RankNext Research · living dataset
The Local AI Index: who does AI recommend for home services?
Every day we ask AI engines the questions American homeowners actually ask — “what are the best HVAC company options in Dallas?” — across 20 trades and 20 metros, store every answer as a receipt, and measure it deterministically. No survey, no vibes: the actual answers.
Current sample: 2678 captured answers · engine: DeepSeek · window: Jul 2 – Aug 15, 2026 · grows daily
94%
of answers name specific businesses
Not advice — actual company names a buyer can call.
3.1
businesses named per answer, average
The shortlist is this small. Being on it is the game.
30%
of answers lean on directories
Yelp, Angi, BBB & co. — the corroboration layer.
1%
of answers hedge with no names
The open-seat markets: nobody has claimed them yet.
Live report per trade
What this means if you own a local business
The AI engines aren’t hedging anymore. When a homeowner asks for the best option in a real trade and a real city, the answer is a shortlist of 3.1 named companies — and every buyer who asks gets one. Either your business is on that shortlist or a competitor is, and nothing in your traffic analytics will ever show you which.
The directory number matters just as much: roughly 30% of answers cite platforms like Yelp, Angi, or the BBB as corroboration. The engines cross-check before they vouch — which is why consistent listings and a steady review stream show up in nearly every named business we capture.
Want to see your own market’s answer? Run the free AI visibility check — same method, your business, one minute.
Sample receipts from the dataset
Every data point opens the actual captured answer.
Methodology — and its limits, stated plainly
The question set.One buyer question per trade-metro pair (“What are the best [trade] options in [metro]? Please name the top few businesses and briefly say why.”) across 20 trades × 20 metros, captured in daily batches, oldest cells refreshed first.
The engines. Currently: DeepSeek. DeepSeek answers from model knowledge (no live web search), so its data reads as what AI recalls about a market; live-search engines are added as the Index grows. Every row is labeled with the engine that produced it.
The measurement. Deterministic parsing — enumerated proper-noun list items count as named businesses; fixed patterns detect directory mentions and hedging. No AI grades the AI. The parser is open to inspection and re-scores the full archive whenever improved.
The limits. Answers regenerate constantly, so rates are rolling-window estimates, not fixed truths; small per-cell samples early on mean trade and metro splits firm up as the dataset grows. We publish the sample size at the top of this page for exactly that reason.
The Index measures the market. The check measures you.
Run your business through the same capture method and see whether the engines name you — with a shareable receipt either way.
Free · about 60 seconds · no call required