Ryan Doser research · Field test · Data collected October 7, 2026

Does ChatGPT Recommend Local Businesses? 250 AI Picks Tested

Claude agents asked four AI assistants to name the best home-service companies in two cities, then checked all 250 recommendations against Google Maps. Across all four assistants, nearly every pick was a real local business, fewer than a quarter came from Google's top 3, and the assistants rarely agreed on who to list first.

  • 250 AI recommendations
  • 12 questions
  • 2 cities
  • 4 assistants
  • Collected Oct 7, 2026

Free to use with a link to this page. Press questions: ryan@ryandoser.com

On this page
97.6%of the 250 AI recommendations were real businesses serving the metro area
22.8%were in Google's top 3 map results. Given how many businesses each answer named, the most possible was about 56%
0 of 12questions where all four AIs listed the same business first
2 of 250recommendations failed outright when re-checked against live listings

Does ChatGPT Recommend Local Businesses That Actually Exist?

Almost always, in this test. Each finding below stands on its own, so you can quote any one of them with its source.

Key findings

  1. In a two-city test on October 7, 2026, 97.6% of 250 local business recommendations from ChatGPT, Claude, Gemini and Perplexity were real businesses serving the metro area, including 98.3% of ChatGPT's 58 picks.
  2. In the same two-city test, only 22.8% of the 250 AI recommendations were in Google's top 3 map results for the matching search, out of a possible 56.4%. ChatGPT matched most often, at 31.0% of its picks.
  3. Across 12 local home-service questions in Omaha and Boise, the four AI assistants never all listed the same business first, and only 6 of 72 head-to-head comparisons of their first picks matched.
  4. In this two-city test, only 2 of 250 AI recommendations failed outright when re-checked against live listings, and both came from Claude.
  5. In this two-city test, phone-number mismatches outnumbered outright failures: in 7 of 250 AI recommendations, the number given did not match the company's live Google listing, against 2 failures.

Where this started

Before the test, I had a Claude agent ask ChatGPT for the best emergency septic service in Des Moines. One of the options it listed was desmoinesseptic.com, which is my own rank and rent site (a real local septic company takes its calls).

That site doesn't name the company behind it, though. By this test's own rules, an outside checker couldn't verify the operator, so it would count as a lead-gen failure.

In other words, ChatGPT put a site with no named operator right next to real septic companies in an emergency answer. My site showing up there is part of why the test ran in Omaha, Nebraska and Boise, Idaho, where I own nothing.

My sites and client sites are excluded from every number on this page.

The same answer also got a real company's details wrong:

Bob's Septic & Portable Restroom Service: ChatGPT's answer vs its Google listing
DetailIn ChatGPT's answerOn Google Maps (October 7, 2026)
Rating1.0 stars from 2 reviews4.3 stars from 84 reviews
PhoneTwo different numbers: (515) 517-2917 on its business card and (515) 262-9174 in the text(515) 517-2917, so only the card's number matched

When the same setup ran the question again after the test, ChatGPT made Bob's its top pick and left my site out of the answer. The 1.0-star card and the two phone numbers were still there.

What Happened to All 250 Recommendations

Each dot below is one recommendation, grouped by what the checks found. A business can appear more than once, since several were named by more than one assistant.

Nearly every AI pick was a real local business

All 250 AI recommendations, one dot each, by what the checks found

  • In Google's top 3: 57
  • Maps listing in the metro area, not Google's top 3: 178
  • Real operator, Maps listing not found: 9
  • Failed a live re-check: 2
  • Outside the metro area or unresolved: 4
In Google's top 3: 57. Maps listing in the metro area, not Google's top 3: 178. Real operator, Maps listing not found: 9. Failed a live re-check: 2. Outside the metro area or unresolved: 4.
One dot per recommendation (phrasing "Who is the best [service] in [City, ST]?", first run). The gray dots cover 3 real businesses based outside the metro area and 1 that could not be resolved either way. Failures and the closed-business flags were re-checked against live listings on October 7, 2026.

Download chart (PNG)

Show this chart as a table
OutcomeRecommendationsShare
In Google's top 35722.8%
Maps listing in the metro area, not Google's top 317871.2%
Real operator, Maps listing not found93.6%
Failed a live re-check20.8%
Outside the metro area or unresolved41.6%

Most AI Picks Are Real, and Most Aren't Google's Top 3

ChatGPT came closest to Google, and even so, only 31.0% of its picks were in Google's top 3. The other three assistants matched less often.

