Two axes decide what AI cites for your category. Geography is a third.
6 measurement windows across 5 Hawaii categories + 1 Austin TX category. Same 7-AI-tool methodology, same 18-question discipline (hash-locked per geo). Own-site share is one axis. Training-data presence is a second axis. The Austin CPA cross-geo measurement adds a third dimension: the training-data axis itself can split by engine and by geography within the same category.
The six-measurement table (web-searching engines)
Own-site share aggregated across the five web-searching AI tools (Perplexity, ChatGPT search, Gemini grounded, Microsoft Copilot, Google AI Overviews). Same week, same hash-locked discipline so every number compares apples to apples within each measurement window.
| Category | Own-site share | 3rd-party share | Cohort size |
|---|---|---|---|
| Hawaii consumer banking | 53% | 42% | 21 firms |
| Hawaii wealth management | 47% | 49% | 42 firms |
| Honolulu dental practices | 44% | 52% | 46 firms |
| Austin TX CPA firms | 40% | 54% | 37 firms |
| Hawaii law firms | 39% | 58% | 33 firms |
| Hawaii CPA firms | 39% | 54% | 41 firms |
14 percentage points separate banking from the bottom of the table. Banking sits alone at the top. The other five measurements cluster between 39% and 47%. Hawaii CPA and Hawaii law tie at the bottom; Austin CPA sits right above them at 40%. The cross-geo CPA data point (Hawaii CPA 39%, Austin CPA 40%) is the most informative single observation here: same category, different geography, essentially the same web-searching own-share. The gradient is category-driven, not geo-driven, on this axis.
The hypothesis on why banking is different
Banks have done two decades of work making their websites the official source of truth for their own products. A bank’s rates, account types, branch locations, and product comparisons live cleanly on the bank’s own site, in structured product pages, refreshed daily. AI tools have nothing equivalent to cite for a wealth firm or law firm or dental practice or CPA firm. The "best wealth manager in Hawaii" question has no canonical source the way "best bank for small business in Hawaii" does, so AI reaches for a wider mix of editorial third-party content.
The honest finding for an operator on the web-searching surface. If you are a Hawaii bank, your own website is doing roughly half the work of getting AI to mention you. Clean it up; the leverage is on-site. If you are a Hawaii wealth firm, dentist, law firm, or CPA, your own site is doing 39-47% of the work and the rest is happening on third-party surfaces (publications, directories, named-attorney bylines, lead-gen platforms, professional association directories, etc.). The 39-47% on-site number means own-site work matters, and the 50%+ off-site number means it is not enough by itself.
The second axis: training-data engines (and the cross-geo split)
Two of the seven AI tools we measure (Claude and Gemma) answer from training data instead of searching the live web. So when they mention a business by name, it means that business has formed a presence in the AI’s training data: through Wikipedia, news coverage, named-founder content, sustained editorial presence over years. This is the slowest signal in the measurement and the most defensible once it forms. A new competitor cannot show up in Claude’s training data overnight. That presence is earned through years of editorial work that an AI model later learned from.
Across six measurement windows (5 Hawaii categories + 1 Austin CPA cross-geo):
| Category | Claude own-site share | Gemma own-site share |
|---|---|---|
| Hawaii consumer banking | 88% | 70% |
| Hawaii law firms | 51% | 78% |
| Honolulu dental practices | 38% | 59% |
| Austin TX CPA firms | 0% | 23% |
| Hawaii wealth management | 16% | 38% |
| Hawaii CPA firms | 1% | 2% |
The training-data picture is not the same picture as the web-searching one. Hawaii banks sit at the top of both. Hawaii CPA firms are tied with Hawaii law on the web-searching surface (both 39%) but collapse to 1% / 2% on the training-data surface, the lowest of any measurement by a wide margin. Law firms are in the middle of the web-searching table but near the top of the training-data table.
