We ran 10,000 local business recommendation queries through ChatGPT and analyzed the patterns.
The question we were trying to answer: What actually determines which businesses ChatGPT recommends? Is it random? Is it gameable? Or does it follow consistent, learnable patterns?
The answer is that ChatGPT follows consistent patterns that are closely correlated with real-world business quality signals — but with predictable gaps and biases that businesses can understand and address.
Study Methodology
We constructed a test set of 10,000 queries across 25 service categories in 40 US markets (metro and mid-sized cities). Queries were run using ChatGPT-4o with web browsing enabled.
For each query, we recorded:
- Which businesses were recommended (by name)
- How many businesses were recommended
- What attributes ChatGPT cited about each recommendation
- Whether the recommended businesses had verifiable quality signals (reviews, GBP completeness, website quality, credentials)
We then compared recommended businesses to non-recommended businesses in the same market/category to identify differentiating characteristics.
Key Findings
Finding 1: ChatGPT Typically Recommends 3-5 Businesses
Across all query types, ChatGPT settled on recommending 3.4 businesses on average. It rarely recommended just 1 (which would feel overconfident) or more than 7 (which would feel unhelpful).
The 3-5 range appears to be a deliberate calibration. For "best [service] in [city]" queries, ChatGPT treats this as "here are the top options" rather than "here is the one best answer."
Implication: There are typically only 3-5 slots available in ChatGPT's recommendation response for any given query. This is why being in the top tier matters far more than marginal improvements in your ranking signals.
Finding 2: Review Count Correlates Strongly With Recommendation
Of the businesses recommended by ChatGPT in our study, 78% had 50+ Google reviews. Of businesses in the same category that were NOT recommended, only 31% had 50+ reviews.
But the correlation wasn't perfectly linear. Above approximately 200 reviews, additional review volume had diminishing impact. The jump from 0-50 reviews was the most impactful, followed by 50-100 reviews.
Review thresholds with meaningful impact:
- 0 → 25 reviews: Large impact
- 25 → 50 reviews: Large impact
- 50 → 100 reviews: Moderate impact
- 100 → 200 reviews: Small impact
- 200+ reviews: Minimal marginal impact
Implication: If you have under 50 reviews, getting to 50 is your highest priority. If you have 200+, additional review volume matters less than other signals.
Finding 3: Business Age Matters Significantly
ChatGPT recommended businesses that had been operating longer at significantly higher rates.
Median years in business:
- Recommended businesses: 9.2 years
- Non-recommended businesses: 4.1 years
This isn't ChatGPT knowing when businesses were founded — it's a proxy for the accumulated web presence, review history, and third-party citations that older businesses have had time to build.
Implication: Newer businesses face an inherent disadvantage in AI recommendations. The workaround is accelerating the accumulation of reviews, citations, and content rather than waiting for time to pass.
Finding 4: ChatGPT Cites Specific Attributes as "Reasons"
When ChatGPT recommended a business, it almost always included a brief reason. We categorized these reason types:
- Review quality/volume (mentioned in 71% of recommendations)
- Specific credentials or certifications (mentioned in 43%)
- Specialization match (mentioned in 38%)
- Location proximity (mentioned in 35%)
- Years in business (mentioned in 29%)
- Specific services offered (mentioned in 27%)
Implication: ChatGPT is specifically looking for credentials and specializations to cite. If it can't find them, it recommends based on reviews alone — and businesses with both reviews AND documented credentials appear with more authoritative descriptions.
Finding 5: Emergency Queries Show Different Patterns
For emergency service queries ("emergency plumber," "24-hour AC repair," "urgent dental care"), ChatGPT's recommendation patterns shifted:
- Location proximity became the most important signal
- "Open 24/7" or "available now" indicators became critical differentiators
- Review count mattered less than for non-emergency queries
- Recent activity (recent GBP posts, recently updated website) mattered more
Implication: For emergency service queries, explicit availability documentation (24/7 in GBP, same-day service on website, emergency phone number prominent) is the primary AI visibility lever.
Finding 6: Negative Reviews Had Lower Impact Than Expected
Businesses with even a handful of negative reviews weren't systematically excluded from recommendations if their overall rating stayed above approximately 4.0.
However, negative reviews with unaddressed owner responses — where the owner argued with the reviewer — were more damaging than the negative review itself.
Pattern: ChatGPT appears to evaluate owner response quality as a signal. Professional, empathetic responses to negative reviews actually partially offset the negative review's impact.
Implication: Respond to every review, including negative ones. The quality of your response matters for AI recommendation consideration.
Finding 7: Category Specificity Matters Enormously
Queries for highly specific services ("orthodontist accepting adult patients," "electrician for solar panel installation," "plumber for gas line repair") showed dramatically different recommendation patterns than generic queries.
For specific queries, ChatGPT was far more likely to recommend businesses that explicitly documented that specific service — even if they had fewer reviews than a more generic competitor.
Implication: Document your specific services explicitly. If you do something particular well, or serve a specific client type, make it explicit everywhere. This is how smaller businesses can compete for specific queries against larger, higher-review competitors.
Finding 8: Information Inconsistencies Harm Recommendation Rates
We identified 340 businesses where ChatGPT recommended them with an inaccurate detail (wrong phone number, discontinued service, outdated pricing). We then tracked how often ChatGPT recommended these same businesses over the following weeks.
Businesses with confirmed information inconsistencies were recommended 34% less frequently than comparable businesses with consistent information.
Implication: NAP consistency (name, address, phone) across all sources isn't just a traditional SEO concern — it materially affects AI recommendation rates.
Industry-Specific Findings
Legal Services: ChatGPT prioritized State Bar Association listings and Avvo profiles for attorney recommendations. Attorneys listed on both with complete profiles had 2.4x higher recommendation rates.
Medical: Healthgrades and Zocdoc profiles were heavily cited. Doctors with complete profiles on these platforms were 1.9x more likely to be recommended.
Restaurants: Recent Yelp reviews were the strongest predictor. Restaurants with 50+ Yelp reviews from the last 6 months had 2.1x higher ChatGPT recommendation rates than those without.
Home Services: Angi (formerly Angie's List) and Thumbtack profiles were frequently cited. The presence or absence of these profiles was a meaningful differentiator.
HVAC/Plumbing/Electrical: NATE, IAPMO, and IBEW credentials were specifically identified and mentioned. Businesses with these credentials in their web content were recommended 1.7x more often.
The Overall Pattern
ChatGPT's local business recommendations are not random. They follow a consistent weighting that prioritizes:
- Review volume (up to a threshold)
- Business age and accumulated web presence
- Documented credentials and certifications
- Service-query specificity match
- NAP consistency across sources
- Emergency availability documentation (for urgent queries)
These signals are all improvable. None of them require a large budget — they require consistent execution and ongoing maintenance.
The businesses that understand these patterns and optimize deliberately are building durable AI visibility. The businesses that don't are leaving a significant and growing referral channel unclaimed.
Use Scope to see exactly which of these signals you're strong on and which you're missing — and to track your ChatGPT recommendation rate as you improve them.