ChatGPT matched Google's top 3 most often

Share of each assistant's recommendations in Google's top 3. Dark bar: the actual share. Light bar: the most possible, given how many businesses each assistant named.

ChatGPT: 31.0% of a possible 62.1%. Perplexity: 26.7% of a possible 73.3%. Claude: 18.5% of a possible 44.4%. Gemini: 18.2% of a possible 54.5%.0%20%40%60%80%ChatGPT31.0%max 62.1%Perplexity26.7%max 73.3%Claude18.5%max 44.4%Gemini18.2%max 54.5%0%20%40%60%80%ChatGPT31.0%max 62.1%Perplexity26.7%max 73.3%Claude18.5%max 44.4%Gemini18.2%max 54.5%
Google's top 3 = the first three local pack results for the matching search, such as "roofing company Omaha NE", pulled the same day. An assistant that names more businesses per answer has a lower ceiling, because the map pack has only three slots.

Download chart (PNG)

Show this chart as a table
AssistantIn Google's top 3Most possible
ChatGPT31.0%62.1%
Perplexity26.7%73.3%
Claude18.5%44.4%
Gemini18.2%54.5%
Recommendations by assistant, checked October 7, 2026.
AssistantRecommendationsIn Google's top 3Matched a Maps listing in the metro areaReal, serving the metro area
ChatGPT5831.0%98.3%98.3%
Perplexity4526.7%95.6%97.8%
Claude8118.5%87.7%95.1%
Gemini6618.2%97.0%100.0%

Claude's lower match rate comes mostly from naming smaller operators that are real but had no Maps listing the checks could find. Counting those, 95.1% of its picks were real businesses serving the metro area.

Overlap by trade

Top-3 overlap varied more across trades than across assistants in this sample. Roofing barely overlapped for any of the four.

Roofing picks almost never matched Google's top 3, for any assistant

Recommendations in Google's top 3, out of each assistant's recommendations for that trade. Darker = higher share.

  • 0
  • Under 15%
  • 15 to 29%
  • 30 to 44%
  • 45% or more
Septic pumping: ChatGPT 4 of 10, Perplexity 3 of 7, Claude 5 of 14, Gemini 4 of 8, all four 16 of 39 (41.0%). Garage door repair: ChatGPT 4 of 10, Perplexity 4 of 8, Claude 2 of 14, Gemini 4 of 11, all four 14 of 43 (32.6%). HVAC repair: ChatGPT 2 of 9, Perplexity 2 of 10, Claude 3 of 15, Gemini 2 of 10, all four 9 of 44 (20.5%). Junk removal: ChatGPT 5 of 11, Perplexity 1 of 9, Claude 2 of 14, Gemini 1 of 12, all four 9 of 46 (19.6%). Emergency plumber: ChatGPT 3 of 10, Perplexity 2 of 4, Claude 3 of 15, Gemini 0 of 12, all four 8 of 41 (19.5%). Roofing: ChatGPT 0 of 8, Perplexity 0 of 7, Claude 0 of 9, Gemini 1 of 13, all four 1 of 37 (2.7%).ChatGPTPerplexityClaudeGeminiAll fourSeptic pumping4 of 103 of 75 of 144 of 816 of 39Garage door repair4 of 104 of 82 of 144 of 1114 of 43HVAC repair2 of 92 of 103 of 152 of 109 of 44Junk removal5 of 111 of 92 of 141 of 129 of 46Emergency plumber3 of 102 of 43 of 150 of 128 of 41Roofing0 of 80 of 70 of 91 of 131 of 37ChatGPTPerplexityClaudeGeminiSeptic pumpingAll four: 41.0%4/103/75/144/8Garage door repairAll four: 32.6%4/104/82/144/11HVAC repairAll four: 20.5%2/92/103/152/10Junk removalAll four: 19.6%5/111/92/141/12Emergency plumberAll four: 19.5%3/102/43/150/12RoofingAll four: 2.7%0/80/70/91/13
Small samples: each assistant's cell covers 4 to 15 recommendations, so read the rows, not single cells. All four = both cities and every assistant combined.

