AI Visibility Study 2026

Who does AI recommend? 10,016 buyer questions about 82 local businesses

We asked an AI assistant with live web search the questions buyers ask before they hire a local business, without naming the business. This is how often it was the answer.

151 of 8,533answers named the independent business being tested
36 of 67independent businesses were not named in a single answer
77 of 82tests where one competitor was named more often than the tested business

Key findings

Between September 24 and November 13, 2026, we asked an AI assistant with live web search 10,016 questions that buyers ask before they hire a local business. The questions came from 82 tests, one test per business, in 8 sectors. No question contained the name of the business being tested. For each answer, code checked whether that business was one of the companies the AI recommended.

  • An independent local business is almost never the answer. The 67 independent businesses were named in 151 of 8,533 answers. 36 of the 67 were not named in a single answer.
  • A known brand name changes the result. The 15 franchise locations and offices of larger brands were named in 299 of 1,483 answers.
  • The same few companies are named again and again. In 48 of 82 tests, one company appeared in at least 1 of every 4 answers. In 77 of 82 tests, the most-named competitor was named more often than the tested business.
  • National chains take a large share of local answers. At least 3,313 of 10,016 answers named a well-known national chain or franchise brand. In 29 of 82 tests, a national brand was the most-named company.
  • Advice questions almost never lead to the tested business. For questions about price, time, rules, risk and how to choose, the tested business was named in 9 of 1,116 answers, even though those answers named companies 2,646 times.
  • Sectors differ, but no sector is safe. Independent law firms were named in 3 of 510 answers and independent real estate and property management businesses in 3 of 740. Home services did better, and still reached only 44 of 3,223.

Why we ran this study

When a person types a question into a search engine, the results page lists ten or more companies and the person chooses. When a person asks an AI assistant the same question, the answer names a few companies and explains why. In this study an answer named three companies at most. A business that is not one of them is not seen by that buyer at all.

Owners of small businesses usually do not know where they stand. They test it by asking an AI assistant about their own company by name, and the assistant answers politely about that company. That tells them nothing about a new customer, who does not know the name yet. This study measures the other case: the question a stranger asks, such as "Who is the best 24/7 emergency plumber in Dallas, TX?" or "How much does roof repair cost in Dallas for a small leak?"

Finding 1: An independent local business is almost never the answer

67 of the 82 tested businesses are independent: a local owner, no franchise and no larger brand behind the name. Together they were tested on 8,533 questions. They were named in 151 of 8,533 answers, and named first in 84.

Independent businessesBusinesses
Not named in any answer36
Named in 1 or 2 answers18
Named in 3 to 9 answers9
Named in 10 or more answers4
All independent businesses67

The few businesses that did well carry most of the total. The 4 independent businesses with 10 or more mentions account for 88 of the 151. The other 63 were named in 63 of 7,719 answers.

This result says nothing about the quality of their work. It says that when a stranger asks, these businesses are not on the pages the AI assistant reads and chooses from.

Finding 2: A known brand name changes the result

15 of the tested businesses are a franchise location or a local office of a larger brand, for example a national home-service franchise or a regional bank. For these, a mention of the brand counted. They were named in 299 of 1,483 answers, and 8 of the 15 were named in 10 or more answers.

One gutter franchise alone was named in 82 of 95 answers. Without it, the other 14 were named in 217 of 1,388 answers, which is still far above the independent group. A brand is not a guarantee: 2 of the 15 were not named at all.

A likely reason: a national brand usually has a page for each city, each service and each location, and it is mentioned on many other websites. That gives the AI assistant something to find for almost any local question in the trade.

Finding 3: The same few companies are named again and again

Across all tests the answers named other companies 24,692 times. Those mentions are not spread evenly.

  • In 48 of 82 tests, one company was named in at least 1 of every 4 answers. In 13 tests, one company was named in at least half of the answers.
  • The three most-named companies of each test collected 6,659 of the 24,692 mentions.
  • At the other end, 5,635 of the 8,857 companies that appeared in a test appeared in only one answer.
  • In 77 of 82 tests, the most-named competitor was named more often than the tested business.

So each local market has a short list of companies the AI assistant returns to for many different questions, and a long list of companies it mentions once.

Finding 4: National chains take a large share of local answers

We counted how often an answer to a local question named a well-known national chain or franchise brand, such as a national handyman, restoration, plumbing or senior-care franchise. At least 3,313 of 10,016 answers did. In 29 of 82 tests the most-named company was one of these brands; among the 67 independent businesses that was true in 21 tests.

"At least" matters here. The count uses a fixed list of brand names that are national without doubt. Regional chains and less famous brands are not on the list, so the real share is higher.

Directories were rare as a named provider: 113 of 10,016 answers named a directory or list site, and in 1 test a directory was the most-named "company". That number is low by design, because the test asks for businesses, not for places to look them up.

Finding 5: Advice questions almost never lead to the tested business

Buyers do not only ask "who is the best". They also ask what something costs, how long it takes, whether they need a permit, what can go wrong and how to choose. We sorted every question by its wording. 8,900 questions ask for a provider. 1,116 are advice questions that do not ask for one; they come from 35 of the 82 tests.

Type of questionQuestions testedAnswers naming the tested business
Asks who to hire8,900441
Price2464
Time420
Rules, licenses, law, insurance260
Risk450
How to choose1121
Other advice (can it be done, how it works)6454
All advice questions1,1169

All 82 tests. A question that asks for a provider and a price in one sentence counts as "asks who to hire".

