website is one that AI agents can read, navigate and act on without a human: every interactive element has a programmatic name, the layout does not shift while the agent is working, forms are labeled and declarative, business facts are machine-readable, and, optionally, key actions are exposed as tools through WebMCP. Chrome added an “Agentic browsing” category to Lighthouse in June 2026 that audits exactly these things. DenverSEO.io builds and retrofits sites to pass it.
Google’s own AI guidance now says agents “may access your website” and describes them analyzing the rendered page and inspecting the DOM to complete tasks. Chrome’s agent-ready toolkit, published June 22, 2026, turns that into a checklist with a score. Perplexity’s Comet browser has been free to everyone since October 2025, OpenAI folded its browser work into the ChatGPT desktop app and a Chrome extension in August 2026, and Google announced Gemini agents in Chrome at I/O. The traffic is small today. The sites that are ready when it is not small will take the bookings.
This page is what the audit checks, what we change on a real site, and what we deliberately do not sell.
What Chrome’s Agentic Browsing audit checks
Lighthouse’s new category (Chrome M150) is informational, meaning it reports rather than weights a score, but its audits are concrete. In plain terms:
- Programmatic names on interactive elements. Every button, link, input and control needs an accessible name an agent can read from the accessibility tree. An icon-only button with no label is invisible to an agent in the same way it is to a screen reader.
- Layout stability. Cumulative Layout Shift is an agent problem as much as a user problem: an agent that located a button and then clicks after the page moved clicks the wrong thing. The audit reuses the CLS measurement.
- Declarative forms. Labels tied to inputs, meaningful names, required fields marked, errors announced. Our earlier note on agent-ready forms is the long version.
- Structured data. Organization, product, service and offer facts in schema.org markup, so an agent does not have to infer your hours from a paragraph.
- WebMCP tools (where present). WebMCP, in origin trial from Chrome 149, lets a site declare actions (“book an appointment,” “get a quote”) as callable tools with typed inputs. The audit checks whether declared tools are well-formed.
- llms.txt, only for errors. Chrome’s documentation calls it an “emerging convention” and the audit only fails if the file returns a server error. A 404 is not applicable. That tells you how much weight to give it.
Read that list again and notice what it is: WCAG-grade accessibility, Core Web Vitals, structured data, and one new protocol. Three of the five we have been doing for years.
What we change on a real site
| Layer | What we do | Verified by |
|---|---|---|
| Accessibility tree | Name every control, fix heading order, make focus visible, replace div-buttons with buttons, add landmarks. | axe scan at 0 violations; Lighthouse accessibility 100 |
| Layout stability | Reserve space for images, ads, seals and late-loading widgets; no content injected above the fold after paint; font fallbacks metric-matched. | CLS under 0.1 in the field (CrUX) and lab |
| Forms | Explicit labels, autocomplete tokens, server-side validation with clear inline errors, honeypot and timing traps that do not block legitimate agents. | Agent dry run: an automated browser completes the form end to end |
| Facts | One schema @graph per site: Organization, LocalBusiness or Service, Offer, OpeningHours, areaServed; the same facts repeated in visible HTML. | Schema validator; Rich Results Test; our schema markup validator output |
| Actions | Where it earns its keep, a WebMCP tool for the one or two actions that matter (quote request, booking), tested in the Chrome origin trial. | Lighthouse Agentic Browsing audit; manual call from an agent harness |
| Readability | Answers in the initial HTML, question-shaped headings, no critical content behind JavaScript. | Machine-readability analyzer; View Source with scripts disabled |
Every site we hand-code already ships this way. For WordPress and Shopify we retrofit it theme by theme; the work is unglamorous and it is mostly templates and labels.
What we do not sell
- “AI agent optimization” as a separate monthly line. It is accessibility, Web Vitals and structured data. If those are in your technical SEO engagement, this is in it too.
- WebMCP on every site. It is in origin trial and most small-business sites have one action worth exposing, if that. We add it where a booking or quote flow exists and the business can act on agent-originated requests.
- A guarantee that agents will use your site. Nobody controls which site an agent picks. What we can guarantee is that when one arrives, it can complete the task.
If you want the business-owner version of this (what an agent-ready site means for a Denver company and what it costs), the design side covers it on AI-ready websites; the agency side lists the surfaces on AI agents for business.
Run the audit yourself first
Open Chrome 150 or later, open DevTools, go to Lighthouse, tick “Agentic browsing” and run it on your quote or contact page. Then run our machine-readability analyzer on the same URL. If both come back clean, you are ahead of nearly everyone. If not, the fix list is the table above, and the free technical review will put it in order for your site.
Questions, answered straight
What is an agent-ready website?
A website that AI agents (in Chrome, ChatGPT, Gemini, Perplexity’s Comet and similar) can read, navigate and act on without a human: named controls, stable layout, labeled declarative forms, machine-readable facts and, optionally, actions exposed through WebMCP.
What does Lighthouse’s Agentic Browsing audit check?
Accessible names on interactive elements, layout stability (CLS), declarative form quality, structured data, well-formed WebMCP tools where declared, and whether llms.txt returns a server error. It shipped in Chrome M150 as an informational category.
Do I need WebMCP?
Only if your site has an action an agent should complete, such as a booking or quote request, and your business can respond to agent-originated requests. It is in origin trial from Chrome 149. Most sites should fix accessibility, layout shift and forms first.
Is an agent-ready site the same as an accessible site?
Largely, yes. Agents read the same accessibility tree screen readers use, so an accessible site is most of the way there. Add stable layout, machine-readable facts and content served in the initial HTML.
How much agent traffic is there today?
Little. SE Ranking measured AI referrals at 0.32% of site traffic in early 2026 and agent-originated actions are a fraction of that. The work is cheap when done inside a normal build and expensive to retrofit under pressure later.
Sources we cite on this page
- Chrome Developers: agent-ready website toolkit (June 22, 2026)
- Chrome Developers: Chrome at I/O 2026 (WebMCP origin trial, Gemini in Chrome)
- Chrome Developers: Lighthouse Agentic Browsing, llms.txt audit
- Google Search Central: AI features and your website (agents section)
- Perplexity: Comet available to everyone (October 2025)
- eWeek: OpenAI retires Atlas, moves browsing into ChatGPT app (2026)
- SE Ranking: AI traffic research study (June 2026)
Let's make your site the one they choose.
Fifteen minutes on the phone usually tells us whether we can help, and how much.