What exactly does agent readiness measure?
Four things, with the weighting stated — and that weighting is here because a composite figure without a published split cannot be recomputed. Actionability weighs heaviest, and that is a position: it is the only dimension almost no site fills in today, and at the same time the only one that matters the moment an agent actually has to arrange something.
| dimension | weight | what gets measured |
|---|---|---|
| readability · readability | 25% | Can an agent get in, and does it recognise what kind of thing you are? Crawler access, an entity schema and a FAQPage. |
| actionability · actionability | 35% | Can it DO something? A machine-readable action, an offer and a contact point. Weighs heaviest, because this is the layer that is new. |
| decision reliability · reliability | 20% | Can it choose without guessing? Offer, price or price indication, and reviews. |
| evidence validity · evidence | 20% | Is what stands there current and checkable? Unknown counts as half here — not knowing is not a fault. |
What is the difference between citable and agent-ready?
Citable is about text, agent-ready about actions. A page that is perfectly citable — clear definition, real table, FAQ that holds up — can still let an agent do nothing: it knows what you sell and finds no way to order it. The other way round, a site full of booking actions can be unreadable to the engine answering the question, and then no agent ever comes by.
Why is a blocked crawler immediately a ceiling?
Because the other three dimensions are then about a building nobody can enter. If your robots.txt shuts the AI crawlers out, every action, every price and every source you publish is out of reach for exactly the party you did it for. A meter that still turns that into a 78 because the markup is tidy is measuring your good intentions.
May an AI agent order from us on a customer's behalf?
We ship the side that depends on us: the markup an agent needs to know what is possible — which offer, which terms, which steps. The ordering itself runs through your own shop and payment provider, and that is where the brake sits today: almost no agent takes that step yet. That is a fact about the market, not about your site, and it is exactly why the markup is already in place: whoever has it is ready on the day it tips.
What does it cost to become agent-ready?
For the first dimension usually nothing: crawler access is a line in a file, and an entity schema plus a FAQPage are already half there on most sites. The price sits in the second: a machine-readable action only exists if something can genuinely happen behind it, and that is often not a markup question but a question for your booking or ordering system.
Frequently asked questions
Do AI agents use this already today?
Barely, and that is not an argument to wait. This is the same trade-off as with llms.txt: the static part costs you nothing beyond the structured data you are placing anyway, and it is there on the day it does get read. The live part is another matter, and there we do wait — for traffic, not for a press release.
Why does actionability weigh more than readability?
Because readability is the entry ticket and actionability is the differentiator. Almost every serious site is readable by now; almost none is actionable. A meter in which the part everybody already has weighs heaviest gives everybody the same score and says nothing.
Which schema.org actions count?
Thirteen by name — reserveaction, orderaction, scheduleaction, bookaction, buyaction and the rest — plus ANY type ending in Action. That second rule is there because schema.org keeps adding action types and a fixed list would quietly go stale. Anyone using a valid action type we do not know by name still counts.
Read on
Free, no account and no card.