Short answer: no, they usually do not agree. In a daily record kept by NameBeam, the AI engines picked the same most-named business in only 14 of 36 markets. In a two-city field test by Ryan Doser, the four assistants never all listed the same business first on any of 12 questions. And in Searchable's data, ChatGPT and Gemini overlapped on a median of 26% of the local service businesses they recommended.
If you run a roofing company, a dental practice, a restaurant, or a retail shop in Northeast Indiana, this changes the question you should be asking. “Does AI recommend us?” used to sound like one question. These studies suggest it is several, and the answers can differ by assistant and by day.
All three studies are from small, vendor-run or individual research efforts, and none of them covers Northeast Indiana. We say what each one measured, where it is thin, and what a local owner can reasonably do about it without paying for anything.
Key Takeaways
- NameBeam asked four AI engines the same buyer questions daily from September 1, 2026. Through October 11, only 14 of 36 markets had two or more engines picking the same top business.
- In Doser's Omaha and Boise test, 97.6% of 250 AI picks were real businesses, but only 22.8% were in Google's top 3 map results.
- Searchable measured a 26% median overlap between ChatGPT and Gemini for local service businesses, against 46% for B2B software companies.
- None of the three studies explains why an engine picks a business. Treat each as a snapshot, not a ranking factor.
- Check each assistant separately, on several days, and fix the basics an assistant would look for: an accurate profile, a consistent phone number, real reviews, and a clear service page.
Do ChatGPT, Gemini, and Perplexity Name the Same Local Business?
The most systematic look comes from NameBeam, a small Wyoming company that publishes a public record of how AI engines name local businesses. Its founder, Terry Carter, summarized the first six weeks of data in an October 2026 article on Digital Information World. Since September 1, 2026, NameBeam has asked ChatGPT, Perplexity, Gemini and Claude the same buyer questions in 46 U.S. market-and-trade pairs, once a day, and saved every answer.
What NameBeam Reported for September 1 to October 11
- Markets with any agreement: in 36 of the 46 markets, two or more engines named at least one business. In only 14 of those 36 did two or more engines pick the same business as their most named answer. The other 22 split.
- How often each engine named anyone: Perplexity named at least one business in 37 of the 46 markets, Gemini in 32, ChatGPT in 24 and Claude in 9.
- How steady the top name was: Perplexity's most-named business appeared on 56% of its answer days on average. Gemini's was 72% across the 40 markets where it answered. ChatGPT's was 20% and Claude's 7%.
That last row is worth reading twice. It is not only that the engines disagree with each other. Some of them also disagree with themselves from one day to the next. An owner who sees their business named once in ChatGPT has learned much less than they think.
NameBeam is careful about what it does not claim. In the article, Carter writes that the record does not say why an engine picks a company and does not say that a website change or a review caused a name to appear. It records what was said, on which day, by which engine. The company also sells that record to agencies and platforms that serve local businesses, so it has a commercial interest in the topic. We use its counts for what they are and nothing more.
Carter's own framing is the practical takeaway. “Does AI recommend us?” is not one question anymore. It is four, and they can have four different answers.
What Did a Two-City Test in Omaha and Boise Show?
A second angle comes from independent researcher Ryan Doser, who published a field test on October 7, 2026 (updated October 8). He asked ChatGPT, Claude, Gemini and Perplexity “Who is the best [service] in [City, ST]?” for six home-service trades in Omaha, Nebraska and Boise, Idaho, which made 12 questions and 250 recommendations. Each recommendation was then checked against live Google listings.
Doser discloses a conflict up front. He owns rank-and-rent sites, and one of them showed up in a ChatGPT answer during a check he ran before the test. He excluded his own sites and any client sites from every number on the page and ran the study in cities where he owns nothing.
The Headline Numbers From the Test
- 97.6% of the 250 recommendations were real businesses serving the metro area. Only 2 of 250 failed outright when re-checked, and both came from Claude.
