How AI Recommends Local Businesses: 5 AIs, 120 Answers, 8 Cities

In our October 2026 test of 120 AI answers across eight cities, 79.4% of the businesses that ChatGPT, Perplexity, Gemini, Claude and Google AI Mode recommended did not appear anywhere in Google's top 20 map results for the same search. The five AIs also barely agreed with each other: 71% of the 558 businesses they named were picked by only one of them. The strongest signal we found was review volume, while star rating and map position mattered much less than most owners assume.

TL;DR

• In our October 2026 test of 120 AI answers, 79.4% of the 558 businesses the five AIs recommended were missing from Google's top 20 map results for the same industry and city.
• Google Maps top-20 businesses with 500+ Google reviews were named by at least one AI 54.8% of the time, against 12.2% for businesses with under 50 reviews.
• Perfect 5.0-star businesses were named only 19% of the time, while businesses rated 4.8 to 4.99 were named 45% of the time, because most 5.0 listings had few reviews.
• Two different AIs asked the same question gave shortlists that overlapped by 17% on average, and only 5 of 558 businesses were named by all five AIs.
• Industry review sites such as EnergySage, WeddingWire and moveBuddha made up 39.6% of everything the AIs cited, the largest single source type.

This page is the full write-up of the Orkkid AI Recommendation Study. We asked five AI assistants the same question a customer would ask ("Who is the best moving company in Denver?") for three industries in eight cities, recorded every business they named and every source they cited, and matched each answer against Google Maps. Every number below comes from that dataset, and you can download the per-business data at the end.

Most AI picks come from outside Google's map top 20

When we matched the 824 recommendations against the Google Maps top 20 for the same industry and city, only 115 of the 558 unique businesses (20.6%) appeared in those 480 map listings. The other 79.4% were businesses Google did not show in its top 20 at all. Looked at from the map side, only 25.4% of the Maps top-20 businesses were named by any of the five AIs.

Even the top of the map gets skipped. Of the 72 businesses in the top three map spots, 45 (62.5%) were named by none of the five AIs. Claude was the most likely to include a top-three map business in its answer (50% of answers), and Google AI Mode the least likely (20.8%), even though AI Mode had the highest share of picks inside the map top 20 overall (36.4%).

The gap varied by city. London stood out, with 92.3% of AI picks missing from the local map top 20.

What it means for you: ranking in the local map pack and being recommended by AI are two separate wins. A strong Google Business Profile still helps (see what AI reads from your Google Business Profile), but most AI picks in our sample came from somewhere else, which the next sections trace.

Reviews matter, star ratings barely do

The clearest pattern in the data is review volume. Among the 480 Google Maps top-20 businesses, the share named by at least one AI rose steadily with review count, from 12.2% under 50 reviews to 54.8% at 500 or more.

Star rating told a different story. Named businesses averaged 4.82 stars and unnamed ones 4.80, which is practically the same. Businesses with a perfect 5.0 were named only 19% of the time, because most of them had few reviews (a median of 51). Businesses rated 4.8 to 4.99, with a median of 215 reviews, were named 45% of the time.

The AIs say this out loud in their answers. Explaining one Tampa pick, Claude wrote: "The rating is slightly lower, but the review count is very large, which suggests a lot of experience with different weddings."

What to do: treat review count as the number to grow. Ask every customer for a Google review, and do it consistently, because businesses rated 4.8 to 4.99 (median 215 reviews) were named more than twice as often as perfect 5.0 listings (median 51). Correlation is not proof of cause, since review volume also tracks business size and age, but it was the strongest single signal we measured.

Map rank helps less than you would think

Higher map positions were named more often, but the effect was modest. 37.5% of businesses in the top three map spots were named by at least one AI, against 24.4% for spots 4 to 10 and 22.5% for spots 11 to 20. In moving, positions 11 to 20 (30%) actually beat positions 4 to 10 (19.6%).

When we split map position by review count, reviews won. A business ranked 11 to 20 with 500+ reviews was named 47.4% of the time, twice the rate of a top-three business with under 50 reviews (23.1%). And a business rated under 4.7 stars with 500+ reviews (42.1%) was named about four times as often as a perfect 5.0 with under 50 (10.7%).

