The four layers of agent readiness
When an AI engine decides whether to recommend a business, it runs four tests. Most businesses fail at least two without knowing the tests exist. Here is each layer, how engines probe it, and the failures we see most often.
Layer one: technical. Can a machine parse you?
Before an engine can recommend you, it has to read you. Not admire your design. Read you: extract your services, your credentials, your prices, your locations, as data. The technical layer is everything that makes that extraction possible or impossible.
What engines look for: structured data (JSON-LD schema for your organization, services, FAQs, and offers), content that exists in the HTML rather than behind JavaScript, a robots.txt that does not block AI crawlers, and pages fast enough that crawlers finish reading them.
The most common failure in healthspan is the beautiful brochure site: a visually stunning page where every service is an image, every price is "contact us," and the schema markup is absent entirely. To a person it signals luxury. To a machine it signals nothing, and nothing is what the machine passes on to the person asking.
Layer two: content. Do you answer the questions engines are asked?
Engines are asked questions, not keywords. "Is this clinic legitimate?" "What does a first visit involve?" "How much does a full assessment cost?" The content layer is whether your own surface contains quotable, accurate answers to the questions your prospective clients actually type.
What engines look for: clear service definitions in plain language, visible FAQ sections, evidence attached to claims, and pricing that can be found rather than inferred. A useful test: could a model quote two sentences from your site, verbatim, and have them stand as a correct answer to a real question? If every sentence needs your brand context to make sense, the answer is no.
The most common failure: atmosphere instead of answers. Pages that say "a transformative journey to your optimal self" and never say what happens, who does it, and what it costs. Engines cannot recommend atmosphere.
Layer three: trust. Do independent sources vouch for you?
Engines do not take your word for it. When composing an answer, they consult independent sources: editorial publications, comparison guides, reference indexes, review platforms. Your presence and consistency across those sources is the trust layer, and it is the layer your own website cannot fix alone.
What engines look for: citations in the publications they already pull from, agreement between what you claim and what the record shows, and presence in neutral reference layers. In our research for the AI Visibility Report 2026, the businesses that appeared in AI answers were the ones multiple independent sources agreed about. Self-description was never enough on its own.
The most common failure: assuming reputation transfers. A clinic can be famous among peers and invisible in the citation pool, because prestige inside an industry and machine-readable independent validation are different assets. The second one is buildable. The first one, alone, does not show up in answers.
Layer four: action. When an agent decides on you, can it act?
This is the youngest layer and the one that will matter most. AI is moving from recommending to doing: booking, quoting, enrolling. When an agent settles on you as the answer, the action layer is whether anything can happen next without a human untangling a contact form.
What engines and agents look for: a booking or inquiry path that can be traversed programmatically or at least described precisely, structured contact data, and stated expectations (response times, availability, what booking requires). Today this layer differentiates. Within two years it will gatekeep.
The most common failure: the phone-only funnel. If the only way in is a phone call during your business hours, an agent working for a client nine time zones away recommends someone it can act on instead.
How the layers interact
The layers compound in one direction and fail in the other. Strong technical work makes your content extractable; extractable content makes independent sources describe you accurately; accurate independent description is what earns the recommendation; and the recommendation is wasted without a path to action. One weak layer caps the value of the other three, which is why point solutions (a schema plugin here, a PR placement there) so often produce no visible movement in answers.
This is also why we scan before we fix. The four scores tell you where the cap is. Fixing the wrong layer first is the most expensive mistake in this discipline, and the most common one.
Want to know your four scores? The scan is free and a person runs it.