Local AI search now delivers specific, actionable business recommendations, and a large-scale study pins down which signals determine who gets recommended. Uberall’s GEO analyst Katya Shishchenko analyzed 120,000+ AI mentions across 3,793 locations in US cities, covering dentists, restaurants, grocery stores, hotels, and banks across five AI models. The headline finding: review volume predicts AI mentions more consistently than star ratings do in four of five business categories.
Note: This research was conducted and published by Uberall, a local marketing platform. Data cited below refers to Uberall’s study.
Do All Five AI Models Recommend Local Businesses the Same Way?
No. Each model has a distinct recommendation “personality,” and understanding the differences matters for where teams allocate effort. Gemini surfaces 8x more unique restaurants than ChatGPT in equivalent queries, largely because it cross-references live Google Maps data. Perplexity searches live and cites its sources, generating the most mentions per run. ChatGPT produces a short, concentrated shortlist and carries the highest hallucination rates (false business recommendations) across all verticals tested.
- Claude: most conservative; favors local and community businesses, largely avoids naming specific healthcare providers.
- Gemini: most diverse; cross-references live Google Maps, surfaces 8x more unique restaurants than ChatGPT in the same study (Uberall).
- ChatGPT: consensus-driven; short sticky shortlist, highest hallucination rates across verticals.
- Grok: references Instagram content and chef qualifications more than any other model; responses read like long-form food journalism.
- Perplexity: searches live, cites sources, generates the most mentions per run; strongest model for businesses with a current, strong web presence.
Which 4 Signals Actually Drive AI Mention Rates?
Uberall’s study identified four consistent signal categories across all five models and all five verticals, labeling them BARS: Business profile completeness, Authority (earned media and editorial lists), Reviews, and Social presence. Brand size and market share were not reliable predictors of recommendation frequency. The signals split into two roles: some determine whether a brand qualifies to be mentioned at all; others determine how often it appears once it qualifies.
How Much Does Google Business Profile Completeness Matter?
Profile completeness acts as a threshold signal in local AI search: it determines entry into the consideration set, before frequency. The impact is steep. A completed GBP description can produce a threefold increase in mention rates for grocery stores, per Uberall’s data. For hotels, increasing listed attributes from the 6-10 range to 31-50 shifts mention probability from 22% to 94%.
Photo count stands out as the single strongest predictor of restaurant mention frequency. Top-mentioned restaurants average three times more photos than those that don’t appear. The same signal predicts both whether a dental practice gets mentioned and how often, and it ranks among the three strongest predictors for banks.
Does Review Volume or Star Rating Matter More for AI Visibility?
Across all five AI models, review volume predicts AI mentions more consistently than star ratings do. This is one of the study’s most counterintuitive findings for teams that have historically spent significant effort chasing rating improvements.
- Grocery: brands with higher volume but lower ratings are mentioned 94.3% of the time; high-rated brands with low volume are mentioned 60.6% of the time (Uberall).
- Dentists: no star rating on any platform reaches statistical significance as a mention factor.
- Banking: higher aggregate ratings on Yelp and TrustPilot correlate negatively with mention frequency. The most-mentioned banks are national giants that accumulate complaints at scale, so the negative correlation reflects size, not a penalty for quality.
- Hotels: the exception. GBP star ratings correlated more strongly with mention probability and frequency than review count, making this the one vertical where rating improvement is the priority signal.
Does Brand Size Influence AI Recommendations?
Less than most assume. Across all five verticals, brand size (store count, market share, room supply) was a poor predictor of AI recommendation frequency. For restaurants and dentists, independent brands can and do outperform chains on mention rates. The authority signals that did show consistent influence:
- Media mentions: brands with 30+ news mentions saw a 15-fold increase in mention frequency for banking and a 100% mention rate in grocery (Uberall).
- Editorial list presence: Bankrate, Forbes Travel Guide, and Michelin recognition function almost like direct ingestion into LLM (large language model, the core technology behind these AI tools) training data. Banks on three or more editorial platforms see a 13-fold mention increase. Michelin recognition appears in 94.5% of Perplexity’s restaurant responses.
