The questions start at 2am. The trust does not.

This is structural, not a matter of quality. The model already has the original. A paraphrase of an explanation available on a hundred other sites contributes nothing that was not already in the pile. There is a Google patent, "Contextual Estimation Of Link Information Gain," that defines the idea exactly: an information gain score reflects "additional information that is included in the document beyond information contained in documents that were previously viewed by the user." Google has not confirmed using it, and the coverage is careful that it does not mean longer pages rank higher. What matters here is the stated application: the patent is framed around automated assistants and chatbots rather than the ordinary list of blue links.

Take it as a writing principle, not a ranking promise. The things only you can say are what you actually do, what it actually costs, what happens at a first visit, who you take and who you turn away, what you tell a frightened person on the phone at 2am, and the specifics of your city, your court, your hospital system. None of that exists anywhere else, which is the only durable reason to be quoted.

If your practice, your own name, and your legal entity read to a machine as three unrelated businesses, no single identity accumulates enough evidence to be named at all. Structured data is the fix for that, and the platforms say so themselves: Google's search team has said structured data gives an advantage in search results, and Microsoft has said schema markup helps its models understand content for Copilot.

What matters is which claim you are being sold. One December 2024 study found no correlation between how much schema a site carried and how often it got cited, and there are still no peer-reviewed studies on schema and AI visibility. Read that for exactly what it measured: piling on more markup does not buy citations. It says nothing about the job structured data actually does, which is making it unambiguous to a machine who you are, where you are, and which facts belong to you. That work is cheap, it is one time, and it holds. Anyone pricing markup by the yard, or quoting you a citation lift per schema type, is selling the thing that got measured and came back empty.

A typical FAQ block is page decoration: a stack of one-line answers that mean nothing once separated from the page around them. Retrieval needs the opposite. It needs two or three sentences that answer the real question standing entirely alone, name the practice and the city, contain at least one fact a stranger could verify, and carry a date and a person's name. Most professional pages do not contain a single passage that clears that bar. The good news is that this is writing, not a redesign. Any one passage is an afternoon's work. Doing it for every question that actually brings you clients, in one consistent voice, without tripping your board's advertising rules, is the part that takes a system.

There is now real plumbing for putting a first-party source directly inside an assistant. Anthropic published the Model Context Protocol in November 2024, an open standard for secure two-way connections between an organization's own data and AI tools. OpenAI's Apps SDK, announced at its developer day in October 2025, is built on that protocol and lets developers make apps discoverable and callable inside ChatGPT conversations.

Now look at who is on the early lists: developer tools, and large consumer brands like Spotify, Figma, Booking.com, and Zillow. Not one independent medical, legal, wealth, or behavioral health practice. This is the most literal version of being the voice in the chatbot. Not hoping to be quoted, but being the source it calls.

The honest caveat is that frontiers have graveyards. Consider llms.txt, a sensible convention proposed in September 2024 for handing language models a curated index of a site. An Ahrefs study of roughly 137,000 domains found about 28 percent now publish the file, and of those with a valid one, 97 percent received no requests at all in a single month. Google's John Mueller has publicly dismissed it. A standard can be real, well-intentioned, widely adopted, and still read by almost nothing. So take the callable surface as a position worth holding early, and never as the thing you bet the practice on.