A running benchmark of what AI actually recommends when real people ask it to name a business, product, or service. We ask the engines. We record what they say. Nothing is estimated.
Everyone is talking about AI search. Almost no one has measured what it actually recommends. So we did — asking live AI engines the same "what's the best…" questions a customer would, and recording, word for word, who got named. This is the inaugural, pilot edition of a benchmark we'll run and grow over time.
Across 12 queries, the engine named businesses in 11 of them — a median of five, and as few as two or three for local services. If you're not on the shortlist, there is no page two to be found on.
Every local query — the lawyer, the jeweler, the HVAC company — returned a geo-personalized "Places" module, and the businesses AI recommended were the ones with the highest star ratings and the most reviews. It said so out loud. Your review profile is now a direct input to whether AI names you.
And AI personalized by location even when we never gave a city — inferring the market from the session and localizing the lawyer, jeweler and HVAC results automatically. If you serve an area, AI is already deciding whether you're "near me."
The recommendations were sourced overwhelmingly to independent review sites, roundups, and forums — not the brands' own homepages. AI didn't recommend the best marketing copy; it recommended the companies other trusted sources already talk about.
The one ambiguous prompt — "best AI marketing agency for a small business" — returned zero named businesses and only a list of selection criteria.
When a category is unclear, AI declines to pick a winner — which means it can't pick you. Ambiguity is invisibility. Being unmistakably categorized is a prerequisite for being recommended at all.
We re-ran five queries on Google's AI Overviews and compared. For established product categories, the two engines shared their top picks. For a fragmented category like franchises, they barely overlapped — so you have to earn each engine separately.
Categorize yourself so clearly that AI never has to hedge about what you are.
Reviews and a clear service area are now direct inputs to the recommendation.
The recommendation is earned off your own site — in the sources AI trusts.
Ranking on Google was about being on the list. Winning in AI is about being the recommendation — and this data shows exactly where that's decided.
Method. 12 natural "recommend me…" prompts run through Perplexity in a single U.S. session on July 30, 2026, with five re-run through Google's AI Overviews for cross-engine comparison. Each result logged verbatim: businesses named, citations, entities. No number is estimated or invented.
Limits. This is a pilot — a small sample, two engines, one location, one point in time. It establishes the method and shows clear, directional patterns. Future editions of the Index scale the sample, add ChatGPT and Gemini, and test multiple markets — published openly, limits and all.