The recommendation for your category differs by geography. The data shows why.
We measured 7 AI tools across 37 Austin CPA firms with the same hash-locked methodology we use for Hawaii categories. The cross-geo data revealed that the original Hawaii CPA training-data finding splits: Claude's collapse generalizes, Gemma's does not. For an Austin firm, this changes which AI engines are worth investing in.
The cross-geo picture, side by side
| Engine | Austin CPA own-share | Hawaii CPA own-share | What it means for an Austin firm |
|---|---|---|---|
| Perplexity | 61% | 46% | Strong reach. On-site work and editorial both move it |
| Gemini grounded | 61% | 59% | Strong reach. Similar to Perplexity in Austin |
| Google AI Overviews | 50% | 45% | Half the citations go to firm sites. The other half to GBP + reviews |
| ChatGPT search | 45% | 60% | Notable Austin/Hawaii asymmetry. Hawaii ranks higher here |
| Microsoft Copilot | 0% | 2% | Universal pattern. First-mover opening on Bing organic |
| Gemma (training data) | 23% | 2% | Reachable for Austin firms. Not reachable for Hawaii firms. |
| Claude (training data) | 0% | 1% | Category-wide blind spot. Collapse holds across geos |
For an Austin CPA firm specifically, the competitive surface is broader than for a Hawaii CPA firm. The four web-searching engines are the primary closable ground (40-61% reach). Gemma is also reachable through sustained editorial presence on Texas-business-news content that Gemma's training cycle ingests. Claude is a blind spot, as it is for Hawaii CPAs. Microsoft Copilot's universal cohort-wide gap is the first-mover opening, same as every other category we have measured.
Why we measured Austin (the cross-geo question)
The Hawaii CPA teardown surfaced what looked like a category-level pattern: training-data AI engines cite Hawaii CPA firms at near-zero rates while web-searching engines reach 45-60%. The original framing called this a "training-data engine collapse" for CPA firms. The natural next question: does the collapse generalize outside Hawaii, or is it Hawaii-specific?
We measured Austin TX as the first cross-geo test because it is a comparable mid-size professional-services market with a direct structural analog (Texas Society of CPAs vs Hawaii Society of CPAs). Same 18-question structural shape, with geographic terms swapped from Hawaii to Austin. Same 7 AI engines, same 3 reps, same hash-locked discipline.
The result split the original Hawaii pattern into two distinct findings. Claude's collapse holds in both geographies (1% Hawaii, 0% Austin). Gemma's collapse does not (2% Hawaii, 23% Austin). One engine produced a category-level pattern. The other produced a geo-specific pattern. This is the kind of distinction that requires cross-geo measurement to surface. A vendor measuring one geography per category would have published the Hawaii result and called it the category-level pattern.
What an engagement looks like for an Austin CPA firm
The standard engagement shape, scoped to an Austin CPA firm:
- Scoping call (30 min). Lock the 18 buyer questions a real Austin CPA buyer would ask AI. Lock the cohort (the firms you compete with for buyer attention, including specialty firms in your service variants). Confirm the questions reflect your specific practice mix (tax planning, audit, advisory, fractional CFO, etc.).
- Three-week kickoff. Measurement across all 7 AI tools, hash-locked, repeated for stability.
- Research memo + prepped punch list. Named competitors, observed gaps per query, per-query playbooks for the weakest queries, schema and page-template recommendations for what your team or agency ships, agency-ready SOW scope per move. The Austin-specific recommendation differs from the Hawaii CPA recommendation in one specific way: Gemma is treated as a reachable engine for Austin firms (23% own-share suggests sustained editorial presence does move the needle on training-data-engine recognition over time).
- Monthly delta memo after the kickoff. What moved, what did not, drift alerts when a competitor moves on your category.
The punch list weights toward the four web-searching engines (Perplexity 61%, Gemini 61%, Google AIO 50%, OpenAI 45%) as the primary closable surface, plus the Microsoft Copilot first-mover push via Bing organic. Gemma is treated as a slower-but-reachable surface that responds to sustained editorial presence in Texas-business-news content over months-to-quarters. Claude is treated as a category-wide blind spot not worth direct investment for any single firm in the short term.
See how 7 AI tools read your firm’s site right now
Free check. Enter your URL and we will show you what AI tools can read from your site, what they cannot, and what is making you harder to find. No signup.
Run the AI search check Get the 1-page diagnosticFree check is automated and instant. The 1-page diagnostic is hand-built and shows which competitor firms AI is naming for your specific buyer questions in Austin.
Honest scope
- 37-firm cohort, expanded across 4 cohort-coverage passes (5 anchors with 1 domain correction, then 16 from run #1, 8 from run #2, 8 from run #3). The full firm list is anonymized in the public teardown (non-customer anonymization rule).
- Austin-specific. National Big-Four and large national or regional firms are excluded from the cohort to keep the comparison Austin-specific. They appear as third-party content occasionally.
- Measurement-only. We do not write content, edit pages, claim profiles, or change your site. Your team or your agency does. That separation is structural and is the engagement.
- The cross-geo finding is n=2 (Hawaii + Austin). Medium confidence. A third measured CPA geo would strengthen or break the cross-geo pattern. The Claude finding (collapse generalizes) is the stronger of the two. The Gemma finding (geo-specific) is the more surprising and would benefit most from a third data point.
NeverRanked home · Full Austin CPA teardown · Hawaii CPA teardown (for comparison) · Cross-category two-axis framework · How we measure