Methodology · Original data · August 2026

Five-Market Dental Reputation and AI-Visibility Analysis: Full Methodology

Published August 3, 2026 · Data collected August 3, 2026 · Companion to How to Get More Dental Patients

This page documents, permanently and in full, how CustomerHalo's five-market dental study was conducted: the queries run, how competitors were selected, what counted as an appearance, when data was collected, which fields were collected, and the anonymized market-level results. The findings themselves are summarized in the guide's benchmarks section. Questions and corrections: [email protected].

Study design

Five independent general dental practices — one in each of the Denver, Atlanta, Columbus, Tampa, and Phoenix metropolitan areas — were analyzed with CustomerHalo's standard production pipeline on August 3, 2026. The subjects were a convenience sample chosen by the study team from ordinary public Google Maps listings: independent (non-chain) general practices with active Google Business Profiles and their own websites. Selection was not randomized, subject practices were not involved or notified, and no non-public information about any practice was collected or used.

Anonymity policy. Subject practices are reported as Markets A–E. The letters are deliberately not in the metro order listed above (and not alphabetical by metro), so that per-market metrics cannot be trivially mapped to a metro. The single subject-practice appearance in an implant-related answer is reported in aggregate and is not attributed to a lettered market, for the same reason. Limit of this policy: all metrics here are drawn from public listings, so a sufficiently determined reader could in principle re-identify practices by exhaustively searching public data; we judged this acceptable because nothing non-public is disclosed and no practice is evaluated negatively by name.

Timestamps

• Scans initiated: August 3, 2026, 19:01:09–19:02:42 UTC.
• Analyses completed: August 3, 2026, 19:08:34–19:13:02 UTC.
• All Google Business Profile data, reviews, and AI answers reflect that collection window. Review counts and answers change over time.

Data fields collected

• Business profile (subject): name, primary Google Business Profile category and category IDs, star rating, total review count, address, coordinates, website URL, and place reference — from public Google Business Profile data via a commercial local-data provider (DataForSEO).
• Subject reviews: the newest reviews, newest-first, up to a requested depth of 300 per practice — date, star rating, and text. Reviewer identities were not used in the analysis.
• Competitor listings and reviews: name, category IDs, review count, and place reference for each selected competitor, plus that competitor's newest 200 reviews.
• Subject website: pages fetched from the practice's own site to identify advertised services.
• AI-answer records: for each query — the query text, intent type, engine identifier, outcome classification, whether the subject was mentioned and/or cited, the domains cited in the answer with a source-type classification, and the businesses named in the answer.

Competitor-selection criteria

Comparison groups were selected automatically by the production pipeline, with no manual curation:

1. Candidate pool: Google Business listings sharing the subject's category IDs within 25 km of the subject's coordinates, with more than 20 reviews, ordered by review count descending, up to 50 candidates.
2. Filters: the subject's own listing and any listing whose website matches the subject's domain are excluded; candidates need at least 40 reviews; recognizable national chains are excluded; duplicate business names are collapsed to the highest-reviewed listing; names appearing three or more times in the pool (multi-location brands) are excluded.
3. Ranking: candidates whose primary category matches the subject's primary category are preferred; up to four competitors are retained per market.
4. Review depth: the newest 200 reviews were retrieved for each selected competitor.

In one of the five markets, no candidates satisfied all criteria, so that market has no comparison group; findings involving comparison groups therefore cover four markets and 16 competitors in total.

The queries

Nine live queries were run per market (45 total), constructed by the production pipeline: one entity query (the practice's name plus its city — a control for whether the engine can find the practice at all), one category query ("best dentist in {city}"), three service buyer queries built from services the practice advertises on its own website, and four focus queries — buyer-phrasing variants and problem phrasings for the focus service (dental implants), generated in patient language. The 25 implant-related buyer searches reported in the guide are, per market, the "dental implants in {city}" service query plus the four focus queries (marked * below). Entity queries are redacted to preserve subject anonymity; all other queries are shown verbatim.

* One of the 25 implant-related buyer searches.

Engine, geographic context, and appearance definitions

• Engine: Google's Gemini with Google Search grounding (pipeline engine identifier gemini_grounded_google_search). One run per query. AI answers are not deterministic; a re-run will produce different answers.
• Geographic context: supplied entirely in the query text ("… in Phoenix, AZ"). No IP-based or profile-based location simulation was used.
• Outcome classification per answer: cited (subject named and its website cited as a source), mentioned (subject named without citation), competitors_only (other businesses named, subject absent), no_businesses (the answer named no businesses).
• "Appearance" means the subject practice was named in the generated answer — i.e., an outcome of cited or mentioned.

Anonymized market-level results

Reading the recency columns: "Service-evidence instances" counts reviews the pipeline's classifier attributed to a specific named service (a single review can qualify for more than one service, so instances can exceed unique reviews). The share column is the percentage of those instances dated within the 12 months before analysis. This is the recency metric reported in the guide. "Reviews retrieved" can be below the listing's total when the provider returns fewer usable reviews than the listing reports; for the two practices above 300 total reviews, retrieval was capped at the newest 300.

Visibility outcomes, in aggregate

• Entity control: all five subject practices were cited when asked about by name — 5 of 5. The engine can find them; the question is whether they surface when patients don't already know the name.
• 25 implant-related buyer searches: subject practices appeared in exactly 1 of 25 (a focus-variant query in one market; not attributed per the anonymity policy).
• "Best dentist in {city}": subjects appeared in 0 of 5.
• Non-implant service searches: subjects appeared in 0 of 10; 2 of these 10 answers named no businesses at all.
• Highest-reviewed competitors: in three of the four markets with comparison groups, the market's highest-reviewed comparable practice appeared in none of that market's implant-related answers; in the fourth market it appeared in one. Two additional comparison-group members in one market appeared in that market's "best dentist" answer.
• What did appear: 65 distinct businesses were named across the 25 implant-related answers (116 across all 45 answers), dominated by implant-focused and oral-surgery providers; the financing platform CareCredit was cited in 5 of the 25 implant-related answers. Booking platforms (Zocdoc, Opencare) were cited in the broader "best dentist" and general service answers rather than the implant answers.

Limitations

• Five practices in five metros is an illustrative snapshot, not a representative sample; subjects were a convenience sample of highly rated practices (4.9–5.0 stars).
• One AI engine, one run per query, on a single date. AI answers vary between runs, engines, and dates; specific appearances should be treated as examples, not stable rankings.
• Service attribution of reviews is performed by an automated classifier and can contain errors; the same classifier was applied uniformly to all practices.
• Competitor selection is algorithmic (criteria above) and may omit practices a patient would consider; in one market it produced no comparison group.
• Review counts, ratings, and answers all drift over time; this page describes August 3, 2026.

Reproducibility

The method above is complete: the queries (except redacted entity controls), engine, geographic scoping, selection criteria, thresholds, and metric definitions are stated exactly as the pipeline executed them. Because grounded AI answers are non-deterministic and public review data drifts daily, a re-run will not reproduce the specific answers — but it will test the same claim the same way. If you find an error in this methodology or its reporting, email [email protected]; the guide's corrections policy applies to this page.

© 2026 CustomerHalo, Inc. · Home · Back to the guide · Terms · Privacy