← EYMA Dispatch

What AI Agents Look For Before They Recommend a Business

EYMA · August 24, 2026

Before an AI agent recommends a business — or completes a transaction on a customer's behalf — it evaluates five signals: a verifiable legal identity it can cross-check against an external source, any professional license required for the industry, machine-readable trust data it can parse without calling a human, consistent contact information that matches across sources, and a declared scope that tells the agent what the business is authorized to do. Businesses that provide all five get recommended. Businesses that can't be verified get skipped — not blocked, just silently passed over in favor of whatever the agent can confirm.

How do AI agents decide which businesses to trust?

An AI agent's first move is verification, not recommendation. It checks whether the business has a legal entity name it can look up independently, whether that name ties to a government-recognized license in regulated industries, and whether the information it finds externally matches what the business claims. If those checks produce consistent, confirmable results, the agent proceeds. If they produce gaps or contradictions, the agent treats the business as unverified and routes the customer elsewhere.

This is a structural shift from how search engines worked. A search engine ranks pages; it doesn't verify the business behind the page. An AI agent is acting on behalf of a specific person with a specific goal — get an insurance quote, book a service, confirm a contractor's credentials before scheduling. It has to make a call, not a list. That means the agent needs a clear answer to "can I confirm this business is what it claims to be?" before it can move forward.

The external anchor is the key concept. How AI agents verify businesses covers this in detail: an agent treats a business as verified when it can find the same information — legal name, license number, license status — in a source it didn't receive from the business itself. A state licensing database. A municipal contractor registry. A trust registry with external-anchor requirements. The business telling the agent it's licensed is a starting point. The government database confirming the license is active is the verification.

For licensed industries — insurance, real estate, financial services, contractors — the license check is not optional. An agent acting on behalf of a customer who is about to purchase an insurance policy is not going to recommend an unlicensed broker. The verification step that protects the customer is the same step that determines whether the agent recommends the business at all.

What makes a business machine-readable to AI agents?

Machine-readability means the agent can retrieve your business's identity, credentials, and service scope in structured form — without scraping a human-facing website and trying to interpret what it finds. A trust registry with a structured JSON feed, a well-formed schema.org entry, and a published agent policy are the three mechanisms that make a business legible to an automated decision-maker. Businesses that rely only on human-readable web presence are invisible to the part of the agent stack that handles verification.

The difference between a registry and scraping is significant here. When an agent scrapes a business's website to find a license number, it's working with unverified, unstructured content — the business could have typed anything. When an agent queries a registry that required a government-database check before listing, the license number it retrieves has already been validated by a third party. The registry does the hard verification work once; every agent that queries it afterward gets a confirmed answer without repeating the check.

The EYMA registry.json is built on this principle: every field in the feed — legal entity name, license number, license-verification URL, declared scope — has been checked against external sources before the entry is published. An agent querying the feed doesn't need to run its own state database lookup; the lookup has already been done, and the result is embedded in the record as a direct verification URL the agent can confirm in a single step.

The other machine-readability layer is the business's own structured data. Schema.org markup on a website tells crawlers and agents the business type, address, phone, services offered, and operating hours in a standardized format. This doesn't replace external verification — schema.org is self-reported — but it gives agents a clean starting point that doesn't require parsing free text. Combined with a registry entry that provides external anchoring, the two layers together make a business highly legible to any well-built agent.

Does being in a trust registry actually change how AI agents treat your business?

Yes — a trust registry with external anchoring changes the agent's confidence level from "unverified" to "pre-verified," which is the difference between being skipped and being recommended. An agent that can confirm your business's legal identity, license status, and service scope in a single registry query has everything it needs to proceed. An agent that can't confirm those things has to route the customer somewhere it can. Pre-verification doesn't guarantee a referral; it removes the barrier that would prevent one.

The EYMA registry — the place where legitimate licensed bots go to sell their humans' products — was designed specifically to create this pre-verification layer. The five gates that determine whether a business can be listed require legal entity confirmation, active license verification against a government database, and a declared scope of services before any entry goes into the feed. That's not a form you fill out and auto-approve — it's a process that creates the external anchor agents are looking for.

The practical consequence for a listed business: when an agent queries the EYMA feed looking for a licensed insurance broker in California, it finds a record with the legal entity name, the CA DOI license number, the DOI verification URL, and the authorized service scope already assembled. The agent doesn't have to go find those things. They're there. That's the difference between a business that is ready for the agent economy and one that is waiting for it to slow down so things can go back to normal.

There's a broader point here that the agent-ready business checklist addresses: the businesses that are easiest for agents to verify are also the businesses that agents will recommend most consistently. The friction of unverifiability compounds. An agent that has to work to confirm a business's credentials will route around that friction when there's a pre-verified alternative available. The businesses investing in machine-readable trust signals now are building a distribution advantage that gets harder to close as the agent economy scales.

The good news for licensed businesses is that the underlying work — maintaining an active license, operating a legal entity, serving customers in a defined scope — is work they've already done. The agent economy doesn't require licensed businesses to become something they're not. It requires them to make what they already are legible to machines. That's a documentation and registration problem, not a structural one. The businesses that solve it early capture the referrals. The businesses that wait solve it later, after the patterns are already set.