The Austin CPA row is the most informative single observation in this teardown. Same category as Hawaii CPA, same query shape (geo-swapped), same week, same methodology. Claude shows nearly identical behavior across the two geographies (Hawaii 1%, Austin 0%). Gemma shows dramatically different behavior (Hawaii 2%, Austin 23%). The training-data axis is not one number; it can split by engine within the same category depending on geography. The original two-axis frame needs a third dimension to accommodate this.
What this means in plain language
Own-site share is decided by current web infrastructure quality. Training-data presence is decided by decades of category-level editorial accumulation. A firm can move the first axis by improving its own website. The second axis moves on a different timescale and through different mechanisms (named editorial coverage, association bylines, sustained public presence). The implication for operators: treating "AI citation share" as one number averages two different surfaces that respond to different work. The cohort-level pattern is the diagnostic that tells you which axis your category sits on and which work pays off there.
The CPA cross-geo finding (the most informative observation)
The Hawaii CPA teardown originally surfaced a striking pattern: Claude cites Hawaii CPA firms 1% of the time, Gemma cites them 2%, while the four web-searching engines reach 45-60% on the same questions on the same week. The original framing called this a "training-data engine collapse" for CPA firms.
The Austin CPA cross-geo measurement refined that framing into two distinct findings. Claude's collapse generalizes; Gemma's does not.
Claude generalizes. Hawaii CPA Claude own-share is 1% (6 cites of 601). Austin CPA Claude own-share is 0% (3 cites of 885). Same engine, same model version, different geographies, essentially zero firm-owned citation rate in both. This is a category-level finding about Claude's training corpus being editorially-thin on CPA firms regardless of geography. Two-geography evidence promotes it from a low-confidence single-geo observation to a medium-confidence cross-geo pattern.
Gemma does not generalize. Hawaii CPA Gemma own-share is 2% (7 cites of 293). Austin CPA Gemma own-share is 23% (73 cites of 313). Same engine, same model version, different geographies, 11x difference in firm-owned citation rate. This is a geography-level finding about Gemma's training corpus: Texas CPA firms are represented substantially better than Hawaii CPA firms. The cause is inferred (likely geo-density of business-news content in Gemma's training data); the observation is direct.
The reason this matters: a vendor measuring one geography per category would have published the Hawaii result and called it the category-level pattern. The cross-geo measurement was the only way to discover that one of the two training-data engines does NOT actually have the category-level pattern the other does. A more precise read of training-data engine behavior emerges only when you measure the same category in multiple geographies.
For an operator in either CPA geography, the closable competitive surface differs in this specific way: Hawaii CPA firms have training-data engines as a category-wide blind spot, and competing inside the web-searching engines is the only short-term move. Austin CPA firms have Claude as a blind spot, but Gemma is reachable through sustained web presence over Gemma's training cycle. The recommendation differs based on which geography the operator sits in.
The Microsoft Copilot pattern (holds across all six measurements, including Austin)
One pattern is consistent across all six measurement windows (5 Hawaii categories + 1 Austin CPA). Microsoft Copilot (which answers using Bing’s organic search results) cites the firms’ own websites 0% to 2% of the time in every measurement.
| Category | Bing/Copilot own-site share |
|---|---|
| Hawaii consumer banking | 0% |
| Hawaii wealth management | 0% |
| Honolulu dental practices | 0% |
| Austin TX CPA firms | 0% |
| Hawaii law firms | 1% |
| Hawaii CPA firms | 2% |
This means: a buyer who asks Microsoft Copilot for a business recommendation in any of these six categories or geographies gets pointed at third-party content (publications, Wikipedia, directories), almost never at the businesses’ own websites. Every business in every category we measured shares this gap. The Austin data point confirms the pattern is not Hawaii-specific.
The first-mover read: whichever firm shows up first in Bing’s organic search results for these queries gets the Copilot answer mostly to itself, because every competitor is also invisible there. The gap is open, and it does not appear to be a category-specific or geo-specific problem. It is a Microsoft Copilot pattern across local-service queries.