The answers to advice questions still recommended companies: 2,646 company mentions in 1,116 answers. The tested business was among them in 9 answers. For time, rules and risk questions together it was 0 of 113. For independent businesses only, it was 4 of 1,046 advice answers, against 147 of 7,487 answers to provider questions.

This is the earliest moment a buyer meets a company: before they are ready to hire, while they are still learning. The companies that answer those questions on their own websites are the ones the AI assistant quotes.

Finding 6: Sectors differ, but no sector is safe

Sector (independent businesses only)BusinessesQuestions testedAnswers naming the businessBusinesses never named
Auto services3428110
Business services880047
Finance, accounting and insurance989994
Health, wellness and coaching71,120404
Home and property services233,2234410
Legal651034
Other6813374
Real estate and property management574033
All independent businesses678,53315136

In health, wellness and coaching, one dental practice accounts for 35 of the 40. "Other" holds a medical transportation network (17) and a youth mental health nonprofit (20).

Read the sector rows with care: some hold only a few businesses, and one strong business can move a whole row. What holds in every row is the size of the gap. In no sector were independent businesses named in more than a small fraction of the answers to their own buyers' questions.

How the study was done

  • The businesses. 82 businesses and organizations that asked for, or were offered, an AI visibility test by Mantas Auk LLC between September and November 2026: 67 independent businesses and 15 franchise locations or offices of larger brands, in 78 different lines of business. 72 of them serve only the Dallas-Fort Worth area; the other 10 serve a wider area or other states.
  • The questions. For each business we wrote a set of questions that its buyers ask before they hire: who does this work, what it costs, how long it takes, what the rules are, what can go wrong, how to choose. The questions use the services and the cities the business really serves. None of them contains the name of the business.
  • The answers. Each question was put to one AI assistant, Claude (made by Anthropic) with live web search, in a separate run. The run was not told which business was being tested. It made one web search for the question and then recommended up to three real businesses found in the results, best first, each with its website. These are answers from Claude with web search. They are not answers from ChatGPT, Gemini or Google.
  • The decision. Whether the tested business was named was decided by code, not by a person: the code compared each recommended company and its website with the name, the known spellings and the domain of the tested business. For a franchise location or a brand office, the brand name counted.
  • Untested questions. The question sets held 11,016 questions. 1,000 could not be tested, mostly because the search returned no businesses at all, only guides or price pages, or because the daily search limit was reached. They are left out. They are never counted as misses. That leaves 10,016 tested questions, from 46 to 447 per business.
  • One test per business. Where a business was tested more than once, we used its most recent test in which every answer came from Claude with web search.
  • Question types. Each question was sorted by fixed wording rules, applied by code. A question that asks for a company or a person is a provider question. Any other question is an advice question, and is then sorted by the first topic it matches: price, time, rules, risk, how to choose, other.
  • National brands. An answer counts as naming a national chain or franchise brand when a recommended company's name contains one of the brand names on a fixed list.

The data tables behind every number in this study are published with it. Businesses are described by trade and area only. No business is named.

Limits of this study

  • Mostly one metro area. 72 of 82 tests cover Dallas-Fort Worth only. We expect the pattern in other large US metro areas to be similar, but this study does not prove it.
  • Not a random sample. The businesses came to one agency, or were approached by it. They may differ from the average local business.
  • One AI assistant, one run. Every answer comes from Claude with web search, asked once. Other assistants read other sources and can name other companies. Answers also vary from one run to the next, so a business with 0 mentions could get 1 or 2 on another day. The size of the gap would not change.
  • Three companies at most. The test asks for up to three recommendations per answer. A live chat answer can be longer or shorter.
  • Answers without a company are left out. A question whose search returned no business was marked as not tested. So this study cannot say how often an AI assistant names no company at all.
  • Brand matching. For franchise locations, a mention of the brand counted even when the answer pointed to another location of the same brand. The franchise numbers are therefore the upper end.
  • Advice questions come from fewer tests. 35 of the 82 tests contain advice questions, and 706 of the 1,116 advice questions come from the 5 largest question sets. Finding 5 leans on those tests.
  • Simple sorting rules. Question types and national brands are decided by word lists. Some questions and some companies are sorted wrong. The lists are in the published data notes.

What a business owner can take from this

  • Do not test yourself by typing your own name. Ask the questions a stranger asks, without your name, and count how often you are in the answer.
  • Find out which companies the AI assistant names in your market. In most tests it was the same short list, and the tested business was not on it.
  • Look at what those companies publish: a page for each service and each city, clear prices, answers to the questions buyers ask before they call, and mentions on other websites. That is the material an AI assistant reads.
  • Measure again every month with the same questions. One run is a snapshot; the trend is what counts.

About the author

Mantas Auk is the founder of Mantas Auk LLC in Dallas, Texas. He has worked in search marketing for more than 15 years and helps local US businesses become visible in search engines and in AI assistants. He built the test method used in this study and a free public version of it, the AI Visibility Check on mantasauk.com.

Questions about the study or the data: mantas@mantasauk.com.

The data

Six CSV files hold everything behind this study. Businesses are described by trade and area only.

To cite: Mantas Auk, “Who Does AI Recommend? 10,016 Buyer Questions About 82 Local Businesses,” Mantas Auk LLC, November 2026, https://mantasauk.com/ai-visibility-study-2026/.

Where does your business stand?

Run the free check: 25 questions your customers ask, asked live, without your name. No email needed.