What to do: keep working your map ranking, because the top three spots still had the highest named rate. Just do not expect a map-pack win to carry you into AI answers on its own. The businesses that had both a top-three spot and 500+ reviews were named 64.7% of the time, which was the best result of any group we measured.

The five AIs disagree with each other

We compared every pair of engines' shortlists for the same industry and city. On average two shortlists overlapped by 17%. The closest pair was ChatGPT and Claude (26%), and the furthest apart was Perplexity and Google AI Mode (8%).

The disagreement shows up business by business too. Of the 558 businesses named, 396 (71%) were named by only one AI, and just 12.2% were named by three or more. Only five businesses were named by all five AIs: Pickfords (moving, London), Solar Power Pros (solar, Denver), Solar Topps (solar, Phoenix), Renu Energy Solutions (solar, Charlotte) and Sunergy Systems (solar, Seattle). Two of those five, Pickfords and Solar Power Pros, were not in the Google Maps top 20 for their city.

Every engine had a large set of picks of its own. Between 38.4% (ChatGPT) and 53.4% (Google AI Mode) of each AI's businesses were named by no other AI in our test.

What to do: never judge your standing from one AI. A business can be ChatGPT's first pick and absent from Perplexity in the same city. Check all five with the same prompts, which our free AI Citation Checker and Prompt Panel Generator are built for, and track the results over time as covered in how to track ChatGPT citations.

Where each AI gets its information

All 120 answers cited at least one source: 613 citations across 229 different domains. Industry review sites were the biggest source type at 39.6%, followed by business websites (16%), best-of lists (11.7%) and media and publishers (11.3%).

Each engine has its own reading habits. Gemini drew 32.7% of its sources from business websites, the most of any engine, and cited Reddit in 16 of its 24 answers (66.7%). ChatGPT cited Reddit in 37.5% of answers, while Perplexity and Claude cited it in none. Perplexity leaned on best-of lists (21.1%) and media (17.8%). Claude's most cited site was yelp.com. Google AI Mode cited google.com in 14 of its 24 answers, and Google properties made up 29.2% of its sources.

ChatGPT described its own process in Denver as "based on current 2026 mover rankings, review aggregators, and local Denver chatter, these are my top picks." That sentence is a fair summary of the whole dataset.

These are the 15 domains cited most often across all 120 answers:

What to do: find out which of these sites cover your industry and city, then make sure you are listed, complete and reviewed on them. A profile on your industry's main review site reaches more AI answers than any single change to your own website. Claude added a useful caution in Toronto: "Many of these lists are paid placements or vendor-directory rankings, so read independent reviews on Google as well." Earn your place on lists honestly, because the AIs are reading them critically.

How long each AI's shortlist is

Claude gave the longest lists, 10.4 businesses per answer on average and up to 16 in one answer. Google AI Mode gave the shortest, 4.9 on average and never more than 6. ChatGPT, Perplexity and Gemini sat in between at 6.3 to 6.4.

Google AI Mode also asked the customer a follow-up question (budget, style, type of move) in 23 of its 24 answers. A short list means fewer seats: on AI Mode, a business has to beat almost everyone in its city to be named at all. Claude's #1 pick was a top-three map business in 25% of answers, the most of any engine, against 4.2% for ChatGPT and Gemini.

What to do: count how many of the five AIs name you, and read each result against the length of that engine's list. Claude's lists of up to 16 are the easiest to get onto and Google AI Mode's lists of four to six are the hardest, so treat a mention on AI Mode as the harder win and a missing spot on Claude as the clearer warning sign.

How the three industries differ

The pattern held in all three industries, but the strength varied a lot. Wedding photography was the most fragmented market: 87.6% of picks were outside the map top 20 and the engines' shortlists overlapped by only 5.3%. Solar was the most consistent, and four of the five businesses named by every AI were solar installers.

Each industry has a gatekeeper site that a large share of AI answers pass through. In solar, energysage.com was cited in 65% of answers. In weddings, three wedding marketplaces (WeddingWire, The Knot and Zola) were each cited in 40% or more. In moving, the AIs read movebuddha.com (47.5%), freightwaves.com (42.5%) and consumeraffairs.com (35%) most.