- Wikipedia presence: a reliable positive factor for hotels, grocery stores, and banks; not significant for dentists.
- Delivery platforms: presence on Instacart, DoorDash, or Mercato does not influence grocery store AI visibility.
What Role Do Facebook and Instagram Play?
The two platforms affect local AI search differently. Facebook enhances mention probability (getting a brand into the AI response at all). Instagram amplifies mention frequency once a brand already qualifies. Dentist practices that receive AI mentions have almost 5x the Facebook followers of those that don’t. Restaurants with a strong Instagram and Yelp presence are mentioned almost 7x more than those without. For boutique hotels, Instagram is the single strongest predictor of AI mentions, outperforming even GBP completeness and editorial signals (Uberall).
What This Means for AI-Search Visibility
The plainest way to read this study: AI models are not building local rankings from scratch. They are connecting dots between the credibility signals that already exist about a business across the web, and coverage across those dots determines who shows up.
Here is the shift that matters most in practice. Many local brands already work on versions of these signals, gathering reviews to improve ratings, posting on social for engagement, listing on directories for brand awareness. What Uberall’s data suggests is that the optimization frame needs adjusting: volume and breadth of signals matter more to AI models than polish and depth on any single one.
The review volume finding makes this concrete. A grocery brand with many reviews and a 4.1 average is mentioned 94.3% of the time. A brand with a 4.6 average but few reviews is mentioned 60.6% of the time. AI models are likely treating review count as a proxy for how well-known a business is across the web, rather than reading reviews the way a consumer does. Ratings signal quality; coverage signals existence.
One caveat this study cannot resolve: model behaviors change with training updates. Perplexity’s live-search architecture will respond fastest to changes in web presence. ChatGPT and Claude, which rely more heavily on training data, may lag. The signals identified here are grounded enough in how AI systems ingest structured data that they are unlikely to become irrelevant. Their relative weight per model will move.
Hingewise’s read: the most durable insight in this dataset is that editorial authority (Michelin, Bankrate, Forbes Travel Guide) functions as a multiplier, but it is not the place most local brands can start. The more actionable near-term lever is the review volume and profile completeness side of the framework, which is also the part most within a brand’s direct control. The 22%-to-94% mention probability swing from adding hotel attributes is the kind of gap that rarely exists in traditional search optimization anymore. In local AI search, it still does.
Quick Checklist: Before Optimizing for Local AI Search
- Audit GBP completeness across all locations: description, categories, and attributes. Note how many attributes each location currently lists (the 31-50 range is the benchmark for 94% mention probability in hotels).
- Count photos per location on GBP and other indexed platforms. Compare against the top-mentioned competitor in your category (top restaurant mentions average 3x more photos than non-mentioned ones).
- Check review volume distribution across the platforms most relevant to your vertical (Yelp, TrustPilot, Google) and identify where volume is thin relative to competitors.
- Identify whether the brand has editorial directory listings applicable to your vertical (Bankrate, Michelin, Forbes Travel Guide) and flag gaps.
- Run a media mention count. If below 30 across indexed news sources, this represents a signal gap for banking and grocery AI visibility.
- Assess Facebook follower counts and Instagram activity per location, not just at brand level.
- Test each AI model with representative local queries in your category to identify which model your brand already appears in and where the largest gap exists.
The data point worth tracking over the next twelve months is whether these signal weightings shift as AI models are updated and as more local queries migrate from search engines to AI assistants. Uberall’s study provides a snapshot of a rapidly moving baseline. How stable these patterns prove across model updates will determine whether BARS becomes a durable local AI search framework or a near-term benchmark.
Source: Sara Vordermeier, “5 AI Models, 4 Signals: What 120K Mentions Reveal About Multi-Location SEO,” Search Engine Journal (sponsored by Uberall). Read the original.
Lam Nguyen is the founder of Hingewise, an AI-search and GEO visibility agency.
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.