Five AI engines, five different shortlists. Your prospects are picking one.
Paid acquisition for commercial insurance brokers got expensive and stopped converting. The traffic migrated into ChatGPT, Claude, Gemini, Perplexity, and Grok — and when we asked all five engines the buyer prompts that decide where a controller starts a brokerage search, they named 72 different brokerages across six commercial buyer-intent questions. Inter-engine agreement was 0.38 (Jaccard, where 1.00 is full overlap). No single broker was unanimous on any prompt. Embroker was the only firm named by all five engines on any one question — the startup cyber and E&O prompt — and even there, ChatGPT routed the same query to Marsh, Aon, and Lockton instead. The shortlist still exists. It is being assembled by something else. This page reports what the engines named, why the answer set fragments by line of business, and what an in-market broker can build to surface where buyers are now starting.
The shift we keep hearing
Your CPL has climbed. Your SEO has flattened. The leads that still come through feel softer — lower-intent, more comparison shoppers, fewer people who actually need a broker. You spent more money to land worse prospects, and the producers who used to fill their pipelines from inbound are quietly working their networks again.
Two pressures compounded. AI Overviews and AI answer surfaces removed the informational queries that used to seed the top of your funnel. At the same time, more B2B buyers are starting product research inside an AI chat interface instead of a search bar — and the firms that surface in that interface enter the consideration set. The firms that don’t, do not. Discovery moved; the bid auction did not.
What we found when we asked AI which broker to use
In May 2026 we ran a bounded pilot capture — six commercial buyer-intent prompts, five AI engines, three runs each (ninety captures, scored line by line). The questions covered a mid-market manufacturer, cyber and E&O for an early-stage startup, a multi-location service business, a construction-liability scenario, a healthcare practice, and a small-cap D&O placement. The results were not subtle.
72
brokers named in total
across six buyer questions
0.38
inter-engine overlap
Jaccard · 1.00 = full agreement
0%
unanimous top broker
on any single prompt
5
engines named Embroker
on startup cyber & E&O
Marsh dominated ChatGPT (named in every run of every prompt where a national broker fit) and then disappeared on Perplexity and Grok, where smaller specialists surfaced instead. Hub International appeared in 83% of Gemini, Perplexity, and Grok answers — and only 50% of ChatGPT. Aon hit 100% on ChatGPT and 33% on Perplexity. The pattern is consistent: there is no broker that all five engines surface together, and no engine pair that agrees on more than half the time.
On startup cyber and E&O, the pattern inverted. Embroker was named by every one of the five engines, every run. Founder Shield and Vouch joined it. The national brokers fell back. That is what an answer-shaped, machine-readable web presence buys: when the buyer’s question matches the content you publish, retrieval flips from molten to settled.
One buyer question. Five engines. Two answers. The full 14-broker exhibit — including how the pattern shifts across mid-market manufacturer, construction-liability, D&O, and four other prompts — is in the Answerability Index entry for commercial insurance brokers.
If your diagnostic shows a similar gap, here is what we usually recommend building
For commercial insurance brokers, the typical post-Diagnostic build queue is six to eight specific assets:
- Three to five answer-shaped pages for the lines of business where AI surfaces a competitor instead of you — cyber/E&O, D&O, construction liability, mid-market manufacturing, healthcare-practice coverage. Buyer-question pages, not “Our Commercial Practice.”
- Entity-graph cleanup — consistent firm identity across IIABA / CIAB / Council of Insurance Agents & Brokers directories, your site, Wikidata, your LinkedIn company page, and broker-rating profiles. The engines stop confusing you with similarly-named regional shops.
- Citation assets — trade-press placements (Business Insurance, Insurance Journal, Risk & Insurance) and the wholesale-broker / association directory presence the engines actually weight.
- Competitor-comparison pages — your firm vs the named consensus players (Marsh / Aon / Hub / Gallagher / Lockton) for the lines you actually compete on. Honest, hedged, sourced.
- Schema package — Organization, InsuranceAgency, Service, FAQPage, BreadcrumbList, named-producer biographies with credentials. Ship-ready JSON-LD.
