WE BUILT A SITE AI ASSISTANTS CAN ACTUALLY READ.

Every brand fact — name, address, phone, area served, what the company knows about, who founded it, what else it is called, what it is not — lives in a single module and is emitted as one canonical organization node referenced by ID from every other schema node on the site. Pages reference the entity; they never re-type it.

One node, not two. The organization is typed as both an Organization and a ProfessionalService, which is a LocalBusiness subtype, so one node carries the brand identity and the local signals together. The site originally shipped a second business node on the homepage on top of it. We removed it. Two nodes describing one company is split-entity risk, not extra signal.

Explicit disambiguation. One short line in the graph stating which entity this is and which similarly-named company it is not. One line — deliberately not an enumeration of every competitor, because a list of rivals in your own graph reads as competitor copy and hands those entities free association with yours.

A real external identity. The node carries a Wikidata item, declared both as a link and as a structured identifier. Wikidata is the structured facts database the knowledge graphs and answer engines reconcile entities against, and declaring the identifier from your own graph closes the loop between your self-claim and theirs.

We wrote our own HTML-to-Markdown converter. 491 lines, no dependencies. A general-purpose converter plus a DOM implementation would have added megabytes to a route that runs on every twin request, to handle markup this site does not emit.

We did not chase page count.The site’s own diagnosis named templated thinness as the cause of its indexation problems. The build spent its effort making existing pages reachable, quotable and schema-complete before adding more.

We treated “what we have not done” as content.Each industry page states plainly what the company has not built in that sector and what adjacent work transfers. And the inverse: there is deliberately no “have you built for this industry before?” FAQ anywhere, because FAQs become FAQPage schema, and that question would ship a machine-readable statement about your gaps straight into the answer engines.

We published no fabricated ratings. There is no self-authored aggregate rating on the organization node. Review markup you write about yourself is worth nothing and risks having your rich results pulled.

Everything above is on-domain engineering. It makes a site citable. It does not make it cited. Language models weight third-party corroboration far above self-claims, and an entity that only asserts things about itself is an entity a model hedges on.

The work that has to happen off the domain, which no amount of schema replaces:

• —Third-party profiles with real content, and byte-identical name, address and phone on every one of them.
• —Genuine reviews on Google Business Profile and the directories — reviews other people write about you, not markup you write about yourself.
• —Real links from real, topically-relevant sites, including credit links on the sites you build.
• —Local institutional presence: chambers of commerce, regional business and technology directories.
• —Original, citable material of your own — data or research a model has a reason to quote.

We sequence it deliberately. The on-domain engineering is a one-time pass with a defined end, and it is the thing that makes every off-domain citation land on a site an assistant can read. Doing the citations first, into a site with no entity graph and no machine-readable surface, wastes them.

That pass is what we sell as the AEO engineering pass — $1,500 to $2,000, one time — and the search half is the SEO pass at the same price.