Each service category has its own short stack of directories and review sites that ChatGPT keeps reading before it names a single local business.
When ChatGPT recommends a local business, it has just finished reading a stack of web pages, and those pages are where the recommendation actually comes from. In our August 2026 study we logged them. Across 215 US markets in the first wave, asking "who is the best {service} in {city}" and recording the ChatGPT web app's own responses, we captured which domains were retrieved and which were cited for every answer.
The pattern that emerged is the most actionable thing in the whole study: every service category has its own short reading list, and some of those lists are nearly universal.
Review platforms and directories dominate the reading
Before the per-industry lists, the general layer. The single most retrieved domain across all 215 markets was reviews.birdeye.com, the review pages that reputation platform Birdeye hosts for local businesses, pulled in 37.7% of markets (81 of 215). Behind it came angi.com at 24.2%, expertise.com at 21.9%, doctor.webmd.com at 18.1%, homeadvisor.com at 17.7%, and zocdoc.com at 16.3%.
The cited layer, the sources ChatGPT actually shows under its answers, told the same story: reviews.birdeye.com led at 19.5% of markets (42 of 215), then expertise.com at 16.3% and threebestrated.com at 13.0%. Not the businesses' own websites. Aggregators, directories, and hosted review pages. The full source tables are on the study page at how ChatGPT picks a local business.
Psychology Today and Zillow anchor entire industries
Averages across industries undersell how concentrated some categories are. Sliced by vertical, with each vertical spanning 18 study markets, the anchor sources jump out:
- Therapists: psychologytoday.com was cited in 100% of markets (18 of 18). Total coverage. A therapist not on Psychology Today was invisible to the most consistent source in the entire study.
- Real estate agents: zillow.com in 94.4% of markets (17 of 18), and the top three sources together accounted for 85.9% of the vertical's citations (73 of 85).
- Personal injury lawyers: superlawyers.com in 83.3% of markets (15 of 18).
For these three industries, the study reads less like a mystery and more like a checklist with one item in bold.
The trades split their citations across a rotation
Home services had no single Psychology Today. Instead, each trade spread its citations across a handful of recurring sources:
- Window cleaning: angi.com in 52.9% of markets (9 of 17) plus threebestrated.com in 47.1% (8 of 17)
- Pool service: threebestrated.com in 50.0% (9 of 18)
- Dentists: reviews.birdeye.com in 50.0% (9 of 18)
- Chiropractors: reviews.birdeye.com in 44.4% (8 of 18)
- HVAC: consumeraffairs.com in 44.4% (8 of 18)
- Plumbers: bestprosintown.com in 44.4% (8 of 18)
- Auto repair: carfax.com in 33.3% (6 of 18)
- Roofing: expertise.com in 33.3% (6 of 18)
- Med spas: medspascout.com and discovermedspa.com at 27.8% each (5 of 18)
Notice how unglamorous some of these are. Best Pros in Town and Three Best Rated are not sites most marketing plans mention, yet they kept showing up as the raw material of ChatGPT's answers for the trades.
Coverage across the list beats one perfect listing
Why does the whole rotation matter instead of just the top source? Because of the churn mechanism I broke down in retrieval churn drives who ChatGPT recommends. Between two identical asks days apart, the set of retrieved domains overlapped at just 0.20 on a 0-to-1 scale (n=197 market pairs). ChatGPT reads a different subset of the rotation almost every time it answers.
A business listed on one source is present only when that source gets drawn. A business listed accurately across the category's whole rotation is present on nearly every draw. That redundancy, not any single placement, is what turns an occasional appearance into a consistent one. This is also the mundane, buildable version of the entity work described in what AI search optimization is: same name, same details, confirmed across many independent sources.
Your own website plays the confirming role
The directories introduce you. Your website vouches for the introduction. Among the businesses ChatGPT displayed, we successfully crawled 827 of 936 websites, and the shown businesses tended toward plain clarity: 63.6% carried the service in the page title, 56.5% carried the city, 44.5% ran LocalBusiness structured data, and 25.6% embedded a review widget. Only 13.8% marked up an aggregate rating.
None of that is exotic. The websites behind recommended businesses mostly just say what the business does and where, in the places machines read first, and let the review platforms carry the social proof.
The questions owners ask most about which sources ChatGPT reads, answered from the study data.
Ready to cover the reading list in your category?
The work here is refreshingly concrete: find your category's anchors, secure and complete your listing on each source in the rotation, keep every detail consistent, and let your website confirm the story plainly. Spearleaf builds exactly this coverage map inside our AI search program, market by market, with the study's source data as the starting point. If you want to know which of these pages already mention you and which are missing you entirely, reach out through our contact page and we will map it together.
This article was written by Joshua Albanese, founder of Spearleaf, a local SEO and marketing agency based in Fort Myers, Florida. Joshua has built six businesses from zero using organic search and content, and he leads Spearleaf's SEO, AI search, and local strategy for service businesses across the U.S.
Frequently asked questions
What sources does ChatGPT use to recommend local businesses?
For local questions it reads live web pages, and in our August 2026 study across 215 US markets those pages were dominated by review platforms and directories. The single most retrieved domain was reviews.birdeye.com, pulled in 37.7% of markets (81 of 215), followed by angi.com at 24.2% and expertise.com at 21.9%. On top of that general layer, each service category had its own anchor sources, like Psychology Today for therapists and Zillow for real estate agents.
Which directories matter most for my industry in AI search?
It depends heavily on the category, which is the point of the study's per-vertical reading lists. Therapists saw psychologytoday.com cited in 100% of their markets (18 of 18), real estate agents saw zillow.com in 94.4% (17 of 18), and personal injury lawyers saw superlawyers.com in 83.3% (15 of 18). Trades were more fragmented, with sites like Angi, Three Best Rated, ConsumerAffairs, and Best Pros in Town splitting the citations. Start with your category's anchors, then cover the rotation behind them.
Does my own website matter if ChatGPT mostly reads directories?
Yes, as the layer that confirms what the directories say. Among businesses ChatGPT displayed in our study, we crawled 827 websites successfully and found 63.6% named their service in the page title and 56.5% named their city, while 44.5% carried LocalBusiness structured data. Your site is where the entity gets verified, so it needs to state plainly who you are, what you do, and where. But a great website with no directory presence leaves you off most of the pages ChatGPT actually reads for local answers.
Should I pay for listings on every directory ChatGPT reads?
Not indiscriminately. The study measured which sources got read and cited, not which paid tiers work, and many of the domains on the lists offer free profiles. The sensible order is to secure accurate free listings across your category's rotation first, especially the anchors, and treat paid upgrades as a separate marketing decision judged on their own leads. Coverage and accuracy are what put you in the pool ChatGPT draws from. Spending more on one directory does not put you on the other nine.
Want to know which sites ChatGPT reads in your category and whether your business is on them? Let's build your coverage map.
Learn more →