- FAQ expansion — passage-citable answer blocks (134–167 words each) on the high-intent coverage questions your prospects actually ask.
Build it in-house from the Diagnostic, or have us build it — a Sprint engagement ($15,000 then $950/mo) is done-for-you remediation over four weeks. Either way, the Diagnostic is the prerequisite: it tells you which six to eight assets matter and which would be wasted effort.
Order the Diagnostic to see your specific build queue — $2,500 →
Why brokerage retrieval fragments — and what to do about it
Commercial insurance is structurally hostile to a canonical answer. Brokerage authority is distributed: each carrier appoints multiple brokers, few firms are canonically associated with a single line, and the engines have no equivalent of a Wikipedia for brokers. They reach instead for adjacent signals — directory listings, trade-press mentions, association membership, the language of the firm’s own website — and those signals fragment across five different evidence systems that weight them differently.
That fragmentation is not a problem in the abstract. It is the reason your firm can appear strongly on one engine and be invisible on another for the same buyer question. The Diagnostic reads the gap and locates it in three places.
| Pillar | The question we answer | What we look for in a broker site |
|---|---|---|
| Content | Can the engines extract a useful answer from your site? | Server-side rendering (many broker sites are JavaScript shells the crawlers cannot read); answer-shaped pages for buyer-intent prompts (“cyber liability for SaaS”, not “Our Commercial Practice”); explicit specialization signals; named producers with credentials. |
| Retrieval | Do you appear when buyers actually ask? | Observed surfacing across the buyer-intent prompts that match the lines you place, on the engines your buyers use. Measured, not theorized. |
| Trust | Are you corroborated in the sources the engines weight? | Presence in the directories, associations, and trade outlets the engines actually cite; consistent firm identity across the long tail of secondary references; named-but-not-cited gaps (the engine knows you exist but does not treat you as a source). |
The shortlist is being assembled without you, every time a prospect opens ChatGPT. The first question is whether you are on it. The second is whether you can change that.
What you receive
Deliverables · 10–14 business days · under MNDA
An evidence-grounded intelligence report on how AI represents your firm
A 30–50-page bound evidence report your team can cite internally.
Every prompt, every engine, every response, scored line by line.
Page-by-page priorities ranked by retrieval impact. Hand it to a writer.
The buyer-intent prompts the engines treat as canonical for your lines.
Which firms hold the answer layer per line, with the evidence trail.
The Visibility Intelligence subscription artifact — what changed and why.
Built for
Brokerages, MGAs, wholesale brokers, and specialty risk advisors already spending on SEO, content, paid search, or producer enablement — and who need to know whether the AI engines are surfacing them in the buyer’s shortlist.
Order the Diagnostic
Stop losing customers to brokers AI happens to know better.
Delivered in 10–14 business days. Five engines. Thirty to fifty buyer-intent prompts measured for your specific firm. Confidential under MNDA. The first month is the full Diagnostic; thereafter the Visibility Intelligence subscription keeps the picture current as the engines move.
$2,500 first month (the Diagnostic) · $950/mo thereafter (Visibility Intelligence: monthly re-runs, deltas, build-queue updates)
- Answerability Index · Commercial Insurance BrokersThe full pilot capture: 14 brokers, 6 buyer questions, the heatmap with evidence drawers and the cross-engine pattern table.
- The Answerability FrameworkThe Content / Retrieval / Trust pillars, applied here and across every Diagnostic.
- How we workThe methodology behind every report — capture protocol, scoring, what we will and will not claim.
- MethodologyThe capture protocol behind every Diagnostic — how we acquire, score, and what we will and will not claim.
A note on scope. Insurance is regulated and varies by state and line of business. Nothing on this page is insurance advice, broker advice, or a solicitation to place coverage. The Diagnostic measures observed surfacing — how AI systems represent firms — not broker quality, broker suitability, recommendation quality, or fit for any specific risk. Surfacing patterns reflect the engines’ current behavior on the captured date and will change as the engines and the underlying web change. Capture: 2026-05-28. Six buyer-intent prompts, five engines, three runs per engine (ninety captures total). Broker names canonicalized; carrier names set aside as a different question.