AEOGeoAI Research Report · July 2026
Pennsylvania AI SearchVisibility Study 2026
An Independent Analysis of AI Citation Patterns Across 163 Pennsylvania Health Practices.
An Independent Analysis of AI Citation Patterns Across 163 Pennsylvania Health Practices.
We analysed 163 independent Pennsylvania health practices to measure whether they appear in AI-generated local recommendations across ChatGPT, Claude and Gemini using the query format: "best [category] in [city] PA". 97.5% showed no genuine citation — but critically, this was rarely because AI had nothing to say. In 96.3% of cases, AI discussed the correct category and often named specific competitors — just never the tested practice.
AI almost never failed to answer local health questions. Instead, it answered them by naming competitors while omitting the tested practice.
The competitors that do get named are overwhelmingly large hospital systems and national retail chains — not comparable independent practices. See Section 3.
This study measures AI citation presence — defined as whether a practice is:
This study measures AI citation presence rather than Google rankings, website quality, reputation, or business tenure.
Each practice was queried individually across three AI systems using a standardised prompt. We tested whether the AI system included the named practice in its recommendation-style answer.
"best [category] in [city] PA"
Each practice category was reduced to its primary field — e.g. "Chiropractic / Rehabilitation" → "chiropractic". Example: a Horsham PA chiropractic practice received the query "best chiropractic in Horsham PA".
A binary name-match alone treats "AI discussed the right category and named three competitors, but not this practice" identically to "AI has never heard of this business or its category" — both would score zero. This study distinguishes the two. A raw score of 35 or higher indicates the practice's name was actually matched in the response — a genuine citation. Scores of 1–34 indicate category-adjacency: the right topic and location were discussed, and often specific competitors were named, but not the tested practice. A score of 0 means no relevant signal was found at all — not even the category or competitors.
| Score | Meaning |
|---|---|
| 0 | No relevant signal — category not discussed |
| 1–34 | Category-adjacent — topic/competitors discussed, practice not named |
| 35–59 | Genuine citation — practice named, limited prominence |
| 60–79 | Genuine citation — consistently included |
| 80–100 | Genuine citation — prominently featured (none observed) |
| Parameter | Value |
|---|---|
| Sample size | 163 practices |
| Geography | Southeastern Pennsylvania — Montgomery County corridor (Horsham, Ambler, Collegeville, Lansdale, Blue Bell, and surrounding areas) |
| Categories | Dentistry, chiropractic, physical therapy, medical spa, ophthalmology/optometry, mental health/behavioral, orthopaedics, pediatrics, others |
| Test period | July 2026 |
| Rate control | 4-second delay per query |
| Tooling | AEOGeoAI visibility scoring system with secondary classifier for category-adjacency and competitor extraction |
Limitations: AI outputs are probabilistic and may vary across time and model updates. Results represent a snapshot of observed behaviour in July 2026 and should not be interpreted as permanent or universal. One record (0.6% of the sample) contained a malformed category field that produced a corrupted query string; its result is excluded from category-level analysis but retained in the overall count.
| Outcome | Meaning | Practices |
|---|---|---|
| True zero | No relevant signal — category not discussed at all | 2 (1.2%) |
| Category-adjacent, unnamed | Category and often competitors discussed — practice never named | 157 (96.3%) |
| Genuinely named | The practice itself was named by at least one model | 4 (2.5%) |
The 96.3% figure is the sharper finding. These practices aren't falling outside AI's field of view — the opposite is true. AI systems are actively discussing their exact category and, in many cases, naming their competitors in the same response. The practice itself is simply never the name that gets surfaced.
Four practices held a genuine, direct name citation. One — Lansdale Dental — was named by two models simultaneously, the only cross-model citation observed in this dataset.
| Practice | Category | City | Claude | Gemini | ChatGPT |
|---|---|---|---|---|---|
| Lansdale Dental | Dentist | Lansdale | 10 | 85 | 50 |
| Ambler Dental Care | Dentist | Ambler | 10 | 75 | 21 |
| Blue Bell Dental Associates | Dentists | Blue Bell | 10 | 60 | 21 |
| Horsham Chiropractic | Chiropractic | Horsham | 9 | 0 | 50 |
All four genuine citations occurred in dentistry or chiropractic — the two highest-volume categories in this dataset (58 and 19 practices tested respectively) — suggesting citation likelihood may correlate with category density rather than practice quality.