Download chart (PNG)

Show this chart as a table
TradeChatGPTPerplexityClaudeGeminiAll four
Septic pumping4 of 103 of 75 of 144 of 816 of 39 (41.0%)
Garage door repair4 of 104 of 82 of 144 of 1114 of 43 (32.6%)
HVAC repair2 of 92 of 103 of 152 of 109 of 44 (20.5%)
Junk removal5 of 111 of 92 of 141 of 129 of 46 (19.6%)
Emergency plumber3 of 102 of 43 of 150 of 128 of 41 (19.5%)
Roofing0 of 80 of 70 of 91 of 131 of 37 (2.7%)

On the original Omaha roofing search, none of the four named any of the three roofers Google showed first. When Claude agents pulled that exact search twice more the same afternoon, three minutes apart, the second pull had swapped one of the three for McCoy Roofing, which was ChatGPT's first pick.

Google's map pack moves too. Treat any overlap figure, including mine, as a snapshot.

The two cities landed within a few points of each other on every measure. With only two cities, that is a consistency check, not proof the pattern holds everywhere.

Omaha vs Boise, all four assistants combined
Share of AI recommendationsOmaha, NE (124)Boise, ID (126)
In Google's top 322.6%23.0%
Matched a Maps listing in the metro area91.9%96.0%
Real businesses serving the metro area96.8%98.4%

Two more numbers came up after the test rather than being planned. Looking from Google's side, the four AIs together mentioned 29 of the 36 businesses in Google's top 3 (three per search, 12 searches) at least once.

Those top-3 businesses were also more likely to be listed first. They made up 35.4% of first picks, compared with 22.8% of all recommendations.

For a related but different question, SOCi studied multi-location brands and found ChatGPT recommends just 1.2% of brand locations, compared with an average 35.9% appearance rate in Google's local 3-pack.

Explore Every Answer

Pick a city and a trade to see Google's top 3 next to every business each assistant recommended, in the order it listed them.

Google's top 3

  1. Crown Pumping (5 stars, 207 reviews)
  2. Omaha Pumping Services (5 stars, 1 review)
  3. Midland Pumping Service (4.8 stars, 39 reviews)

ChatGPT

  1. Crown PumpingGoogle's top 3
  2. Midland Pumping ServiceGoogle's top 3
  3. S & S Pumping Service
  4. AAA Sewer & Septic LLC
  5. Nebraska SepticOutside metro area

Claude

  1. Midland Pumping ServiceGoogle's top 3
  2. Crown Pumping LLCGoogle's top 3
  3. CB Septic IncMaps listing not found
  4. Great Plains Septic
  5. S & S Pumping Service
  6. Bill's Sewer ServiceMaps listing not found
  7. Allen SepticOutside metro area
  8. Ruda SepticOutside metro area

Gemini

  1. Crown Pumping LLCGoogle's top 3
  2. Norm's Pumping & Septic Service
  3. Omaha PumpingGoogle's top 3
  4. G&T Services

Perplexity

  1. Crown Pumping LLCGoogle's top 3
  2. Midland Pumping ServiceGoogle's top 3

Google search: "septic tank pumping company Omaha NE". Business names are as each assistant wrote them. Untagged names matched a Google Maps listing in the metro area, per the Claude-agent checks. Google rounds review counts above 1,000.

The Four AIs Rarely List the Same Local Business First

Across these 12 questions, the four assistants usually listed different businesses first. Each assistant's first pick was compared with every other assistant's first pick, question by question.

6 of 72

head-to-head comparisons of first picks matched (8.3%). The grid below shows where.

Only 4 of the 12 questions had any two AIs list the same business first

Each dot is one assistant's first pick for one question.