The implication for where to do the work, by category and geography
The two-axis pattern, refined by the cross-geo CPA measurement, produces six different recommendations:
- Hawaii consumer bank. Own-site work pays off the most directly on both axes. AI is reading the bank’s website 53% on web-search and 70-88% on training-data. Cleaner structured data, sharper product pages, better FAQ markup. Both axes respond.
- Hawaii wealth firm. Own-site work matters (47%), off-site editorial matters (49%), and training-data presence is in the bottom tier (16-38%). The on-site cleanup is the necessary baseline; the harder work is getting into the third-party surfaces AI uses for the category (lead-gen platforms, named-planner editorial, professional bylines).
- Honolulu dental practice. Similar shape to wealth on the web-searching surface (44%), middle of the cluster on training-data (38-59%). Own-site work and review-directory presence both matter. The Microsoft Copilot first-mover opening is open here as in every category.
- Hawaii law firm. 39% on the web-searching surface, but training-data presence is high (51-78%) because attorneys generate long-form editorial content (Chambers profiles, BestLawyers entries, named-attorney bylines). The implication: own-site cleanup plus continued investment in the editorial surfaces that feed into training data over time.
- Hawaii CPA firm. 39% on the web-searching surface, near-zero on BOTH training-data engines (Claude 1%, Gemma 2%). Web-searching engines are where the competitive game plays. Training-data engines are a category-and-geo blind spot that no individual firm can close in the short term. Editorial work would help over years; in the meantime, the four web-searching engines are the addressable surface.
- Austin TX CPA firm. 40% on the web-searching surface (essentially identical to Hawaii CPA). Claude is a blind spot (0%) as it is in Hawaii. Gemma is reachable (23%): an Austin CPA firm CAN move the Gemma needle through sustained web presence in Texas-business-news content that Gemma’s training cycle ingests. This is the materially different recommendation between Hawaii and Austin CPAs. The cross-geo data surfaced it.
The non-obvious read across measurements: the conventional SEO instinct (clean up your website and AI will cite you) is correct in proportion to where your category sits on the web-searching axis, but it has nothing to say about the training-data axis. And the training-data axis is not one number; it can split by engine and by geography within the same category. A firm in a category like CPA in Hawaii could perfect its on-site infrastructure and not move either Claude or Gemma. The same firm operating in Texas would not move Claude but would have a path to moving Gemma. Both surfaces respond to different work; both surfaces can vary by geography in ways that aren't visible from a single-geo measurement.
The cohort sizes (transparency note)
The six cohorts were not the same size. Cohort size is determined by how many distinct businesses AI tools mention with enough frequency to register as recurring, then expanded through cohort-coverage scans across multiple runs. Categories with more independent businesses (dental practices, CPAs, individual law firms) surface larger cohorts. Categories with consolidation (banking, wealth management) surface smaller ones. This is a feature of the measurement, not a bias: the cohort reflects who AI actually cites, not who we picked in advance.
- Hawaii consumer banking: 21 banks and credit unions, named in full at teardown 01.
- Honolulu dental practices: 46 dental practices (39 surfaced from run #1, 7 more added after run #2 cohort-coverage scan).
- Hawaii law firms: 33 firms (5 originally registered, 25 surfaced through run #1 cohort-coverage, 3 more after run #3).
- Hawaii wealth management: 42 firms (15 originally registered, 27 surfaced through three rounds of cohort-coverage scans). The expansion lifted the own-site share from 27% on run 1 to 47% after the third expansion, illustrating the cohort-coverage discipline.
- Hawaii CPA firms: 41 firms (5 anchors, 15 surfaced from run #1, 8 from run #2, 11 from run #3, with 2 anchor domain variants corrected). The expansion lifted the own-site share from 27% on run 1 to 39% after the third expansion. See teardown 06 for the per-category breakdown.
- Austin TX CPA firms: 37 firms (5 anchors with 1 domain correction, 16 surfaced from run #1, 8 from run #2, 8 from run #3). The expansion pattern matched the Hawaii CPA category closely. See teardown 07 for the per-category breakdown and the cross-geo finding details.
All individual firms outside the banking cohort are kept anonymized in this public artifact per the non-customer anonymization rule.
