Data report · updated as new leaderboards land
The State of AI Search for Local Business
Before customers call anyone, they ask an AI assistant who’s best. We ran the questions they ask, unprimed, across ChatGPT and Gemini, and recorded who those assistants actually recommend. This is what 840 leaderboards covering 21 local-service trades and 40 U.S. cities reveal about how AI picks winners, and how open those answers still are.
Dataset window June 18, 2026 to August 24, 2026
Finding 01
AI answers are wide open. Nobody owns them.
In an average market, the single most-recommended business captured only 24% of all AI mentions. The assistants spread recommendations across roughly 10 businesses per question set. Unlike Google’s map pack, there is no entrenched “#1”, which means the answers are still winnable.
Finding 02
AI recommends local, not national.
Of the 11,678 distinct businesses the assistants named, 90% appeared in only one city’s answers. National chains show up, but they don’t dominate. A strong local reputation still beats brand recognition in AI recommendations.
- 01Orkin938 mentions
- 02Terminix894 mentions
- 03Two Men and a Truck798 mentions
- 04The Cleaning Authority734 mentions
- 05Roto-Rooter Plumbing & Water Cleanup495 mentions
- 06H&R Block494 mentions
- 07A1 Garage Door Service415 mentions
- 08The Joint Chiropractic339 mentions
Finding 03
Some trades are locked up. Others are up for grabs.
“Leader share” is how much of a market’s AI mentions its top business captures. Higher share means AI concentrates on a clear favorite; lower share means the field is fragmented and open. Here’s where each end of the spectrum sits.
- Chiropractors31% leader share
- Locksmiths30% leader share
- Dentists28% leader share
- Real Estate Agents26% leader share
- Moving Companies26% leader share
- Salons & Barbershops21% leader share
- Landscapers21% leader share
- Auto Repair Shops22% leader share
- Gyms & Fitness Studios22% leader share
- HVAC Companies23% leader share
Methodology
How we measured this
- Real answers, not surveys. Every figure is computed from 2,261 stored measurements of 840 leaderboards in our production database. Nothing here is estimated or illustrative.
- Unprimed prompting. Each of 5 real buyer questions per market (e.g. “best plumber in ___”) was asked across 2 assistants, ChatGPT and Gemini, without naming any business, so results reflect what the models recommend on their own. That is 22,610 recorded answers. Claude and Perplexity are not in this corpus; your own free scan adds both.
- Coverage. 21 local-service trades across 40 U.S. cities. Markets are re-measured on a rolling cycle rather than all at once, so every leaderboard page carries the date it was last measured, and these numbers update as the corpus grows.
- Limits we’re honest about. Mention counts are aggregated across the 2 assistants, so this report does not break out per-engine visibility. Business names come from model output and are normalized but not manually deduplicated, so distinct-business counts are a close estimate, not an audited registry.
Is AI recommending you?
Run the same measurement on your own business, free, across all four assistants. See whether ChatGPT, Claude, Gemini and Perplexity name you when your customers ask, and exactly how to get named when they don’t.
Run my free scan →