- 22.8% were in Google's top 3 map results for the matching search, out of a possible 56.4% given how many businesses each answer named.
- 0 of 12 questions had all four assistants list the same business first. Only 6 of 72 head-to-head comparisons of first picks matched (8.3%), and only 4 of the 12 questions had any two assistants agree on the first pick.
- 7 of 250 recommendations gave a phone number that did not match the company's live Google listing.
Two findings matter for a small business. The first is reassuring: the assistants in this test were not inventing companies. Nearly every pick was a real business with a real listing. The second is less comfortable: being a strong local-map result did not make a business the AI's pick. Doser did find that the businesses in Google's top 3 were more likely to be listed first, making up 35.4% of first picks versus 22.8% of all recommendations, and that the four assistants together mentioned 29 of the 36 top-3 businesses at least once. So the map pack and the AI answers are related, but they are far from the same list.
Doser also shows how wrong a single detail can be. In a Des Moines check run before the main test, a ChatGPT answer showed a septic company with a 1.0-star rating from 2 reviews and two different phone numbers, when the company's Google listing showed 4.3 stars from 84 reviews. Only one of the two numbers matched. That is one answer, not a rate, but it is the kind of error an owner would never see unless they looked. If you have been through our post on how customers verify AI picks before they call, this is the other half of the story: the customer checks, and sometimes the check finds a wrong number.
For context, Doser also points to a separate SOCi study of multi-location brands, which found that ChatGPT recommends just 1.2% of brand locations compared with an average 35.9% appearance rate in Google's local 3-pack. That is a different measure on a different sample, so we cite it as Doser reports it and do not combine it with the numbers above.
Does the Disagreement Differ From Trade to Trade?
Yes, in these samples, and by a lot. Doser broke out how many AI picks landed in Google's top 3 by trade. Septic pumping had the most overlap, with 16 of 39 picks (41.0%) in the top 3. Garage door repair followed at 14 of 43 (32.6%), HVAC repair at 9 of 44 (20.5%), junk removal at 9 of 46 (19.6%), and emergency plumbing at 8 of 41 (19.5%). Roofing sat at the bottom with 1 of 37 (2.7%).
On the original Omaha roofing search, none of the four assistants named any of the three roofers Google showed first. Doser then had agents pull the same Google search twice more that afternoon, three minutes apart, and the second pull swapped one of the three for McCoy Roofing, which had been ChatGPT's first pick. His own caution applies: Google's map pack moves too, so treat any overlap figure, his included, as a snapshot.
NameBeam's roofing data points the same direction from a different angle. In Dallas, Perplexity named Arrington Roofing on 27 of its 37 answer days. In Denver it named Gates Enterprises on 35 of 37, and in Miami it named Extreme Roofing on 26 of 37. ChatGPT and Claude named none of those three companies on any of their 40 answer days, and Gemini, which was asked fewer times because of its quota, named them on 0 of 8, 0 of 9 and 0 of 10 days. Same city, same question, same six weeks: one engine returned a steady name, and the others returned nothing for it at all.
We would not read much into why roofing looks different. Doser warns that each assistant's cell covers only 4 to 15 recommendations, so the rows are more trustworthy than any single cell, and the two cities he tested landed within a few points of each other on every measure but are still only two cities. What the roofing rows do show is that a trade with high-ticket jobs and a lot of competition can look very different across engines. If you work in a trade like that, a single check will not tell you much.
The ads side of the same trades is moving as well. We covered one change in our post on Local Services Ads missed-call charges. The point is the same in both places: the way a customer first meets your business now has more doors than it used to.
What Did Searchable Find About ChatGPT and Gemini Overlap?
Searchable, a company that sells AI-visibility tracking, published a comparison of the businesses ChatGPT and Gemini consistently mention for the same unbranded prompts. For each prompt, it looked at the businesses each engine mentioned in at least half of its answers, averaged that overlap for each brand, and took the median across brands.