Genuine-citation absence rate by category (share of practices with no genuine name citation on any model):
| Category | Tested | No genuine citation | Rate |
|---|---|---|---|
| Dentistry | 58 | 55 | 95% |
| Other specialties | 39 | 39 | 100% |
| Chiropractic | 19 | 18 | 95% |
| Mental health / Behavioral | 19 | 19 | 100% |
| Physical therapy | 11 | 11 | 100% |
| Ophthalmology / Optometry | 8 | 8 | 100% |
| Medical spa / Aesthetics | 5 | 5 | 100% |
| Orthopaedics / Sports medicine | 3 | 3 | 100% |
| Pediatrics | 1 | 1 | 100% |
Because our detection method captures category-adjacent responses rather than discarding them, we can extract exactly which businesses AI names in place of the tested practice — using only names that actually appeared in the model's response, not inferred or invented data.
| Business named | Type | Times cited |
|---|---|---|
| Main Line Health | Hospital system | 9 |
| Penn Medicine | Hospital system | 6 |
| Abington Health | Hospital system | 6 |
| Psychology Today | Directory platform | 6 |
| Lansdale Hospital | Hospital | 5 |
| Abington Hospital–Jefferson Health | Hospital system | 5 |
| LensCrafters | National retail chain | 5 |
| Pearle Vision | National retail chain | 4 |
| Visionworks | National retail chain | 4 |
AI defaults to institutions, not independents. When an independent practice isn't named, AI's fallback is almost never "a different independent competitor." It's a hospital system, a national chain, or in mental health specifically, a directory platform (Psychology Today) rather than any named provider at all. For an independent practice, the competition for an AI citation isn't really the practice down the street — it's Main Line Health.
Where AI names a specific alternative, the qualities it associates with them cluster around a small set of themes:
| Attribute | Times mentioned |
|---|---|
| Friendly staff | 19 |
| Gentle approach | 13 |
| Personalized care | 11 |
| Comprehensive services / care | 14 (combined) |
| Multiple locations | 7 |
| Patient-centered approach / care | 12 (combined) |
| Modern technology / facilities | 8 (combined) |
None of these are differentiators unique to large institutions — "friendly staff" and "personalized care" are exactly the kind of claims an independent practice's own website already makes. The gap isn't in what independent practices offer; it's in whether AI has independent, third-party confirmation of it.
AI systems construct local recommendations by assembling entities from indexed third-party sources. Inclusion requires:
Most practices in this dataset lack sufficient third-party entity signals for inclusion in AI recommendations — independent of Google rankings, website quality, reputation, and tenure.
Signal gap vs ranking gap. This is not a ranking problem. It is an entity visibility problem — AI systems surface entities with sufficient external confirmation signals, and hospital systems and national chains simply have vastly more of that confirmation than any independent practice, regardless of care quality.
Only one practice (Lansdale Dental) achieved citation on more than one model, indicating that even genuine citation is largely model-specific rather than a stable, transferable form of entity authority.
All four genuine citations occurred in the two categories with the largest sample sizes (dentistry, chiropractic). Whether this reflects a real effect (more indexed dental/chiropractic content generally) or a sampling artifact of this dataset's category mix is a question for future research with balanced category sizes.
Using the same query methodology and detection thresholds, we can compare this Pennsylvania dataset directly against our earlier 216-practice New Jersey study.
| Metric | New Jersey (n=216) | Pennsylvania (n=163) |
|---|---|---|
| True zero (no signal at all) | 0.9% | 1.2% |
| Category-adjacent, unnamed | 98.6% | 96.3% |
| Genuinely named (1+ model) | 0.5% | 2.5% |
| Genuine cross-model citation | 0 practices | 1 practice |
The overall pattern replicates closely across two independent states tested a month apart: roughly 97–99% of independent health practices have no genuine AI citation, and the near-total majority of that group is category-adjacent rather than genuinely invisible. This consistency across states — rather than a single-state anomaly — is what suggests a structural feature of how AI systems build local recommendations, not a New Jersey- or Pennsylvania-specific quirk. That said, the finding has only been tested in these two states; broader geographic generalisation would require additional data.
The two states are not identical, however — Pennsylvania showed a higher genuine-citation rate (2.5% vs 0.5%) and produced this study's only cross-model citation, where New Jersey had none. With two data points, it isn't yet possible to say whether this reflects a real difference in publication density between the two markets or ordinary sample variation at this scale.
Practices without AI citation presence risk reduced discovery in AI-native search, and increasing dependence on traditional SEO channels as AI-driven local discovery grows — while their most likely "competitor" in an AI answer is a large institution with a permanent structural advantage in third-party coverage.
Based on the observed citation patterns and inspection of the cited entities in Section 3, practices that were genuinely cited tended to have broader third-party representation, including:
This is our interpretation of the pattern, not a factor we directly measured or isolated in this dataset.
Third-party citation signals increase the likelihood of AI citation presence but do not guarantee inclusion in any specific AI-generated answer. AI model outputs are probabilistic and change over time.
We publish structured entity articles about PA health practices on verified local publications already indexed by AI systems — creating the third-party citation signals AI relies on when constructing recommendation responses.
View PA service and pricing →From $199 · One-time publication · No retainer · Montgomery County corridor