How AI Recommends Businesses: A Live Share-of-Voice Sample
When you ask an AI engine to recommend a business, how stable is its answer? We measure it directly: the same buying question, asked many times, across real verticals. This is our own data, refreshed weekly.
Why We Measure This Ourselves
Most writing about AI visibility describes how the systems work in the abstract. This piece does something different: it measures what they actually do. We ask AI answer engines real buying questions, the kind a customer would ask, and we record which businesses they name. The result is primary research, our own data, not a summary of anyone else's.
The reason to measure it directly is that the behavior is counterintuitive. People assume an AI engine has a settled view of "the best" businesses in a place, the way a directory has a fixed list. It does not. Asked the same question twice, an engine often answers differently. The shape of that variation, where it is stable and where it is not, is the thing worth knowing, and the only way to know it is to sample.
AI recommendation share of voice
The share of an AI answer engine's responses to a given buying question in which a particular business is named, measured by asking the same question many times and dividing appearances by the number of answers.
This is the measurable outcome of Trust Visibility: the clearer and more trusted a business is to an AI system, the larger its share of the answers tends to be.
How This Is Measured
The method is deliberately plain, so the numbers can be trusted and reproduced.
- For each vertical, we choose one realistic money-question, the sort a customer types when they are ready to act.
- We ask an AI engine that exact question several times, with web search enabled so it names real, current businesses rather than inventing them.
- We record the businesses named in each answer and count how many answers named each one. A business's share is its appearances divided by the number of answers.
- The engine used for this sample is Claude, queried through the Claude command-line tool on a standard subscription. The approach is engine-agnostic and the same questions can be run against other engines.
Three honesty notes apply to every figure below. First, AI answers vary between runs, so these are sampled shares, not a fixed ranking; a different week will produce somewhat different numbers, and that is expected. Second, competitor names are withheld and shown as anonymized labels, the counts and shares are real and unaltered, but the public page reports the pattern, not a leaderboard of named firms. Third, a business listed under slightly different names in different answers is counted separately unless the names are trivially identical, so the true concentration is, if anything, marginally higher than the tables show.
The Data
Data as of 2026-06-19. 32 total answers sampled across 4 local-service verticals, 8 answers per question. Competitor names are withheld; only the recommendation pattern is shown.
Estate planning attorneys in Thousand Oaks California
Money-question: "best estate planning attorney in Thousand Oaks California". 8 answers sampled, 13 distinct businesses named, 0 named in a majority of answers.
| Rank | Business | Appeared in | Share of answers |
|---|---|---|---|
| A | Business A | 3 of 8 | 38% |
| B | Business B | 3 of 8 | 38% |
| C | Business C | 3 of 8 | 38% |
| D | Business D | 3 of 8 | 38% |
| E | Business E | 1 of 8 | 13% |
| F | Business F | 1 of 8 | 13% |
| G | Business G | 1 of 8 | 13% |
| H | Business H | 1 of 8 | 13% |
Dentists in Pasadena California
Money-question: "best family dentist in Pasadena California". 8 answers sampled, 14 distinct businesses named, 0 named in a majority of answers.
| Rank | Business | Appeared in | Share of answers |
|---|---|---|---|
| A | Business A | 4 of 8 | 50% |
| B | Business B | 4 of 8 | 50% |
| C | Business C | 3 of 8 | 38% |
| D | Business D | 2 of 8 | 25% |
| E | Business E | 2 of 8 | 25% |
| F | Business F | 2 of 8 | 25% |
| G | Business G | 2 of 8 | 25% |
| H | Business H | 2 of 8 | 25% |
Financial advisors in Austin Texas
Money-question: "best fee-only financial advisor in Austin Texas". 8 answers sampled, 14 distinct businesses named, 0 named in a majority of answers.