Searchable's Overlap Figures
- Local service businesses: a median overlap of 26% from July 28 to October 5, 2026, and 33% in the ten weeks to September 15.
- B2B software companies: 46% in both periods.
- Sample: 29 local service businesses and 41 B2B software companies in the later window, and 28 and 44 in the earlier one, all brands tracked in Searchable's own product.
- Size of the gap: between 8 and 21 percentage points depending on the analysis settings, which Searchable describes as a consistent direction with an uncertain size.
Searchable also tested whether location alone explained the gap, since local businesses are tied to a place. It does not, according to the company, because hotels, which are just as tied to a place, showed overlap as high as B2B software. The company assigned industries by hand from each brand's name and website, and 29 businesses is a small group, so this is a hint more than a finding. It does line up with the other two studies, which is why we mention it.
The advice Searchable draws is the simplest in this post: check ChatGPT, Gemini and Google's AI Overviews separately, rather than assuming one stands in for the rest. We agree, and we would add Perplexity. We also covered a related effect on Google's own side in our post on AI Overviews and business-name searches, and the way ChatGPT leans on one particular listing source in our post on ChatGPT local results and Yelp.
What Do These Studies Not Prove?
Quite a lot, which is worth saying plainly before anyone builds a plan on them. Here are the limits we would keep in mind.
Six Limits Worth Keeping in Mind
- None of them is about Northeast Indiana. NameBeam covers 46 U.S. market-and-trade pairs, Doser tested Omaha and Boise, and Searchable tracks brands in its own product. We do not know which markets are in NameBeam's set beyond the ones named.
- They are small samples. Doser's test is 12 questions and 250 recommendations. Searchable compares 29 local businesses. NameBeam counts days for individual businesses.
- Two of the three are from vendors. NameBeam sells its record to agencies and platforms, and Searchable sells AI-visibility tracking. Both sell products in this space. Doser is independent but owns rank-and-rent lead-generation sites.
- They do not say why. No study tests what makes an engine pick one business over another. NameBeam says so itself.
- They are snapshots. The assistants change, and so does Google's map pack. Doser saw it change within minutes.
- They count who was named, not who got the call. Nothing here measures how many customers used an AI answer to choose a business.
The honest summary is that three small, imperfect measurements all point the same way. When different people use different methods and land in the same direction, that is worth acting on carefully. It is not worth acting on dramatically.
How Can a Northeast Indiana Business Check Its Own Picture?
This section is our judgment, built from what the three studies did, not a research finding. You can run the same kind of check yourself in an afternoon with no paid tool.
A Simple Four-Assistant Spot Check
- Write three questions the way a customer would. For example, “Who is the best roofer in Auburn, Indiana?” or “Which dentist in Fort Wayne takes emergency appointments?” Use your own trade and town. Doser's wording was simply “Who is the best [service] in [City, ST]?”
- Ask each one in ChatGPT, Gemini, Perplexity and Claude. Record who is named first, who else is named, and whether you appear. Do the same Google search and note the top three map results.
- Repeat on three different days. NameBeam's data shows some engines' top names move day to day, so one look is not a result.
- Check every detail the assistant gives about you. Compare the phone number, hours, address and rating against your live Google Business Profile. Doser found phone mismatches in 7 of 250 picks, more often than outright failures.
- Write down the pattern, not the score. You are looking for which assistants know you exist and which get something wrong.
Then fix the basics an assistant could be working from. We cannot say these cause a business to be named, because none of the studies tested that. But they are cheap, they are worth doing for customers anyway, and they take away easy ways for an answer to be wrong.
Make your business facts identical everywhere. The same name, address, phone number and hours on your website, your Google Business Profile, and the major directories. A mismatch between what an assistant reads and what Google shows is exactly the kind of error Doser found. Our local SEO service starts with this kind of cleanup.
Be specific about what you do and where. A page that names your services and the towns you serve in plain sentences gives an assistant something concrete to quote. We looked at the services-and-description side of this in our post on the Google Business Profile service description field.
