Original research

AI visibility, measured twice: 12,792 AI answers in September, 4,291 in August

We measure which businesses AI names and which pages it cites, using data we collect ourselves. In September 2026 we collected 12,792 AI answers from 3 AI models and compared 3,127 business websites across 8 service industries and 12 US metros. In August 2026 we analysed 4,291 answers across ChatGPT, Claude, Gemini and Perplexity and tested 60 businesses across 41 niches. None of the numbers below come from industry reports.

The two studies used different engines, questions and samples, so every figure names its study and the totals are never added together. Last updated September 2026.

September 2026 · Newest study

What 12,792 AI answers and 3,127 business websites reveal about local businesses

12,792 AI answers
3,127 business websites
8 service industries
12 US metros

We asked OpenAI GPT-5.6, Anthropic Claude Sonnet 5 and Google Gemini 3.6 Flash about 8 local services in 12 US metros, with web search off and on, then compared the websites of the businesses AI named against the ones it never named. We collected every answer ourselves, through an API, between 5 and 10 September 2026.

2 of the 6 findings

59.6%

AI named the business with more content in 6 of 10 matchups with the same service, same city and similar reviews.

To compare like with like, we paired businesses head to head: same service, same city, similar number of Google reviews. In 59.6% of those pairs, the business AI named had more website pages.

That is not a coin flip. The difference is statistically significant.

348 pairs, 4 ties excluded. Clearest when AI answered from existing knowledge; the web search comparison was inconclusive.

7 of 96

More reviews didn't win. The review leader was AI's top pick in just 7 of 96 markets.

Reviews still matter. But having the most reviews rarely made a business the one AI named most often.

AI's existing knowledge, with web search turned off.

The website findings are observational. They show what the businesses AI named had in common, not that adding pages will make AI name a business.

The other 4 answer these

Which page does AI link to when a buyer asks what something costs?
Where does AI actually get its answers?
Does leading Google Maps decide who AI links to?
How much more content do the businesses AI names have?

The full study also includes:

3 examples of what AI answered and which local-business pages it cited, each with a to-do for your website
Every comparison behind the 6 findings, with its sample and scope
What the findings mean for your website, and where to start
The full method: models, dates, metros, industries and quality checks

See all 6 findings

August 2026

What 4,291 AI answers say about which businesses get named

Our first study. We collected 4,291 answers to real buyer questions from ChatGPT, Claude, Gemini and Perplexity between 28 April and 10 August 2026, roughly 1,070 per engine. Separately, we ran 60 real businesses across 41 niches against the buyer questions in their own category between 3 June and 14 August 2026.

21-33%

of answers named any specific business, depending on the engine

30 of 60

businesses tested never appeared once in their own category

1.8 vs 9.9

sources cited per answer, ChatGPT against Perplexity

Read the August 2026 study, engine by engine

Two studies, one pattern

Where both studies agree

Different engines, questions and samples, so these are two separate measurements pointing the same way, not one trend over time.

Going unnamed is normal

August 2026

30 of the 60 businesses we tested never appeared once in answers to their own category's buyer questions.

September 2026

2,369 of the 3,127 businesses whose websites we compared were never named in the experiment. That means absent from this experiment, not invisible to every AI system.

The question decides, not the company

August 2026

Our reading of the pattern: AI builds answers from sources it trusts on the specific question, so the unit of competition is the question.

September 2026

We measured it on 288 answers to cost and comparison questions with web search on, and the type of page AI linked to changed with the question. See which pages.

Leading somewhere else does not carry over

August 2026

Our reading of the pattern: a ranked list and a generated answer are separate results with separate winners.

September 2026

The business with the most Google reviews was AI's most-named business in only 7 of 96 markets. We also checked the local businesses AI linked to against the Google Maps results for their service and city. See how many were missing from Maps.

The mechanism

What makes AI name one business over another?

Neither study found a single switch. What both point to is the question. A business that has published a clear answer to a buyer's question gives AI something of its own to cite. Being cited and being named are different outcomes, and we measure them separately.

Answer Content Engine is built around that pattern: 7 agents find the questions your buyers ask, write the answers with your proof, publish them and track where you show up. It was not tested in either study.

01 · Research

Question Agent

Finds the real questions your buyers ask before they buy.

02 · Research

Research Agent

Ranks questions by buying intent and finds the angle competitors missed.

03 · Create

Content Agent

Writes the answer, built to get cited by AI.

04 · Create

Media Agent

Adds the images and video that ride with every answer.