What to do by industry: movers should start with their moveBuddha, ConsumerAffairs and Google review counts (our moving company program covers this). Wedding photographers should treat WeddingWire, The Knot and Zola profiles as AI inputs, since those three dominate the sources (see SEO for wedding vendors and our wedding vendor program). Solar installers should prioritize an active, well-reviewed EnergySage profile (our solar program).

What the websites of named businesses have

We checked the websites of the Google Maps top-20 businesses for common technical signals: 448 sites, of which 358 were reachable (94 named by AI, 264 not named). The differences were small. LocalBusiness schema showed the widest gap (71.3% of named sites against 62.1%), followed by review stars in schema (17% against 12.5%) and FAQ schema (13.8% against 9.5%).

Two popular tactics showed no effect in our sample. Named and unnamed sites had an llms.txt file at almost the same rate (29.8% against 28.8%). Blocking AI crawlers was rare: 5 of the 358 sites (1.4%) shut out AI crawlers, three by naming bots such as GPTBot and two by blocking every bot, and none of those five businesses were named. Having a website at all mattered more, with 98.4% of named Maps businesses listing a website against 91.6% of unnamed ones (97.8% against 82.6% in solar).

Most llms.txt files we found were never a deliberate choice: 57 of the 103 (55.3%) were generated automatically by Wix or an SEO plugin. Our beginner's guide to llms.txt explains what the file does and does not do, and the llms.txt spec is the primary reference.

What to do: add LocalBusiness (or your specific type, such as MovingCompany) schema with your reviews, list your website on your Google Business Profile, and confirm with our AI Crawler Checker that you are not blocking AI crawlers by accident. Treat llms.txt as optional housekeeping, because it made no measurable difference here.

Every business the AIs recommended

These tables list every business named in the study, how many of the five AIs named it, and its Google Maps rank and review count where it appeared in the top 20. Filter by city, search by name, or sort by any column. A blank map rank means the business was outside the Maps top 20 for that search. The same data is in the CSV download.

Moving companies (164 businesses)

Wedding photographers (267 businesses)

Solar installers (127 businesses)

What this means for your business

Each action below is tied to a finding in this study. None of them is a guarantee, because these are correlations from a single run, but they are where the evidence points.

1. Grow your Google review count every month. Maps businesses with 500+ reviews were named 54.8% of the time and those under 50 just 12.2%. Volume mattered more than rating, so a steady flow of honest reviews beats chasing a perfect 5.0. A direct review link, such as one from our Google Review Link Generator, makes asking easier.
2. Get listed and reviewed on your industry's gatekeeper site. Industry review sites were 39.6% of all citations, and one site per industry (EnergySage, WeddingWire, moveBuddha) appeared in roughly half or more of that industry's answers.
3. Earn a place on local best-of lists and in local media. Best-of lists (11.7%) and media and publishers (11.3%) together made up nearly a quarter of sources, and Perplexity relied on them for 38.9% of its citations.
4. Take part in Reddit threads about your city and trade. Gemini cited Reddit in 66.7% of its answers and ChatGPT in 37.5%. Helpful answers from a named owner in local threads are the honest way to show up there.
5. Check all five AIs separately, every quarter. Shortlists overlapped by only 17% on average, so a single engine says little about the others. Use the same prompts each time so you can see change.
6. Fix the website basics that showed a real gap. LocalBusiness schema (71.3% vs 62.1%) and review markup (17% vs 12.5%) were more common on named sites, and nearly every named business listed a website on its map listing. Our LocalBusiness Schema Generator writes the markup for you.
7. Keep your Google Business Profile strong for Google AI Mode. AI Mode had the highest share of picks inside the map top 20 (36.4%) and drew 29.2% of its sources from Google itself.

This is also what our generative engine optimization work is built on, and it aligns with Google's own advice to create helpful, reliable, people-first content and with its published local ranking factors of relevance, distance and prominence. For the engine-level view, see how ChatGPT decides which businesses to recommend and our Perplexity SEO guide.

How we got this data

Date and engines. All answers were collected on 7 October 2026 through DataForSEO. ChatGPT (gpt-5.5), Perplexity (sonar-pro), Gemini (gemini-3.8-flash) and Claude (claude-sonnet-5-5) were queried with web search switched on and no location parameter, using the DataForSEO LLM Responses endpoint. Google AI Mode was queried through the DataForSEO Google AI Mode endpoint with the search "best {industry} in {city}" and the city set as the location. Each engine's web search works differently; see the documentation for OpenAI web search, Claude web search, Gemini grounding with Google Search and Perplexity Sonar Pro.

