Agent discovery/Schema coverage/Five-move readiness
Why AI Agents Decide Who Gets Considered
Independent structured-data crawls put schema.org coverage at roughly 30 percent of the web, which means an estimated 70 percent of pages carry nothing a machine can parse without guessing. When a builder asks an AI assistant to pick an API or a data source, the agent reads that machine surface first, and a page with no schema, no llms.txt, and no discoverable manifest is often out of consideration before a human ever sees a result.
The audience shift
Who Is Actually Reading Your Landing Page First?
A meaningful share of the people evaluating a new API right now are not reading it themselves. Roughly 63 percent of vibe coders are non-traditional developers, and a growing share of the research and integration work in a project gets delegated to an AI assistant that reads documentation, picks an endpoint, and writes the client code. That assistant is doing exactly what this piece describes: reading the machine surface first, then reporting back in plain language.
The practical effect is that a page written only for a careful human reader can still fail with this audience. An assistant skimming raw HTML for structured facts does not scroll, does not infer tone, and does not give a vague page the benefit of the doubt. It either finds a parseable answer or it moves to the next candidate. For a builder shipping a data API into this world, that means the developer docs and the machine-readable surface are doing first-pass sales work that used to belong to the homepage alone.
See how this plays out in practice in the docs, or read the entity building piece for the layer below schema: how an agent decides whether an entity is even real.
The machine surface
What Does an Agent Actually Look For?
Three layers get checked, in roughly this order: structured data that names what a page is, a plain-text discovery file that lists what a company offers, and a protocol manifest that describes how to transact if the answer is yes. Missing any one layer does not make a page invisible, but it does make the agent do more inference with less confidence, and less confidence means a lower spot in whatever shortlist it returns.
The cost of invisibility
What Happens If a Page Has None of This?
An estimated 69 percent of searches already end without a click, and AI Overviews now appear on about 83 percent of informational queries, so the pattern of a machine reading a summary instead of a human reading a page is not a future problem. A page with no structured data does not get excluded outright, but it forces the reading system to guess, and a guess is a worse answer that is less likely to get repeated or cited the next time someone asks a similar question.
The fix compounds in the other direction too. A page that answers its own question in the first paragraph, in plain text, with a number attached to a real source, gives an agent something it can quote directly instead of paraphrase. That is the same discipline behind every piece in this hub: a sourced number up front, one clear answer before the elaboration, and a plain-text paragraph an agent can lift without misquoting it. Read the compliance page for how the same discipline applies to provenance claims, or the glossary for the terms this piece assumes.
Questions, answered plainly
Does this mean my landing page copy does not matter anymore?
It matters to the human who reads the agent's summary, not to the agent doing the first pass of elimination. Copy still closes the decision. Structured data, answer-first paragraphs, and a discoverable manifest are what get a page into the set an agent is willing to summarize in the first place.
What is the fastest way to check if my own site is agent-readable?
Look at the raw HTML your server returns before any script runs, since that is what most crawlers read. If the meaningful content only appears after JavaScript executes, or if there is no schema.org markup anywhere in that HTML, an agent has to guess at what your page offers instead of reading it directly.
Is schema.org markup enough on its own?
No. Schema tells an agent what a page is about, but agents built for commerce and tool use also look for llms.txt, an agent.json or ai-plugin.json file, and protocol manifests like UCP and ACP that describe capabilities and checkout in a fixed shape. Markup and manifests answer different questions and both are checked.
Who should care about this: the founder or the engineer?
Both, but the fix is an engineering task. Adding schema, an llms.txt route, and a manifest is a few hundred lines in most frameworks, not a marketing rewrite. The five-move sequence in this piece is written so an engineer can implement it in an afternoon.
Build the surface agents can read.
Accounts are onboarded in launch order and API keys are issued at activation, but the documentation and the protocol surface are public today, so whatever is evaluating this API on your behalf can read exactly what it does before you ever sign in.