Local businesses can reach more customers by understanding how often AI platforms return the same recommendations. A new study of 200,085 prompts and 1.9 million citations found that the same search run multiple times returns a different set of local businesses each time, with Google Maps surfacing tracked businesses 66% of the time compared to 32-38% for AI surfaces. The work was carried out using the Local AI Visibility Tracker across ChatGPT, Google AI Mode, and Google AI Overviews.<\/p>\n
Researchers ran 200,085 non-branded prompts across three AI platforms: ChatGPT, Google AI Mode, and Google AI Overviews. The prompts covered searches that a local business could realistically be recommended for, spread across 1,300 business locations. Each prompt was then re-run several times over a 60-day period, and again from multiple points on a map within the same city, to measure how much the list of recommended businesses changed between runs (variance) and how often a single business reappeared (persistence).<\/p>\n
For each prompt run, the team logged which businesses were named, how many were mentioned, and how many businesses appeared across the different sets of answers.<\/p>\n
There is significant variance in the businesses that AI platforms recommend when the same prompt is run repeatedly. Only 20-33% of businesses overlap between repeat runs on average (ChatGPT 23%, AI Mode 21%, AI Overview 33%). A business that appears once is seen again in about half of subsequent attempts (50% persistence for ChatGPT and AI Mode, 58% for AI Overview). 71-80% of businesses appear in half the attempts or fewer, so inconsistent inclusion remains the norm.<\/p>\n
A worked example shows why. For the prompt “best pizza place in manhattan for a tourist that only has time to try one pizza” run four times on ChatGPT, the four responses named different sets of businesses: one response added three extra businesses that appeared only once. Across the four runs the average similarity was 43% and the average persistence 50%.<\/p>\n
AI Overview had the highest consistency of the three platforms, but it also returned the fewest businesses per response. Part of the reason is that AI Overview grounds its answers in traditional search results and only fires for a smaller share of queries, while ChatGPT and AI Mode produce a response every time.<\/p>\n
ChatGPT surfaces the most businesses per response, but the median across platforms is 2-4. ChatGPT averages 4.1 businesses per response (range 1-7, median 4), Google AI Mode averages 3.5 (range 0-6, median 4), and Google AI Overview averages 2.5 (range 0-4, median 3). Empty responses are not rare either: 10% for ChatGPT and 11% for both Google surfaces.<\/p>\n
This tighter count from AI Overview mirrors the size of a traditional Local Pack, while ChatGPT’s longer lists are one of the main reasons its answers look less consistent from run to run.<\/p>\n
Ranking well on Google Maps does not translate directly into being recommended by AI. Google Maps mentioned a tracked business in 66% of searches, compared to 38% for AI Overview, 33% for ChatGPT, and 32% for AI Mode. Maps is also far more stable: 99% average agreement across attempts vs 91-96% for the AI platforms.<\/p>\n
The platforms disagree with each other as well. AI Mode and AI Overview share only 29% of named businesses on average, despite both relying heavily on Google Business Profile. ChatGPT overlaps only 19-20% with either Google surface, reflecting its heavier use of Yelp and Bing sources rather than Google Business Profile.<\/p>\n
When an AI answers a prompt, it runs a “fan-out” of related sub-queries before blending the results. A prompt like “best lawyer in downtown LA for family law” gets broken into searches such as “best family law attorney Downtown Los Angeles,” “divorce custody reviews Los Angeles family law attorneys,” and “certified family law specialist downtown Los Angeles.” This is different from a single keyword match on Maps and helps explain why strong Maps rankings do not guarantee AI inclusion.<\/p>\n
Like traditional local search, moving the searcher across a city changes what AI surfaces, and ChatGPT is the most volatile of the three. Overlap across different points in the same town runs at 36.2% for ChatGPT, 46.9% for AI Mode, and 47.4% for AI Overview. Only 3.5% of ChatGPT results are always present across points, compared to 8.8% for AI Mode and 25.0% for AI Overview.<\/p>\n
Proximity still matters. 52.8-59.3% of business picks fall within 5 km of the searcher, and 71.7-78.2% fall within 10 km. Both AI Overview and AI Mode keep tight radii, with around 5% of picks within 1 km. ChatGPT starts close to this but weakens with distance; its “circle of influence,” the radius covering 90% of its results, stretches to 287 km, compared to 64 km for AI Mode and 38 km for AI Overview.<\/p>\n
The study also tallied which sources the AIs cited. Across all three platforms, business websites dominated: 93% of all unique domains cited were business sites, and they accounted for 42% of all citations. Google Business Profile was the single largest source at 28.5% of all citations across the three platforms. All three platforms rely on Yelp, with the Google surfaces leaning on Facebook as well, while ChatGPT pulls more from Yelp and Bing.<\/p>\n
The same prompt will return a different shortlist each time it is run, and a single business that appears once will reappear in roughly half of subsequent attempts. Treating that volatility as background noise is risky when visibility drops well below 50%. Tracking which prompts surface a business, how often, and alongside which competitors makes it possible to act on the pattern rather than leave it to chance.<\/p>\n
Practical steps that come out of the data: make the business address and NAP (name, address, phone number) easy to find and consistent across the website and off-site listings, keep location pages clear, build reviews, get listed on relevant local and industry directories, and review the Google Business Profile category. These are the same fundamentals that feed traditional local search, and the study shows they feed AI citation sources too.<\/p>\n
For businesses that want to see how their rankings shift across a service area rather than one city average, a geo grid report shows rank position mapped across towns and suburbs, and SEOScanPro’s GEO Grids do this.<\/p>\n
About 50% of the time for ChatGPT and AI Mode, and 58% of the time for AI Overview. A business that appears once is seen again in roughly half of subsequent attempts.<\/p>\n
Not directly. Google Maps mentions a tracked business in 66% of searches, compared to 32-38% for AI platforms (38% AI Overview, 33% ChatGPT, 32% AI Mode). Maps rankings and AI recommendations are driven by different processes.<\/p>\n
ChatGPT averages 4.1 businesses per response (range 1-7, median 4), Google AI Mode averages 3.5 (range 0-6, median 4), and Google AI Overview averages 2.5 (range 0-4, median 3). Empty responses occur in 10-11% of runs across the three.<\/p>\n