Building the Knowledge Foundation for AI Discovery: From Schema to Entities

For more than two decades, being found meant earning a place on the results page. Publish pages, target keywords, compete for rankings, count the clicks. Entire teams and budgets were built around that model, and it is now giving way faster than most organizations planned for. 

What changed is not another algorithm update. It is the interface itself. AI engines, among them Google AI Overviews, ChatGPT, Gemini, Claude, and Microsoft Copilot, increasingly answer the customer’s question directly, reading across many sources and returning one synthesized response without the customer ever visiting a site. When that answer names your business, you win the moment. When it does not, the customer may never learn you existed. 

The goal is no longer to rank for a term. It is to become the verified entity an AI engine trusts enough to recommend. That discipline did not appear from nowhere. It grew out of structured data. Schema used to earn rich results. Now it builds the knowledge foundation that decides whether AI engines can find, understand, and recommend you at all. 

What actually changed

Three shifts are happening at once, and they compound. 

From keywords to entities. Traditional search matched the words on a page to the words in a query. AI engines build an internal model of your business first: who you are, what you offer, where you operate, and how your products, locations, and policies relate. This is the move from strings to things to entities, from matching text, to recognizing distinct objects, to reasoning over a connected ecosystem of them. A page can rank for a keyword and still be invisible to a system that cannot resolve what your business actually is. 

From links to answers. Discovery used to end on the results page and continue on your site. Now the engine resolves the question in place. Visibility no longer means a ranking; it means being the source the answer is built from. Rank well on a page few people reach, and it is worth less every quarter. 

From people to agents. Increasingly the visitor is not a person but an AI agent acting for the customer, comparing options, checking availability, completing tasks. Agents do not browse the way people do. They find, read, and decide in one pass, and they need information that is structured, retrievable, and verifiable. The businesses an agent can read cleanly are the ones it will choose. 

Together these raise the bar. Being discoverable no longer means being indexed. It means being a well-understood, consistently described, verifiable entity, a harder bar than ranking but a more durable one, because it rewards clarity over tactics that reset with every update. 

Dig deeper: Winning the AI decision layer: From AI discovery to agentic commerce

The framework: build, maintain, prepare

Winning here is not a campaign. It is a discipline that runs in three connected stages, and skipping ahead is the most common way organizations stall. 

Build a knowledge foundation 

An AI engine cannot recommend a business it cannot resolve. Start with a clean, connected model of your organization. 

1. Map entities, not pages. Model how your organization, brands, products, and locations relate. Build top down, from the organization to the offer to the price that changes most often, so every detail traces back to one hierarchy instead of living in scattered pages.

2. Give every entity one permanent address. Anchor each page to a single authoritative record with a stable identifier, an @id. When pages point back to the same reference instead of re-describing the business, an engine can resolve, with confidence, that they all describe the same company.

3. Deploy the foundation first. Website, Organization, WebPage, and Breadcrumb markup can go live immediately, so value starts on day one while richer product and location detail layers on afterward. Do not wait for a perfect model to ship the base.

Maintain trusted, current knowledge 

An entity model is only as valuable as it is accurate, and AI engines lose confidence in sources that drift or contradict themselves. 

1. Watch for drift. Monitor the entity layer for conflicting names, stale descriptions, and mismatched details, and correct them before they spread into the engines reading you. Accuracy becomes a continuous state rather than a periodic audit.

2. Pull volatile data from the source. Prices, availability, and policies change faster than any manual process can track. Feed them directly from your source systems on a schedule so your entities reflect reality within hours, and signal each change the moment it happens so the engines that matter see it in near real time.

3. Govern what changes and how. Let low-risk fields update automatically while pricing and policy require approval and deliver server-side or at the CDN edge so the information renders before the page loads and AI engines see it every time. Client-side is a fine start but treat it as a stepping stone.

Prepare for agentic discovery 

The first two stages make you readable to today’s engines. The third readies you for the visitor coming next. 

1. Audit the way an agent reads you. Score your site across the pillars an agent actually moves through: whether it can find, understand, retrieve, trust, choose, and act on your content. Knowing where you are strong and where you are opaque is the starting point for everything else.

2. Make your site conversational for machines. Approaches such as NLWeb let an agent draw accurate answers from your own information rather than scraping and guessing. Treat it as a new front door, not a side experiment.

3. Publish verifiable offerings. Emerging standards such as Agentic Resource Discovery (ARD) let an agent confirm what you offer by reading a structured file you publish under your own domain. It is your entity foundation, extended to agents so they can act on it with confidence.

How Milestone is operationalizing this

 The framework above is where the industry is heading. It is also what has guided the latest generation of Milestone Schema Manager, which has moved from a tool that generates valid markup to one that helps AI engines understand, trust, and continuously interpret a business as it changes. 

Building better knowledge. Milestone Schema Manager models the relationships between your organization, brands, products, services, locations, and offers, so AI engines see a coherent organization rather than scattered facts. It builds your entity registry from the top down, establishing the hierarchy of Organization, Brand, Product, Offer, and Price before any detail is filled in, and assigns each entity a permanent @id so every page refers back to one authoritative record instead of re-describing the business. 

Maintaining trusted knowledge. Foundational types, Website, Organization, WebPage, and Breadcrumb, deploy immediately so value starts on day one while richer detail layers on afterward. Milestone Schema Manager watches the entity layer for drift and flags inconsistencies before they spread, pulls volatile data such as prices and availability directly from your source systems on a schedule, and pairs with IndexNow to notify AI engines the moment something changes. Governed field rules decide what updates automatically versus what needs approval, and server-side or CDN-edge delivery renders your information before the page loads, so AI engines see it every time. 

Preparing for agentic discovery. A unified AI Readiness audit scores your site across the six pillars an agent moves through, find, understand, retrieve, trust, choose, and act, reading each page the way an agent does. NLWeb turns your site into something an agent can converse with, drawing answers from your own information rather than scraping and guessing. And support for Agentic Resource Discovery lets agents verify what you offer from a structured file you publish under your own domain. Together they turn the knowledge you have built into something agents can read, trust, and act on.

Where to start

Narrow it to one honest test: can an AI engine explain your business back to you accurately today? Ask the engines your customers use who you are, what you offer, and how you compare. The gaps are your roadmap, and they almost always trace to a foundation that is incomplete, inconsistent, or out of date. 

From there, sequence the work: build the entity foundation, wire it to live data with governance around it, then prepare for agents. Assign an owner. This is not a markup project you finish and forget. It is a standing capability, closer to how you treat data quality than how you once treated a keyword campaign. And it works only as one loop. Content, listings, structured data, and reporting are read together by the engine at once, and wherever they disagree, it loses confidence and recommends a competitor it trusts more. 

The bottom line

Search did not disappear. It changed shape. The page gave way to the answer, the keyword to the entity, and the human visitor is being joined by the agent. The businesses recommended in this era will not be the ones with the most pages or the cleverest tactics. They will be the ones an AI engine can understand clearly, trust consistently, and verify on demand. 

The knowledge foundation you build now is not a hedge against the next update. It is the asset that compounds, because clarity and consistency do not go out of style the way tactics do. Build it, keep it accurate, ready it for agents, and you stop chasing visibility and start owning it.