The 2026 AI Recommendation Study

Five headline findings

• Across 100 live category searches (Gemini 2.5 Flash with Google Search grounding), the engine named an average of 12.53 businesses per query (range 4-44) and grounded 100% of answers in real web sources.
• 569 unique source domains were cited in total (770 citations), but citations concentrate hard: only 35 domains were cited across 3+ different searches, while 534 appeared just once or twice.
• The most-cited domains are third-party directory, ranking and review platforms — the top five were bestlawfirms.com, reddit.com, forbes.com, superlawyers.com, justia.com — not the businesses' own websites.
• Directory and ranking sites were cited in 78.0% of answers and review/community sites (led by Reddit) in 21.0%; the strongest lever to get named is being listed and reviewed on those platforms, not just your own site.
• The single most-cited domain was bestlawfirms.com, appearing in 16 of 100 answers.

What we measured

We sampled 100 vertical-and-city combinations evenly across 10 service verticals — chiropractors, cosmetic dentists, dental implant clinics, HVAC companies, law firms, med spas, personal injury lawyers, plastic surgeons, real estate agents, and veterinary clinics — and 25 US cities. For each, we sent one natural buyer prompt: "What are the best [vertical] in [city]? List specific businesses and cite your sources."

Every response was logged verbatim. We recorded the businesses the engine named and, critically, the real source domains behind the answer — pulled from Gemini's grounding metadata, which lists the web pages the model actually retrieved. This is not a survey. It is the engine's live output, captured and counted.

Finding 1: a handful of third-party platforms carry the citations

The clearest pattern in the citation data is concentration. Of 569 distinct domains cited across the study, only 35 were cited across three or more different searches; the other 534 appeared just once or twice. Those few repeat-cited domains are almost entirely directories, ranking lists, and review platforms — bestlawfirms.com, reddit.com, forbes.com, superlawyers.com, justia.com — not individual businesses' own sites.

Directory and ranking sites were cited in 78.0% of answers (78 of 100) and review/community sites in 21.0%. Individual business websites do get cited, but almost always in that long tail — each appearing in a single answer — so no business's own site rises to the level of a top source. The takeaway is blunt: getting AI to name you is largely about getting listed and reviewed on the platforms AI already trusts.

Source types are bucketed from the raw cited domains by rule; a single answer can cite several types, so shares do not sum to 100%.

Finding 2: the domains AI cites most

Ranked by how many of the 100 answers cited them at least once, these are the domains doing the heaviest lifting in AI recommendations for local service categories.

Finding 3: a small set of businesses recur across markets

Counting how often each specific business was named across all 100 searches, a shortlist of recognizable, well-corroborated names recurs — the businesses with the broadest, most consistent web footprints.

Generic-sounding names that recur verbatim across unrelated cities can indicate the engine generating a plausible label rather than a verified business; we report the raw counts without smoothing.

Per-vertical breakdown

Methodology

What this means if you run a business

If AI leans on directories, review sites, and community threads more than on your own website, then optimizing only your website is optimizing the wrong surface. The compounding work is earning consistent, accurate mentions on the third-party sources the model already trusts — the exact signals we build and measure when we run an audit.

See the companion dataset, the State of AI Visibility 2026, for how many names the shortlist holds, and the AI visibility glossary for the mechanics behind every term here. Then run your own category live with our free AI Visibility Finder.