| Rank | Business | Appeared in | Share of answers |
|---|---|---|---|
| A | Business A | 4 of 8 | 50% |
| B | Business B | 4 of 8 | 50% |
| C | Business C | 3 of 8 | 38% |
| D | Business D | 3 of 8 | 38% |
| E | Business E | 2 of 8 | 25% |
| F | Business F | 1 of 8 | 13% |
| G | Business G | 1 of 8 | 13% |
| H | Business H | 1 of 8 | 13% |
Personal injury lawyers in Phoenix Arizona
Money-question: "best personal injury lawyer in Phoenix Arizona". 8 answers sampled, 8 distinct businesses named, 0 named in a majority of answers.
| Rank | Business | Appeared in | Share of answers |
|---|---|---|---|
| A | Business A | 4 of 8 | 50% |
| B | Business B | 4 of 8 | 50% |
| C | Business C | 4 of 8 | 50% |
| D | Business D | 2 of 8 | 25% |
| E | Business E | 2 of 8 | 25% |
| F | Business F | 2 of 8 | 25% |
| G | Business G | 1 of 8 | 13% |
| H | Business H | 1 of 8 | 13% |
What the Data Shows
The clearest finding is how little the answer is settled. Across these samples, even the single most-recommended business in a vertical usually appears in only a minority to about half of the answers, and in many verticals no business reaches a majority at all. Asked the same question eight times, the engine frequently names a largely different set each time. The recommendation is reassembled, not retrieved.
The degree of consensus also varies, both between verticals and between weeks: sometimes a small group of names recurs often enough to form a consensus set, and sometimes the answers are scattered with no clear front runner. That instability is the signal worth reading. Where answers are scattered, visibility is contestable: no one owns the answer, and a clearly legible, well-trusted business can win a share the incumbents have not locked up. Where a consensus set has formed, the goal is to become one of the few names it contains. The table above shows which situation each vertical is in for the current sample.
The consensus set
The small group of businesses an AI answer engine names in a majority of its responses to the same buying question: the businesses it treats as the dependable answer rather than an occasional mention.
Joining the consensus set is the concrete objective of AI visibility work. It is earned through the signals the six pillars measure, not bought, and it is what separates being mentioned from being recommended.
What It Means for a Business
The variance between runs is not noise to be wished away; it is the opportunity. A fixed ranking would mean the order is settled and hard to move. A variable answer means the engine is still deciding each time, weighing what it can read about each business, and that the businesses it can read most clearly and trust most easily earn the larger share over many answers.
That is the bridge from this data back to the rest of the AIOInsights framework. Share of voice in AI answers is the visible result; trust visibility is the cause. A business that is clearly categorized, consistently identified, precisely described, and verifiably trusted gives the engine fewer reasons to leave it out, run after run, and a larger share is the cumulative effect.
An AI engine does not hold a fixed list of the best businesses in a place. It reassembles an answer each time it is asked, and the businesses it can read most clearly are the ones that keep reappearing.
A fixed ranking rewards whoever got there first. A variable answer rewards whoever is easiest to understand and trust, every time the question is asked. AI recommendation is the second kind, which is why clarity compounds and position cannot be hoarded.
This is a living data piece. The sample is re-collected every week by the AIOInsights share-of-voice run, and the figures above update with it. The method does not change between runs; only the data does.
It is the share of an AI answer engine's responses to a buying question in which a particular business is named. We measure it by asking the same question many times with web search enabled, then dividing each business's appearances by the number of answers. A 50 percent share means the business was named in half of the answers.
AI answers are not fixed. Asked the same question repeatedly, an engine names different businesses in different runs, drawing on different sources and phrasings each time. That variance is the finding, not a flaw, which is why we report a sampled share across many answers rather than a single ranking.
Yes. Every figure is computed from real answers collected by asking AI engines real buying questions, with the date, the number of answers, and the method stated alongside it. Competitor names are withheld and shown as anonymized labels, but the counts and shares are unaltered. Nothing on this page is estimated or invented.