Research archive / July 2026
AI answers vary.
Study what repeats.
An independent study of cosmetic-dentistry recommendations in Philadelphia and the Main Line. This is an example of measurement and quantitative evaluation, not a client implementation.
- 18
- Predetermined discovery questions
- 4
- Monitored AI search platforms
- 7
- Dates of observation
How the study worked
From July 25 to 31, 2026, I observed responses to the same unbranded discovery questions across ChatGPT, Google, Perplexity, and Copilot using Otterly as the collection platform. Questions covered Philadelphia, the Main Line, Delaware County, and nearby western suburbs. These monitored outputs are not assumed to be identical to every consumer interface.
Of 504 scheduled discovery responses, one failed, leaving 503 eligible observations. I retained response text, available citations, dates, and reviewed classifications. A strong recommendation or viable option counted; a passing mention did not. Each provider received at most one credit per eligible response.
What the reviewed record contains
The final reviewed dataset contains 3,510 provider recommendation credits across 495 analytical provider identities. Repeated observations support comparisons across questions, platforms, and dates. Those counts describe this sample, not a permanent ranking.
What this demonstrates
A defined test panel, repeat observations, traceable classifications, and a reconciled denominator make a result inspectable. The same discipline matters when deciding whether an AI system is reliable enough for operational use.
Limits of the evidence
The study does not explain why a platform recommended a practice or predict future answers. AI search visibility does not measure clinical quality, patient outcomes, demand, leads, or revenue. The results are limited to the tested questions, conditions, platforms, and dates.
This research is archived. Firefly Clover no longer offers dental visibility services or private practice findings.