Methodology summary
Each category was measured the same way:
- 7 AI tools: Perplexity (sonar API), ChatGPT search (gpt-4o-mini-search-preview), Gemini grounded (gemini-2.5-flash with Google search), Microsoft Copilot via Bing organic results, Google AI Overviews, Claude (claude-haiku-4-5), Gemma (open-weight, via Together AI).
- 18 questions per category, locked by hash so every run compares apples to apples. The questions mirror how a real buyer in each category would actually search.
- 3 repetitions per question per AI tool to separate signal from noise.
- 3 usable runs per category across the dataset. All five categories cleared the MOAT.md rule 5 pattern-readiness threshold. Banking, dental, and law numbers each held within one to two percentage points across their runs (stability signal at clean cohort coverage). Wealth's number rose from 27% on run 1 to 47% after three rounds of cohort expansion. CPA's number rose from 27% on run 1 to 39% after three rounds of cohort expansion. Both are the cohort-coverage discipline working as designed: more runs surface more cohort members, the headline number trends toward true ceiling rather than away from it.
Full methodology, source bucket definitions, and pattern-readiness rules at /methodology/. Public source code at github.com/LanceRoylo/neverranked-outreach. Gemma is open-weight, so any auditor (compliance team, outside reviewer, agency data lead, anyone) can independently re-run the same questions and verify the training-data numbers.
Honest scope and what this does not prove
Five categories is enough to establish a two-axis pattern but not enough to declare a universal law. The pattern is consistent in direction (web-searching and training-data engines respond to different mechanisms) and in magnitude (the spread is wide enough to matter) but it lives inside a small sample of categories, all in Hawaii, all measured in a roughly two-week window.
What is not in this teardown:
- National categories. Everything here is Hawaii-specific. Categories in larger markets may show different patterns because cohort dynamics differ.
- Categories outside professional services and consumer financial services. The five measured here cluster narrowly. B2B SaaS, restaurants, real estate brokerages, healthcare specialties beyond dental, all unmeasured here.
- Whether the pattern holds over time. AI tools refresh their training data and search indices on schedules we do not control. The same categories measured six months from now may produce different numbers.
- Causation. We measured what AI cites. We did not test whether changing what is on a firm’s site or in third-party publications would cause AI to cite differently. That requires pre-registered experiments against named businesses with control for confounds. Different scope, different methodology.
What an engagement does with this
NeverRanked engagements measure your specific category every week for as long as the engagement runs. For an engagement in a category that sits high on both axes (banking), the punch list weights toward on-site work. For an engagement in a category that splits between axes (law, where on-site is mid and training-data is high), the punch list covers both surfaces. For an engagement in a category that collapses on the training-data axis (CPA), the punch list focuses on the four web-searching engines because that is where the work pays off. The two-axis pattern above is one input to the recommendation; the per-buyer-question detail and the per-engine breakdown are the other. The work the data points at is named specifically. Whether the work moves the needle is a measurement question we keep answering month over month.
Measurement windows: all five categories measured across 3 usable runs each, spanning 2026-05-23 to 2026-05-26. Pattern-readiness cleared per MOAT.md rule 5 for every category in this teardown. Wealth's own-site share rose from 27% on run 1 to 47% on run 3 after three rounds of cohort expansion. CPA's own-site share rose from 27% on run 1 to 39% on run 3 after three rounds of cohort expansion. Both are the cohort-coverage methodology working as designed: more runs surface more cohort members, the headline number trends toward true ceiling rather than away from it. Banking, dental, and law numbers held within one to two percentage points across their three runs.
Substantiation: question sets locked by hash, open-source measurement code, named AI tools on named dates. The fact-checker (also public source) rejected zero claims in this teardown.
Anonymization: the Hawaii consumer banking cohort is named in full in teardown 01 because that artifact already established public naming for the cohort. The other four categories’ cohorts are anonymized here per the rule that non-customer businesses appear named only in 1:1 deliverables. Categories and counts are public; individual firm names are not.