Machine readiness is the way Topkay measures how well your site works for the machines that now stand between you and your audience: search crawlers, AI answer engines and the agents that browse and act on a person's behalf. SEO, AEO and GEO each look at one slice of this. Machine readiness is the whole thing.
Why one score
The reader has changed. For twenty years the job was to rank a page for a person who clicks. Then answer engines started lifting a passage and stating it without the click. Now agents discover services, judge whether to trust them, pick one and invoke it, sometimes without a web page at all. These are not three separate problems with three separate acronyms. They are one question asked at different depths: can a machine find you, understand you, trust you, choose you and use you. Machine readiness is that question, made measurable.
A bot reads, an agent acts
It is worth separating the two kinds of machine, because they want different things and "AI" blurs them. A bot is a crawler: it fetches a page and reads it, the way GPTBot, Googlebot or ClaudeBot do, to index the page or answer from it. It is passive, and it lives in the first three pillars, Discover, Understand and Trust. An agent is autonomous: it browses, decides and takes an action on a person's behalf, choosing a service and calling it, often through a tool or an endpoint rather than a page. It is active, and it lives in the last two pillars, Select and Invoke. Most sites are built only for the bot and have nothing for the agent, which is why Select and Invoke are where the ground is emptiest.
The five pillars
Every machine interaction with your site passes through the same five stages, so machine readiness scores each one.
- Discover. Can a machine reach and index your content at all. Robots access, crawlability, sitemaps, internal links and crawl efficiency live here. If you cannot be discovered, nothing else counts.
- Understand. Once reached, can a machine make sense of a page and lift a clean answer from it. This is your answerability: passages that stand on their own, structured data, clear topics. It is the same number as your Answerability Index.
- Trust. Does the machine have reason to believe you. A clear brand and entity, one canonical home instead of duplicates, a signed identity and consistent signals across the site. Self-declared identity is treated as weak, so proof matters.
- Select. When a machine has several options, does it choose you. This is competitive: whether your passages beat the field for a query, and whether your capabilities match what is being asked for. Being understood is not the same as being picked.
- Invoke. Can an agent actually use you. In the agent world the discoverable unit is often an endpoint or a tool, not a page: an Agent Card an agent can read, capabilities it can call and a way to authenticate, plus WebMCP that marks a form up as a tool an agent can invoke in the page. Most sites score low here today, because they have never published either. That is the frontier.
How the score works
The headline is not one number that hides the five pillars behind it. Topkay reports how many of the five it could actually measure, as a coverage figure, and shows each pillar on its own. Discover, Understand and Trust are measured from your crawl. Select fills in once you run AI visibility, and Invoke once you publish an Agent Card, so a fresh site reads as three of five measured. The headline number is the average of only the pillars that ran, shown next to that coverage figure, so it is never read as a finished five-pillar score. Select and Invoke are shown as their own pillars with an unlock path, never counted as zero, because a site that has not published an Agent Card is not failing Invoke, it simply has not been measured on it yet.
Treat the number as a calibrated readiness signal, not a guarantee. A higher score means your site is easier for machines to find, read and trust, which is the groundwork for being cited and chosen. It is not a promise that a given passage will be cited: retrieval depends on the query, the competition and the engine on the day. Optimise the pillars because they make the site genuinely better for a machine reader, and treat citation as the outcome you are working towards, not a figure the score entitles you to.
Machine readiness and SEO, AEO, GEO
Machine readiness does not replace SEO, AEO or GEO, it contains them. Classic SEO is most of Discover and part of Understand. AEO is Understand and part of Select. GEO spans Understand, Trust and Select across generative engines. Agent readiness is Invoke, and much of Select, and it is the newest ground: free scanners from Cloudflare, Chrome and Shopify now check whether the pieces are present, but presence is only the start, and knowing which of them actually change whether an agent chooses you is the part still unmeasured. If you already do SEO and AEO well, machine readiness shows you the ground you have not covered.
The signal that this is real
This is not a bet against the grain. Chrome now ships an experimental Agentic Browsing category in Lighthouse that scores, with deterministic checks, how well a site is built for machine interaction: whether its forms are marked up as tools an agent can invoke, whether its accessibility tree is complete enough for an agent to read, whether the page is stable enough for an agent to click the right thing. That is the Invoke and Understand pillars, measured by Google. The ground is moving under this exactly the way machine readiness predicts.
Optimise the proxy and you game a number. Optimise for machine readiness and you are building a site that machines can find, read, believe, choose and use. That is the outcome, not the metric.