  • Another assistant listed the same business first
  • No other assistant listed it first
Omaha septic pumping: ChatGPT Crown Pumping (shared), Perplexity Crown Pumping LLC (shared), Claude Midland Pumping Service, Gemini Crown Pumping LLC (shared). Omaha garage door repair: ChatGPT Spring Kings Omaha, Perplexity Garage Doors Of Omaha, Claude Norm's Door Service, Gemini Omaha Door & Window. Omaha hvac repair: ChatGPT Skradski Heating & Cooling, Perplexity Xtreme Heating & Cooling, Claude Getzschman Heating & Air Conditioning, Gemini Complete Comfort Heating and Cooling. Omaha junk removal: ChatGPT 1-800-GOT-JUNK? Omaha, Perplexity Horizon Junk Pros, Claude R & R Hauling, Gemini Junk King Omaha. Omaha emergency plumber: ChatGPT Omaha Emergency Neighborhood Plumbers, Perplexity PlumbRite, Claude Roto-Rooter, Gemini Eyman Plumbing Heating & Air. Omaha roofing: ChatGPT McCoy Roofing, Perplexity Erie Home, Claude Royalty Roofing (shared), Gemini Royalty Roofing & Renovations (shared). Boise septic pumping: ChatGPT ABC Pumping Service (shared), Perplexity American Septic, Claude ABC Pumping Service (shared), Gemini Sweet's Services. Boise garage door repair: ChatGPT A1 Garage Door Service, Perplexity Garage Door Store Boise, Claude Legacy Garage Doors, Gemini Overhead Door Company of Southwestern Idaho. Boise hvac repair: ChatGPT Owyhee Heating and Air Conditioning, Perplexity Wickstrom Plumbing Heating Cooling, Claude Western Heating & Air Conditioning, Gemini A-1 Heating, Air Conditioning & Electric. Boise junk removal: ChatGPT Junk Holler, Perplexity 1-800-GOT-JUNK? Boise, Claude Junk King Boise, Gemini Full House Junk Removal. Boise emergency plumber: ChatGPT Roto-Rooter Plumbing & Water Cleanup (shared), Perplexity Roto-Rooter (shared), Claude All Star Plumbing & Restoration, Gemini Diamond Heating, Cooling, Plumbing & Electric. Boise roofing: ChatGPT Bork Roofing, Perplexity Blue Goat Roofing, Claude E & H Roofing, Gemini Signature Roofing.ChatGPTPerplexityClaudeGeminiOMAHA, NESeptic pumpingGarage door repairHVAC repairJunk removalEmergency plumberRoofingBOISE, IDSeptic pumpingGarage door repairHVAC repairJunk removalEmergency plumberRoofingChatGPTPerplexityClaudeGeminiOMAHA, NESeptic pumpingGarage door repairHVAC repairJunk removalEmergency plumberRoofingBOISE, IDSeptic pumpingGarage door repairHVAC repairJunk removalEmergency plumberRoofing
Matches use the business name each assistant gave, cleaned of suffixes like "LLC". Hover or tap a dot to see the business. The full list is in the table below and in the explorer above.

Download chart (PNG)

Show this chart as a table
QuestionChatGPT's first pickPerplexity's first pickClaude's first pickGemini's first pick
Omaha septic pumpingCrown Pumping (shared)Crown Pumping LLC (shared)Midland Pumping ServiceCrown Pumping LLC (shared)
Omaha garage door repairSpring Kings OmahaGarage Doors Of OmahaNorm's Door ServiceOmaha Door & Window
Omaha hvac repairSkradski Heating & CoolingXtreme Heating & CoolingGetzschman Heating & Air ConditioningComplete Comfort Heating and Cooling
Omaha junk removal1-800-GOT-JUNK? OmahaHorizon Junk ProsR & R HaulingJunk King Omaha
Omaha emergency plumberOmaha Emergency Neighborhood PlumbersPlumbRiteRoto-RooterEyman Plumbing Heating & Air
Omaha roofingMcCoy RoofingErie HomeRoyalty Roofing (shared)Royalty Roofing & Renovations (shared)
Boise septic pumpingABC Pumping Service (shared)American SepticABC Pumping Service (shared)Sweet's Services
Boise garage door repairA1 Garage Door ServiceGarage Door Store BoiseLegacy Garage DoorsOverhead Door Company of Southwestern Idaho
Boise hvac repairOwyhee Heating and Air ConditioningWickstrom Plumbing Heating CoolingWestern Heating & Air ConditioningA-1 Heating, Air Conditioning & Electric
Boise junk removalJunk Holler1-800-GOT-JUNK? BoiseJunk King BoiseFull House Junk Removal
Boise emergency plumberRoto-Rooter Plumbing & Water Cleanup (shared)Roto-Rooter (shared)All Star Plumbing & RestorationDiamond Heating, Cooling, Plumbing & Electric
Boise roofingBork RoofingBlue Goat RoofingE & H RoofingSignature Roofing

The clearest agreement was Omaha septic, where ChatGPT, Gemini and Perplexity all put Crown Pumping first. Claude said there was no single best and listed Midland Pumping Service first.

