The web-searching engines find Hawaii CPAs. The training-data engines don't know any of them.

41-firm cohort, 18 hash-locked questions, 3 usable runs on 2026-05-26. Pattern-readiness cleared. Individual firms anonymized; counts and distributions named.

Why this category matters as a measurement subject

Hawaii CPA firms are a category where buyer trust signals matter more than commodity attributes. A buyer choosing a CPA is choosing a named partner's judgment on tax, audit, or advisory work that touches the rest of the business. The AI-citation surface reflects that. Firms compete for inclusion in editorial sources, professional associations, and category-specific directories in ways that purely transactional categories do not need to. With 41 firms in the cohort, the data shows where that competition actually lands.

CPA Hawaii is the fifth vertical NeverRanked has measured to pattern-readiness, after Honolulu consumer banking, Hawaii wealth management, Honolulu dental, and Hawaii law firms. The cross-category teardown reads CPA against the other four. The most distinctive structural pattern, surfaced only by measuring against the prior four, is the training-data blind spot named above.

Methodology summary

Same 7-AI-tool methodology applied across all NeverRanked teardowns:

• 5 web-searching 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.
• 2 training-data AI tools: Claude (claude-haiku-4-5, via Anthropic API), Gemma (open-weight, via Together AI).

18 questions a Hawaii CPA buyer would actually ask AI, locked at hash dc6ae677... so every run compares apples to apples. 3 repetitions per question per AI tool. 3 usable runs on 2026-05-26 (run #1, #2, #3 fired in succession). Pattern-readiness rule of 3 usable runs cleared per MOAT.md rule 5.

The 41-firm cohort was built in four passes. 5 anchor firms registered before run 1. 15 additional firms surfaced through the run #1 within-citation scan. 8 more surfaced after run #2 cohort-coverage. 11 final additions after run #3. National Big-4 firms (KPMG, PwC, Deloitte, BDO, CLA) and out-of-state firms were excluded from the cohort to keep the comparison Hawaii-specific. State professional bodies (Hawaii Society of CPAs, the UH Shidler business school), generic firm directories (Goodfirms, Clutch), neighborhood directories, and government domains were deliberately not registered as competitors; they appear in the data as third-party content sources.

Full methodology + open-source measurement code at /methodology/.

Source-type distribution (cohort-wide)

Across all 41 firms and all 7 AI tools, 5,809 total citations, AI pulled answers from these source types:

This puts Hawaii CPA firms at the lower end of the cross-category cluster, tied with law firms at 39% firm-owned. For comparison: Hawaii consumer banking is 53%, Hawaii wealth management 47%, Honolulu dental 44%, Hawaii law firms 39%. The professional-services cluster (wealth, dental, law, CPA) all sit between 39% and 47%, with CPA tied for the lowest. The review-directory share at 5% is also the highest of any measured category, driven primarily by Clutch (101 cites), Yelp (66 cites combined desktop and mobile), Expertise (35), and BBB (30).

Per-AI-tool breakdown, the training-data collapse

This per-engine table is the most distinctive finding in the data. The four web-searching engines (OpenAI, Gemini, Perplexity, Google AIO) all reach between 45% and 60% firm-owned share. The two training-data engines (Claude and Gemma) reach 1% and 2%. Microsoft Copilot sits at 2%, consistent with the cohort-wide Copilot pattern observed in dental, wealth, and law categories.

For comparison: in the Hawaii law-firm teardown, Gemma was the highest own-share engine at 78% and Claude was second at 51%. In Honolulu dental, Gemma was the top own-share engine. In Hawaii consumer banking, Claude reached 71%. Across every other category measured, the training-data engines knew the cohort firms by name and cited them frequently. For Hawaii CPAs, both training-data engines collapse to nearly zero.

The structural reading is that Hawaii CPA firms have substantially less brand presence in the training-data corpora than the equivalent firms in adjacent professional-services categories. CPA work is more transactional, less editorially covered, and less likely to surface in the kind of broad-web content training-data engines memorize. A firm cannot close this gap by changing what is on its website. Closing it would require category-wide editorial coverage of Hawaii CPAs in the kind of sources training data ingests (national publications, professional-association editorial, academic-adjacent content), which is outside the scope of what any individual firm controls.

Top recurring firms (anonymized)

The 5 firms AI cited most often across the 18 questions and 7 tools.

The top 5 firms account for 35% of all firm-owned mentions (789 of 2,244). For comparison: Hawaii law firms top 5 = 66%, Hawaii consumer banking top 5 = 71%, Honolulu dental top 5 = 49%, Hawaii wealth top 5 = 41%. Hawaii CPA is the least concentrated of the five categories measured. The competitive distribution is flatter: more firms with meaningful mention counts, no single dominant firm pulling away.

All 5 top firms appeared in all 3 measurement runs (consistency signal, not run-to-run noise). The remaining 36 firms in the cohort have meaningful mention counts but at a noticeably lower frequency. The structural reading: for a firm outside the top 5, the gap to the leader is narrower than in any other category we have measured. There is no dominant tier to break into; the head of the distribution is shallow.

Where AI pulls from when it cites non-firm content

The 3,133 third-party-content mentions are not all the same shape. Top recurring sources across runs:

The Hawaii Society of CPAs (hscpa.org) appearing 130 times across 3 runs is the most structurally significant third-party source. For a buyer asking AI about Hawaii CPA firms, the state professional association is cited more often than any individual firm's own website. Editorial directories (Clutch, Expertise) and review platforms (Yelp, BBB) account for the next tier. Generic Hawaii reference sites (gohawaii.com, hawaii.com, kaimukihawaii.com) appear frequently because AI often grounds Hawaii-specific queries in state-level sources before naming individual firms. A firm not represented in HSCPA's public directory is effectively invisible across roughly 6% of all citations in the data.

What this teardown does and does not prove

What it does support:

• The 41-firm cohort is a representative slice of Hawaii CPA firms AI tools actually cite for buyer-shaped questions.
• The 39%/54% firm-own / third-party split is stable across the expanded cohort and all 3 runs.
• The training-data engine collapse (Claude 1%, Gemma 2%) is the most distinctive structural pattern, unique to this category among the five measured.
• The four web-searching engines reach between 45% and 60% firm-owned share, the highest engine-cluster reach we have observed for a professional-services category.
• The top 5 concentration (35% of mentions) is the lowest of the five categories measured; the competitive distribution is flatter than in adjacent professional services.
• HSCPA (the state professional association) and Clutch (B2B review directory) are measurable AEO surfaces for this category that most firms likely treat as a separate practice-development effort, not an AEO one.

What it does not yet support:

• That the training-data collapse is permanent. AI training data refreshes on schedules outside our control. A surge in editorial coverage of Hawaii CPA firms could shift Claude and Gemma's awareness. The monthly memo cadence is how we would observe that.
• That changing a firm's content, directory presence, or third-party listings would cause AI to cite differently. We measured what AI cites. Causation requires pre-registered experiments. Different scope.
• That the closable ground inside the four web-searching engines is actually closable for any specific firm. What conditions move web-searching citation rates is its own measurement question and varies by engine.

Why this is anonymized

None of the 41 firms in this cohort are paying NeverRanked customers. The non-customer anonymization rule applies: counts, distributions, source-type breakdowns, and per-AI-tool numbers are public; individual firm names are not. The pattern is what is informative on a public surface. The named cohort lives only inside paid engagement deliverables, where the named firm is the customer authorizing the use.

A firm that becomes a NeverRanked customer gets a 1:1 deliverable that names every firm in the cohort, names the queries the customer is missing on, and ranks the closable conditions. That deliverable is private to the customer.