Latest · Experimental · Published 11 Sep 2026 · GA Applications editorial

If a person can use it, an agent probably can too.

AI readiness is not a badge, a plugin or a hidden file of text for bots. Google’s own guidance for AI features requires the same fundamentals as search: crawlable, clear, people-first content. The practical test is parity — whatever a human can understand and do on your site, a machine should be able to understand and do: read the meaning, see the evidence, complete the action, confirm the result.

Run the parity test Get a parity reading

Google’s AI guidance and the WebMCP talk are cited with dates. WebMCP is emerging; no discovery guarantee exists or is implied.

Orientation

What “AI-ready” actually means

Two machine audiences now read your site. The first is here: search AI features that summarise and cite. Google’s 2026 guidance for those features is almost disappointingly plain — no special markup, no new file format, the same crawlable, people-first fundamentals as search.

The second audience is arriving: agents that act on a user’s behalf — find, compare, enquire, book. Tara Agyemang’s AI Engineer Europe 2026 talk, “The agent-ready web: Simplify user actions with WebMCP”, proposes one concrete pattern: sites exposing their actions directly to agents through a standard interface. Treat that as what it is — an emerging proposal with serious people behind it, not a discovery guarantee and not yet something to rebuild around.

What both audiences reward is the same thing your customers do: meaning that is actually in the page, evidence that can be checked, and actions that behave predictably. That is why the working test is parity.

The claims going around

Myth versus evidence

Verdicts current as of 11 September 2026, reviewed monthly while this territory moves.

The claim	What the evidence says	Verdict
“You need an AI-optimisation plugin or badge”	Google’s guidance lists fundamentals, not plugins; a badge changes nothing a crawler reads	Avoid
“Publish hidden text or shadow pages for bots”	Content hidden from users sits close to cloaking in Google’s spam policies; parity is the safe pattern	Avoid
“A special markdown file makes you agent-legible”	GA has seen no published guarantee from any major platform that such files drive discovery; meaning machines can use must live where people read	Unproven — GA position
“Semantic HTML and supported structured data still help”	Both remain documented, legitimate practice in Google’s guidance	Adopt
“Agents will transact on sites via standards like WebMCP”	A real, dated talk and active experimentation; an emerging pattern, not a guaranteed channel	Watch and prepare

“Avoid” rows are GA judgement applied to the cited policies, not a quote from them. The WebMCP row is deliberately cautious: prepare through parity, not through speculative rebuilds.

Four checks

The human/agent parity test

Run all four groups on your key pages. Whatever fails for a machine usually fails for a person on a bad phone in a hurry, too.

0 of 9 complete

Semantic: the page’s main meaning is real text in the HTML, not baked into images or videoSelect-and-copy the key claim. If you cannot, a machine cannot read it either.
Semantic: headings describe the section beneath them, so a machine could outline the page correctlyRead just the headings as a table of contents.
Semantic: key facts — services, coverage, prices, hours — appear as text, consistently phrasedThe same fact should read the same way on every page.
Visual: nothing essential is revealed only on hover, in a carousel, or behind a clickMachines do not hover; neither do most customers.
Visual: contrast and size keep the same content readable on a phone in sunlightAccessibility and machine-legibility are the same work viewed from two sides.
Action: every primary action — call, enquire, book, buy — is a real link or form control with a labelAn agent (and a keyboard user) needs a control, not a painted div.
Action: forms state what they need and why; errors explain the fix, not just the failure“Invalid” teaches nothing. “Phone numbers need 8–10 digits” completes tasks.
Confirmation: completed actions produce a visible confirmation and a reference a human could quoteAn agent needs to know the action landed; so does your customer.
Confirmation: the same state is true everywhere — what the confirmation says matches what the office seesIf the site says “booked” and the office sees nothing, that is the first fix.

Parity is the principle: no shadow content for machines, no degraded experience for people. Fix failures in the human interface first — the machine inherits the improvement.

Honest position

What GA is doing about the agent path

This site practises the parity position: semantic HTML, evidence in the open, actions as real controls, confirmations that match office state. Ask Maven is GA’s supervised-assistant demonstration — grounded, with sources and limits shown, and escalation to a human.

On WebMCP specifically: GA is tracking the proposal and will test it when a stable implementation surface exists. Until then, no client site will be sold “agent integration” on the strength of an emerging talk. The AI-readable structure resource holds the stable version of this advice; this article tracks the moving edge.

Evidence honesty

Sources, method and what would change this

Verified fact
Google’s AI-feature guidance
Google’s documentation for generative AI features in Search requires no special markup or new file formats; the same people-first fundamentals apply.
Google Search Central, “Optimizing your website for generative AI features”, 2026
Verified fact
Hidden-content risk
Google’s spam policies treat content shown to crawlers but hidden from users as abuse. Shadow content for bots is a risk, not a strategy.
Google Search Central, “Spam policies for Google web search”, 2026
Verified fact
WebMCP and the agent-ready web
Tara Agyemang’s talk proposes WebMCP as a way to expose site actions to agents. It is an emerging pattern — not a discovery guarantee, and not yet a standard GA will sell against.
Tara Agyemang, “The agent-ready web”, AI Engineer Europe 2026
Proposal
The parity test
The four-check parity test is GA’s formulation, published as a design to be run and judged. It encodes the cited guidance into something a business can run in an afternoon.
GA Applications method
Proposal
What would change this conclusion
If a major platform published an agent-specific discovery standard with measurable uptake, or WebMCP reached stable cross-vendor support, this piece and the parity test would be revised and re-dated.
GA Applications editorial

Get a parity reading on your key pages

GA runs the parity test on the pages that carry your enquiries and returns the failures in fix order. Most are cheap; all of them help people too.

Request a parity reading Read the AI-readable structure guide

No AI badges sold. No hidden files installed. Meaning and action, verifiable by you.