How a Firm's Entities Resolve in Search and AI Answers

A professional-services firm is not one entity. It is a cluster: the firm itself, every named practitioner inside it, the office locations, the credentials each person carries, and the matter types the firm actually handles. Search engines and AI answer systems resolve that cluster independently of each other, from whatever public evidence they can corroborate.

When the cluster is coherent, a machine can answer both kinds of question a prospective client asks — the branded one ("who is this firm?") and the intent one ("who handles this problem, near me, with this credential?"). When it is incoherent, the branded answer still works and the intent answer quietly fails. That failure is invisible from inside the firm, because everyone who tests it already knows the name.

§ 01The two questions machines are asked

Branded queries

Someone already has the name. They heard it from a referral, saw it on a filing, or read it in a directory. The machine's job is retrieval: find the firm, confirm it exists, return the address, phone, and role. This resolves off basic corroboration — a consistent listing across two or three public sources is usually enough.

Intent queries

Someone has a problem and no name. "Board-certified family law attorney downtown." "Securities lawyer for a Form S-1." "High-conflict custody, county court." The machine's job here is selection: build a candidate set, then rank it by how confidently each candidate matches the described need.

Selection needs more than existence. It needs the machine to have already classified the firm as the kind of entity that handles that specific matter, with that specific credential, in that specific jurisdiction — and to have seen that classification corroborated somewhere other than the firm's own website.

§ 02Why the branded answer succeeds first

Branded resolution is the easy half and it arrives early, which makes it dangerous. A firm can watch an AI system describe it accurately, name its founder, cite its bar role, and surface its address — and conclude that visibility is solved.

It isn't. The branded answer proves only that the entity exists and is correctly labeled. It says nothing about whether the firm appears in a candidate set when its name is absent from the prompt. Those are different machine operations running against different evidence. This is the distinction the seven-stage path separates: Recognition and Citation are not Inclusion and Selection.

§ 03Case study: a credentialed firm that machines find but clients can't reach

An anonymized professional-services firm — a regional family-law practice with genuinely rare credentials — was audited across both query types. The credentials were real: a board-certified specialist in the practice area, a former circuit-level magistrate on staff, national press coverage, five attorneys each carrying something distinguishing.

What the branded query returned

A branded search triggered a Google AI Overview that correctly identified the founder as an attorney in the named city, cited her bar-association leadership role, and surfaced the office address and phone with a directory citation. Accurate, specific, and sourced. The entity resolved.

What the intent queries returned

The same audit ran the queries a prospective client would actually type, with no firm name in the prompt:

Branded resolution confirmed; intent resolution not confirmed. Same firm, same week, same systems.

What the gap was made of

Four structural conditions, none of them ranking problems:

1. 01Credentials lived only on the firm's own site. Board certification, the magistrate background, and the bilingual capability appeared in the firm's own prose and nowhere a machine treats as third-party corroboration — not in the bar directory profile, the legal aggregators, or the firm's own structured data.
2. 02The practitioners were not modeled as entities. Five attorneys existed as page copy, not as resolvable named entities with stable identifiers, credentials, and a link back to the firm.
3. 03Identity signals disagreed. Multiple phone numbers across the site and directories, and a firm name written three slightly different ways, gave the corroboration step contradictory evidence to reconcile.
4. 04Matter types were implied, never stated. The site described a philosophy of practice; it never listed, in machine-readable form, the specific matters the firm takes.

The result of those four together: a correctly labeled entity that is under-corroborated. Machines can confirm the firm when handed the name. They cannot assemble a case for including it when the name is withheld.

What the fix looks like

The intervention is not more content. It is evidence architecture — the same five steps as the Entity Lock Protocol, applied to a cluster instead of a person:

• One canonical form of every identity signal. Firm name, address, and phone identical across the site, the bar directory, every legal aggregator, and the business profile. Contradiction is the single most expensive thing in the cluster.
• Each practitioner modeled as a named entity. A stable identifier, the distinguishing credential attached, and an explicit relationship to the firm — so a credential query has something to select rather than a paragraph to interpret.
• Credentials corroborated off-site. A certification asserted only on your own domain is a claim. The same certification present in the issuing body's directory, in the aggregator profiles, and in third-party coverage is evidence.
• Matter types stated explicitly. Named, enumerated, machine-readable — so self-selection happens before the call, and so the classification step has something to classify.
• Structured data that matches the visible page. Schema that asserts credentials, practitioners, and service areas the page itself does not display is a contradiction, and contradictions get discounted.

§ 04What to measure instead of rankings

Ranking reports will not show this gap, because the branded terms a firm tracks are exactly the terms that already work. Run two fixed prompt sets on a cadence and compare them:

• Set A — branded. Firm name, founder name, name plus city. Confirms the entity exists and is labeled correctly.
• Set B — intent. Credential queries, situation queries, jurisdiction queries. No firm name anywhere in the prompt.

Record whether the firm appears in the answer, whether it appears in the cited sources, and whether the credential is attributed to the right person. The distance between Set A and Set B is the actual visibility problem, and it is the only number worth reporting to a managing partner.

§ 05The short version

Getting found by name is retrieval. Getting selected without a name is a different machine operation with a higher evidence bar. Most credentialed firms have already cleared the first and have never measured the second — and the fix is corroboration architecture, not more pages.