This lines up with the Steady Demand AI Citation Ledger, which found ChatGPT and Gemini name the same top business only 4.9% of the time.

The answers also drift. The test reran 12 of the 48 answers, and an average of 79.1% of the businesses from the first run showed up again.

Where the AI Recommendations Went Wrong

A bad recommendation can waste a homeowner's call and send the job away from a real business. These were the clearest misses, each re-checked by Claude agents against live listings on October 7, 2026.

FailedClaude · Boise HVAC

Carlon Heating & Air

What Claude saidIt has 300+ reviews, was established in 1948, holds an A+ BBB rating.
What the live check found

No current Google listing. An older Maps record marks it permanently closed, and its old web address was serving a gambling spam page when checked.

FailedClaude · Omaha garage door repair

A1 Garage Door Service, LLC

What the live check found

A real multi-state company, but its Omaha location shows as closed on Yelp and its own site lists no Nebraska service areas.

Wrong detailPerplexity · Omaha junk removal

Horizon Junk Removal (Perplexity called it “Horizon Junk Pros”)

What the live check found

The phone number Perplexity gave matched neither the company's Google listing nor its Facebook page.

Wrong detailChatGPT · Omaha HVAC

Anderson Comfort

What the live check found

A real company with a 5.0 rating, but the website ChatGPT gave for it now redirects to a domain marketplace page.

Correction to this testClaude · Boise septic

ASAP Septic

What Claude saida 5.0 rating across 47 reviews
What the live check found

An old cached record said permanently closed. The live listing shows it open with 5.0 stars from 47 reviews. Claude was right, and the first-pass check was wrong.

The 7 phone-number mismatches can mean an old number, a tracking line or a plain mistake. Either way, call the number on the listing.

The ASAP Septic card is a correction of the test itself. The Claude agents' first pass flagged 7 recommendations as closed, not found or lead-gen, and a live re-check cleared 4 of them, with 1 more unresolved either way.

For a separate check of a different engine, Whito tested Google AI Mode in 20 UK towns and found 5.2% of the businesses it called trustworthy could not be verified as active companies trading in that town. It checked company registry records, so its unverifiable category is not directly comparable to the 2 failures here.

What This Means If You Run a Local Business

The errors documented here included stale or conflicting details: an old closed listing, a dead domain, a phone number that differs from the listing. The test can't prove what caused each error, but those gaps are the ones you control.

Does ChatGPT recommend local businesses diagram: check an AI pick against its Maps listing, website and phone before calling
The misses in this test involved a listing, a website or a phone number. If your listing, site and phone agree, the AI has less to get wrong.
  • Keep your Google Business Profile current, especially your hours and phone number.
  • Use one main phone number everywhere: your listing, your site, Facebook and every directory.
  • Keep your domain renewed. A dead domain can still get handed to customers.
  • Close or merge old listings when you move, rebrand or sell, so a stale record cannot outlive the real one.

None of that is new. My take from this pilot: for local service businesses, getting named correctly by AI starts with the same fundamentals that already pay off on Google.

Getting named more often is a different question. I covered which sources ChatGPT actually reads in my test of how to rank in ChatGPT.

If you are on the other side of this, looking for a plumber at midnight, open the business's Google listing before you call the number an AI gave you.