Keep collecting real reviews. In Doser's septic example, the answer showed 2 reviews where Google showed 84. We cannot say why, but a steady flow of genuine reviews at least gives every source the same story to tell.
Re-check monthly. Put the spot check on your calendar. Because the answers move, the trend over a few months says more than any one afternoon. If you want help running it or reading the results, that is part of what our answer engine optimization service covers.
Want to Know What the Four Assistants Say About You?
Button Block runs this kind of spot check for Northeast Indiana businesses, shows you exactly what each assistant says, and fixes the profile and website details that are wrong. No promises about rankings, just an accurate picture and a to-do list.
Frequently Asked Questions
- Do ChatGPT and Gemini recommend the same local businesses?
- Usually not. Searchable, an AI-visibility tracking company, found that the median overlap between the local service businesses ChatGPT and Gemini recommended for the same prompts was 26% from July 28 to October 5, 2026 (33% in the ten weeks to September 15). For B2B software companies it was 46%. The sample was small, 29 local service businesses, and Searchable says the size of the gap ranged from 8 to 21 percentage points depending on how it was analyzed.
- Why do AI assistants name different businesses for the same question?
- None of the three studies in this post tests why, and we are not going to guess. NameBeam says outright that its record does not say why an engine picks a company. What the data does show is that the engines disagree often, that their answers also change from day to day, and that being named by one tells you little about the others.
- Are AI local recommendations real businesses or made up?
- In one October 2026 field test by Ryan Doser, 97.6% of 250 recommendations from ChatGPT, Claude, Gemini and Perplexity were real businesses serving the metro area, and only 2 of 250 failed outright when re-checked against live listings. Details were a bigger problem than existence: in 7 of 250 recommendations the phone number given did not match the business's live Google listing.
- Do the AI picks match what Google shows in the local map pack?
- Mostly not. In the same Omaha and Boise 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%, and Gemini least at 18.2%. Separately, SOCi has reported that ChatGPT recommends just 1.2% of brand locations versus a 35.9% average appearance rate in Google's local 3-pack, which is a different measure on a different sample.
- How do I check whether AI assistants recommend my business?
- Ask the same plain question a customer would, such as the service plus your town, in ChatGPT, Gemini, Perplexity and Claude, and record who is named first and whether you appear. Repeat it on several different days, because the answers change. Keep a dated log. A one-time check is a snapshot, not a result.
- Is one AI engine more important than the others for local businesses?
- The evidence in these studies does not rank them for you. NameBeam found Perplexity named at least one business in 37 of 46 markets, Gemini in 32, ChatGPT in 24 and Claude in 9, but that describes how often each answered with a name, not how many of your customers use each one. Find out which tools your own customers mention, and check all of them.
- Can a small business control which businesses an AI recommends?
- No one can promise that, and none of these studies tested what causes a business to be named. What you can control is the evidence an assistant could find: a complete and accurate Google Business Profile, a consistent name, address and phone number everywhere, real reviews, and a website that states your services and service area clearly. Treat those as good practice, not a guarantee.
Sources & Further Reading
- Terry Carter (NameBeam), “Four AI engines were asked the same local question every day. They rarely agreed,” Digital Information World: digitalinformationworld.com (Oct. 2026): the 46-market daily record, the 14-of-36 agreement figure, and the roofing examples. NameBeam sells its data to agencies and platforms.
- Ryan Doser, “Does ChatGPT Recommend Local Businesses? 250 AI Picks Tested”: ryandoser.com (published Oct. 7, 2026, updated Oct. 8): the Omaha and Boise field test, trade-by-trade overlap, and the phone-number mismatches. The SOCi comparison is cited as Doser reports it.
- Searchable, “AI Search Optimization for Local Businesses in 2026”: searchable.com (Oct. 2026): the ChatGPT and Gemini overlap figures, sample sizes, and method. Searchable sells AI-visibility tracking.