05 · Create

Distribution Agent

Publishes to every platform the moment you approve.

06 · Analyze

Optimization Agent

Improves every piece in a self-learning loop that compounds.

07 · Analyze

Visibility Agent

Tracks where you show up versus competitors, and finds the gaps to own.

Client results

What this looks like for real businesses

Three Answer Content Engine clients in three different markets. Every number comes from their live dashboards, April to June 2026, which track AI mentions, Google top-10 placements and share of category mentions every week. These are client results, not results from either study.

64%

Real estate

Of category mentions, 3.5x the nearest competitor

Storage facility

Share of voice, plus 32 first-page Google placements

71%

Sports consulting

Of category mentions, nearly 3x the nearest rival

All case studies

Questions and answers

Common questions about AI visibility

How often do AI assistants name a specific business?

Between 21% and 33% of the time, depending on the engine, in our August 2026 study. Across 4,291 answers collected between 28 April and 10 August 2026, Perplexity and Gemini named a specific business in 33% of answers, Claude in 31%, and ChatGPT in 21%. Most AI answers about a buying decision in that study named no business at all.

Which AI engine recommends businesses most often?

In our August 2026 study, Perplexity and Gemini, tied at 33% of answers naming a specific business. ChatGPT was the outlier at 21%, roughly a third less often than either. If you judge your AI visibility by what ChatGPT says about you, you are reading the least generous of the four engines we measured.

How many sources does an AI answer cite?

It varies a lot by engine. In our August 2026 study, Perplexity cited an average of 9.9 sources per answer while ChatGPT cited 1.8, a spread of more than five to one. Being one of several acceptable sources can be enough to get surfaced in Perplexity, while ChatGPT behaves much closer to winner-take-all.

How many businesses are invisible to AI search?

Half in one study and most in the other. In August 2026, 30 of 60 businesses tested across 41 niches never appeared once in answers to their own category's buyer questions. In September 2026, 2,369 of the 3,127 businesses whose websites we compared were never named in that experiment, which is not the same as invisible to every AI system. The two studies used different samples, so the figures are not comparable or additive.

Does ranking in Google mean you appear in AI answers?

Not reliably. In our September 2026 study, AI linked to the websites of 129 local businesses when buyers asked cost and comparison questions, and a large share of them were not in the first 60 Google Maps results we collected for that service and city. That does not make Maps visibility irrelevant. Maps results depend on the query, location and collection time.

The full finding is in the September 2026 study

What makes an AI assistant recommend one business over another?

No single factor, but website content stood out in our September 2026 study. In 348 matched pairs with the same service, same city and review counts within 25% of each other, the business AI named had more pages 59.6% of the time (4 ties excluded), which is statistically significant. The result was clearest when AI answered from existing knowledge; the web search comparison was inconclusive. That is an association, not proof that adding pages causes AI to name a business.

The full finding is in the September 2026 study

Can brand share in AI answers be measured?

Yes, by asking the same buyer questions repeatedly and counting which businesses get named. One answer is one draw, so repetition is the point. In our September 2026 study, the main experiment asked 248 distinct questions repeated across 4 days and collected 12,504 answers. In our August 2026 study, the average business owned 9.4% of the answers in its category and the strongest reached 40%.

Do more Google reviews make AI recommend a business?

Not on their own. In our September 2026 study we compared the business with the most Google reviews against the business AI named most often in 96 local markets, with web search turned off. They matched in 7. Reviews still matter, particularly for whether a business is named at all. It is not a reason to stop collecting them.

The full finding is in the September 2026 study

Methodology

How we run this research

We collect the data ourselves. Every AI answer is one we asked for, inside a stated date window. Nothing is taken from a vendor dashboard or an industry report.

Every number carries its sample. A figure is reported with the answers, businesses or pages it was measured on, and the qualification sits beside it, not in a footnote.

Studies stay separate. August and September used different engines, questions and samples. Their figures are never added or read as a trend.

We leave out what we cannot trust.In August we withheld the per-engine rate at which AI cited a business's own site, because the stored figure did not survive a recount.

No sampled business is named. The businesses in both studies stay anonymous.

Limits. Neither study uses a random sample of businesses, and AI behaviour changes over time. These figures describe dated windows, not constants.

Full methods: September 2026 study and August 2026 study.

If this is your category

Find out where you show up

The free scan runs your business against AI engines and shows you which buyer questions you appear in and which ones a competitor owns.

Run the free scan Book a strategy call