Plexa Index methodology

The index does not say "AI recommends X." It says "X appears in N% of its category's answers." It is share of voice: how often AI names each business when asked about a local service. It is not a quality ranking or an award; it is what the models say today, counted honestly.

Today we measure two surfaces: ChatGPT and Google AI Overviews. Google AI is queried as AI Overviews through a monitoring API; the measurement is faithful to the engine's behavior, not an exact replica of every individual session.

Not yet: Perplexity, Gemini, or Meta AI. When a surface enters the measurement, it is declared here and in the corresponding edition. We do not report an engine we did not run: the list above is the complete list.

For each combo (vertical × city) we run 10 to 15 real questions — the ones a customer would ask, not the ones that flatter us. The question set is deterministic: the category defines it, not a language model. The same battery runs identically in every wave, so waves are comparable to one another.

A mention is when the AI answer identifiably names a business that is in the Plexa directory. If the answer names a business not in the directory, or names none, that answer counts as "no directory mention" — what we call the silence. We count at the answer level, not per session or per click.

A business's AI Visibility Score = the fraction of its category's answers it appears in, ×100. A business that appears in half of the answers for its vertical and city has a Score of 50.

The silence is measured at the answer level (answers that name no business from the directory), not as a "% of invisible businesses." The monitoring sees whom AI cites, not the full universe of businesses that exist; that is why we do not invent a denominator of the invisible. The figure we report is the one the data supports, not one more.

Canonical definition of silence: numerator = answers where no directory business is resolved; denominator = all answers measured in the window. "Resolved" means the name AI said was matched to a directory business after normalizing accents, symbols, and city or descriptor suffixes —by equality or by distinctive containment, never by a generic category word ("dental clinic," "auto repair"), so it neither under- nor over-counts. If the matching changes, the number changes: that is why the definition lives here, and each edition keeps its own.

The current edition covers 25 combos (vertical × city) across 8 verticals and 11 cities. Within the window — between June 14, 2026 and July 14, 2026 — we measured 922 AI answers and recorded 841 mentions of 214 distinct businesses from the directory. Each market was measured in one or more runs across the window; the number of answers per market reflects how many times it was measured, not a difference in method.

The sample is concentrated: 50.3% of the answers (464 of 922) come from a single metropolitan area, Cabo San Lucas. The global aggregates are weighted toward that market; the study's by-city cuts let you read each market without that weighting. We state it plainly: the Plexa Index does not yet have even national coverage, and it grows market by market.

Position is not for sale: no business can pay to appear or to improve its place. The index measures organic mentions only; some engines already run paid ads, and those don't count. We do not invent data: if the monitoring didn't see it, we don't assert it.

House honesty: in this wave, Plexa itself was cited in 0.1% of answers (1). We are entrants; we measure ourselves by the same yardstick.

The per-combo rankings are recalculated periodically; each aggregate edition of the index is dated and published with a DOI on Zenodo, under a Creative Commons BY 4.0 license. The first edition lives with DOI 10.5281/zenodo.21364735. Each edition keeps its own window and sample; the figures are not rewritten in hindsight.