Prompt and sample. The four chat assistants received the identical prompt "Who is the best {moving company | wedding photographer | solar installer} in {City}, {Region}? Give me your top picks." in eight cities, written as Denver, Colorado; Austin, Texas; Phoenix, Arizona; Charlotte, North Carolina; Seattle, Washington; Tampa, Florida; London, UK; and Toronto, Canada. That is 3 industries times 8 cities times 5 engines, or 120 answers, which produced 824 recommendations of 558 unique businesses. A business counted as named when the answer recommended it by name as a pick for that city; directories, marketplaces and review sites were recorded as sources, not picks. All 120 answers named at least four businesses, and none declined.

Google Maps matching. For each industry and city we pulled the Google Maps top 20 with the DataForSEO Google Maps endpoint (480 listings in total), using the industry as the keyword and the city as the location. Names were matched after lowercasing and removing legal suffixes, generic trade words (moving, photography, solar and similar) and city names. Two names matched when their core words were identical, when one contained the other (five or more characters), when they shared a distinctive first word of six or more letters, or when at least two thirds of their words overlapped.

Sources and websites. Each cited domain was counted once per answer and classified into nine types (industry review sites, business websites, best-of lists, media and publishers, directories, Reddit and social, Google, government and trade bodies, other) with fixed rules applied in that order. We then fetched the homepage, robots.txt and llms.txt of every Maps top-20 business website: 448 sites, 358 reachable. We recorded whether an llms.txt file existed, robots.txt rules that fully block 11 AI crawlers (including GPTBot, ClaudeBot, PerplexityBot and Google-Extended), JSON-LD schema types, the city name in the homepage title, and homepage word count.

Limits. This is one run per prompt and engine, so it is a snapshot; AI answers are non-deterministic and a re-run will differ in detail. The Maps top 20 reflects one search per city from the city location, and a customer searching from a suburb would see a different map. The study covers three industries and eight cities, so it describes these markets, not every local business. Every relationship reported here is a correlation. We plan to re-run the same prompts in the same cities each quarter and publish the trend.

Download the data (CSV): one row per business with industry, city, business name, the number of AIs naming it, which AIs, Google Maps rank and Google review count. You are welcome to use and cite it with a link to this page.

Free GEO audit

Be the business AI recommends

More customers now ask AI who to hire before they ever search. Your free GEO audit shows exactly where your business stands in those answers today.

Tested across ChatGPT · Perplexity · Google AI Overviews · Gemini

See who AI recommends

FAQ

Which AI follows Google Maps most closely?

Google AI Mode. In our October 2026 test, 36.4% of its picks appeared in the Google Maps top 20 for the same search, against 21.9% to 27.2% for ChatGPT, Perplexity, Gemini and Claude. It also gave the shortest lists (4.9 businesses per answer) and drew 29.2% of its sources from Google.

Do Google reviews affect whether ChatGPT and other AIs recommend a business?

In our data they were the strongest signal we measured. Google Maps top-20 businesses with 500+ reviews were named by at least one AI 54.8% of the time, against 12.2% for those with under 50. Star rating made little difference: named businesses averaged 4.82 stars and unnamed ones 4.80.

Does ranking #1 on Google Maps mean AI will recommend me?

No. Of the 72 businesses in the top three map spots across our 24 searches, 45 (62.5%) were named by none of the five AIs. Top-three businesses were named more often (37.5%) than those ranked 11 to 20 (22.5%), but review volume mattered more than position.

Does adding an llms.txt file help a business get recommended by AI?

Not in our sample. 29.8% of named businesses and 28.8% of unnamed businesses had an llms.txt file, which is almost identical. Most of the files we found (57 of 103) were generated automatically by Wix or an SEO plugin.

Can I use this data in my own article or research?

Yes. The per-business data is free to download as a CSV from the method section above, and every chart on this page has a "Cite this chart" button. Please credit "Orkkid AI Recommendation Study, October 2026" and link to this page.