Home › Research › How AI decides who to recommend
SEMPITE Research · AI Visibility Index · August 2026
More Than 2 in 5 of Google’s Top-Ranked Local Businesses Are Invisible to ChatGPT
We pulled the top-3 map-pack businesses for 50 category–city markets (5 industries × 10 major U.S. cities), then asked AI engines to recommend providers in each. They named specific businesses every single time — just not always the ones winning on Google. This is our second monthly run, and every category we track lost ground.
ChatGPT visibility by category
Share of Google’s top-3 businesses that ChatGPT actually mentions:
The everyday trades still fare worst — two-thirds of top-ranked plumbers never appear in ChatGPT’s recommendations — and every category slipped between July and August.
What this means
AI assistants are already answering “who should I hire?” with confident, specific recommendations — and those recommendations are not simply Google’s rankings restated. Businesses that dominate the map pack are being skipped by the fastest-growing recommendation channel on earth, while competitors with clearer entities, stronger citations, and more AI-legible content take their place.
The Google AI Mode number deserves its own sentence: even Google’s own AI declines to mention 15% of the businesses Google’s own map pack ranks top-3. But the gap between the two engines is the real finding. Google’s AI Mode still names the #1 map-pack business 90% of the time — it is largely Google’s rankings in a conversational voice. ChatGPT names that same business 58% of the time. If you treat “AI search” as one channel, you are averaging two engines that disagree sharply about who deserves the recommendation.
If you’ve invested years in ranking on Google, this is the blind spot worth checking first. The playbook for fixing it is in our AI Search Visibility guide.
July → August change
Same 50 markets, same prompts, same engines, one month apart:
A caution we’d rather state than bury: this is two data points, not a trend. Language models are non-deterministic, and some of this movement is ordinary run-to-run variance rather than the engines genuinely re-ranking local businesses. What we can say is that the direction was consistent across all five categories and both engines, which variance alone would not usually produce. We’ll know more with a third run in September, and we’ll publish it whichever way it moves.
Methodology
August 2026 full run (data collected 1 August 2026): 50 category–city markets — plumber, dentist, personal injury lawyer, real estate agent, and HVAC contractor across New York, Los Angeles, Chicago, Houston, Phoenix, Philadelphia, San Antonio, San Diego, Dallas, and Miami. Baseline = top-3 Google Maps results per market (via DataForSEO). Engines: ChatGPT (gpt-4o-mini, prompted “I need a [category] in [city]… recommend specific businesses by name”) and Google AI Mode (query “best [category] in [city]”). Mentions matched on distinctive business-name tokens; an engine that returns no answer for a market is excluded from that market’s stats, never counted as a miss. July comparison figures come from the identical run configuration executed in July 2026. Claude is not in this month’s panel — it is paused pending API funding, and we report it as absent rather than as a zero. The index repeats monthly. Data available on request under CC-BY.
Is your business in the answers?
Run the free 60-second check — the same engine, pointed at you.