Methodology
The initial benchmark will use a predefined panel of unbranded discovery questions, repeated observations across multiple AI platforms, and documented testing conditions. It will report how often sampled practices appear, how results vary, and which sources are cited when available. The final report will disclose the prompt panel, observation period, sample sizes, scoring rules, and limitations.
Version 1 will use a predefined panel of unbranded local-business discovery questions. The panel will cover several relevant prompt families and phrasings selected before testing begins. The final report will publish the panel and explain how questions were selected.
Questions will be observed repeatedly over a documented observation period and across more than one AI platform. Testing conditions will be kept consistent and documented where reasonably controllable. The report will identify observation dates and explain material differences in platform behavior or available features.
The benchmark will measure recommendation or qualifying-mention frequency and the variation and consistency of practice appearances across observations. When an AI platform provides cited websites or sources, those citations will also be recorded and summarized.
Reported figures will include their sample sizes and observation dates. The final report will disclose the methodology, scoring rules, and limitations needed to interpret the findings. Exact sample sizes, platform coverage, repetition counts, and testing schedule will not be published until those decisions are finalized.
Testing records will be preserved for the documented observation period, including the questions submitted, observed responses, available citations, observation dates, and the scoring decisions applied. The benchmark will not promise that any platform will recommend any business.
Questions about the methodology? hello@fireflyclover.com