How I Ran This Field Test

  • Questions. "Who is the best [service] in [City, ST]?" for 6 urgent home services (septic pumping, emergency plumber, HVAC repair, roofer, junk removal, garage door repair) in Omaha and Boise. That is 12 questions and 48 answers.
  • Assistants. ChatGPT through its consumer app with web search on, collected with DataForSEO's AI Optimization API. Claude, Gemini and Perplexity through their APIs, with web search confirmed on every answer but no user location. The results apply to these four setups and may differ in the consumer apps.
  • Recommendations. Every business an answer presented as a recommendation, which came to 250 entries covering about 160 different businesses. Businesses named only as warnings did not count, and Claude's first-listed business counted as its first pick.
  • Google comparison. The local pack top 3 for the matching Google search (for example, "roofing company Omaha NE"), pulled the same day in each city.
  • Checks. Claude agents matched each recommendation to Google Maps listings and websites, checked the 57 top-3 matches against the saved search results, live-checked every closed, lead-gen or not-found flag, and confirmed the 7 phone mismatches against live listings.
  • Reliability. A separate Claude model re-checked 50 randomly drawn recommendations against live listings or the saved search results and agreed with the labels on 48 (96%), above the 90% bar set in advance. Both disagreements were the same plumber, a real business matched to the wrong Maps listing on its name alone. That fix is in the numbers above, but other name-only Maps matches were not re-checked one by one, so the 94.0% Maps figure could run slightly high.
  • One inconsistency. Claude sometimes included names from a third-party list. Eight of those were counted in three answers but left out of a roofing answer, and removing them moves every headline figure by about a point or less.
  • Disclosure. Claude ran the checks and is also one of the four assistants tested, and I sell a Claude Code product. The same rules applied to every assistant, and Claude finished near the bottom on top-3 overlap and had both hard failures.

Before collecting any data, the plan set criteria for expanding to a full 10-city study. Top-3 overlap had to land at 50% or above, or 20% or below, or hard failures had to reach 10% (closed, not found, or a lead-gen site with no verifiable operating company behind it).

Overlap came in at 22.8% and hard failures far below 10%, and the pilot also ran over the plan's cost and workload limits. So the test stopped at two cities, and I'm reporting it as a field test.

The agreement figures, the 29-of-36 coverage figure, the first-pick figure, the city comparison and the trade-by-assistant grid were not part of that plan, so treat them as observations. The bigger limits: two cities, 12 questions, one day.

The Picks Were Real. The Details Weren't Always.

Does ChatGPT recommend local businesses that are real? In this test, the four assistants named real, local businesses 97.6% of the time.

I ran this test from Claude Code. The marketing skills I run there every week are in my Claude Code Skills Stack, which is my own product, so weigh my Claude Code bias accordingly.

The assistants often disagreed with Google and with each other, and the details they repeated were not always current. The part you control is whether those details are right when an AI does name you.

Does ChatGPT Recommend Local Businesses FAQs

Can ChatGPT recommend a local business that is closed?

This test did not catch ChatGPT recommending a closed business. Across 250 recommendations from ChatGPT, Claude, Gemini and Perplexity, only 2 failed outright, and both came from Claude. ChatGPT's misses were mostly about details, like a website that now redirects to a domain marketplace and phone numbers that did not match the live Google listing, plus one business based outside the metro area.

Do ChatGPT, Gemini, Claude and Perplexity recommend the same businesses?

Rarely as the first business they list. Across 12 local questions, the four assistants never all listed the same business first, and only 6 of 72 head-to-head comparisons of first picks matched. Each answer named about five businesses, so a company can show up in several answers without being anyone's first choice. Steady Demand found a similar pattern for ChatGPT and Gemini, at 4.9%.

What share of AI local recommendations are in Google's top 3?

In this two-city test, 22.8% of 250 AI recommendations were in Google's top 3 for the matching search, out of a possible 56.4%. ChatGPT matched most often, at 31.0% of its picks. The four assistants still mentioned 29 of Google's 36 top-3 businesses at least once, and 35.4% of their first picks came from that top 3. This test cannot show that the ranking caused the mention.

How often do AI assistants give the wrong phone number for a local business?

In this test, 7 of 250 recommendations (2.8%) came with a phone number that did not match the main number on the company's live Google listing. Some of those may be tracking lines rather than errors, so check the number on the business's Google listing before you call.

Cite This Field Test

You are welcome to use any figure, chart or table here with a link to this page. Please keep the collection date and the two-city scope with any figure you use, and find the full dataset in the downloads at the top.

Doser, R. (2026). Does ChatGPT recommend local businesses? A field test of 250 AI recommendations in Omaha and Boise. ryandoser.com. https://ryandoser.com/research/does-chatgpt-recommend-local-businesses/

Chart images

Each chart below is a 1600-pixel PNG with its title and source line built in.

The AI SEO System I use to rank in Google & AI search. Runs in Claude or any AI setup.

Get my system

Free AI Marketing Guide

Get the free AI Marketing Guide

The systems I use to run my agency solo. No coding needed to start:

Join over 10,000 subscribers getting my weekly AI marketing newsletter. Unsubscribe anytime. Privacy policy.