AI endorsement varies greatly by engine.
This study analyzed 3,234 AI answers and 10,133 citations to 72 patient questions on ChatGPT, Gemini, Perplexity and Google AI Overview, checked against 18,492 practice listings in five US metro markets: suburban Chicago, Philadelphia-South Jersey, Tampa-St. Petersburg, Phoenix West Valley and suburban Houston.
Asked whether it recommends a specific practice for a condition, ChatGPT said it could not verify the practice in 31% of 45 answers in Javvo's study, while Gemini endorsed 71% and never said it could not verify. Google AI Overview appeared on only 64% of these questions, and endorsed in 62% of the answers where it appeared.
We tested questions where the patient asks AI whether it recommends a specific practice for a given condition. Across the 135 answers, the engines behaved very differently.
ChatGPT is the skeptic. Gemini always endorses or verifies
ChatGPT, the most used AI engine among consumers, is the most skeptical: it endorses only when the evidence it sees is verifiable and specific, and in almost one out of three answers it said it could not verify the practice. Gemini endorsed the same practices warmly on the same questions and never said it could not verify. Perplexity sits between the two, endorsing over half the time.
Outcomes when an engine is asked about a named practice
- Endorsed
- Described factually
- Said it could not verify
| Engine | Endorsed | Described factually | Said it could not verify |
|---|---|---|---|
| ChatGPT | 28.9% | 40% | 31.1% |
| Perplexity | 55.6% | 40% | 4.4% |
| Gemini | 71.1% | 26.7% | 0% |
| Google AI Overview* | 62.1% | 31% | 0% |
*AI Overview appeared on 64% of these questions. Its rates cover the answers where it appeared. Remaining shares are rare negative answers.
One question, different endorsement
The same practice and the same question produce a different answer for the patient depending on the app they opened:
I couldn't verify a strong, specific reputation for [Practice] on its own for this condition.
Yes — [Practice] appears to be a good fit for plantar fasciitis.
[Practice] emerges as a promising option.
Each engine reads a different mix of a practice's online presence, and weighs it differently
The differences are not confined to direct questions about a practice. Across every question type in the study, each engine built its endorsements from a different slice of the practice's online presence. ChatGPT leans on review-site ratings and hedges when it cannot verify them. Gemini reads reviews and practice profiles and paraphrases their sentiment into warm recommendations. Perplexity compiles ratings and factual data into its answers. Google AI Overview draws mostly on practice websites. A practice that looks strong on one surface and thin on another might get a different assessment from different engines, which is why getting recommended consistently means getting the whole online presence right.
- Previous finding: AI almost never speaks negatively about a practice.
- Next finding: Four AI engines, four personalities.
- Read all findings → How Javvo helps clients →
See how AI answers questions about your practice today.
Get your free AI Visibility SnapshotRun AI visibility for the practices you serve.
AEO for partnersAugust 2026 edition.