Deutsch
Français
Español
Português (Brasil)
ItalianoHow AI recommends brands
in the United States
Which brands ChatGPT, Gemini and Google AI Mode recommend across 35 US markets, and which sources they rely on.
- Markets
- 35
- Questions
- 18,400
- Answers
- 55,198
- Citations
- 608,256
01 · Categories
Pick a category
35 measured US markets in 9 categories. Open a category, or jump straight to a market. Leaders on ChatGPT.
- Home Services 7 markets
- Healthcare 6 markets
- Addiction Treatment ServicesLeader: Hazelden Betty Ford45%
- Dental & Orthodontic ServicesLeader: Aspen Dental19.5%
- Dermatology & Aesthetic Medical ServicesLeader: Mount Sinai19%
- Physical Therapy & RehabilitationLeader: Hospital for Special Surgery25.5%
- Psychiatric Care ServicesLeader: Talkiatry23.5%
- Psychological Therapy & Assessment ServicesLeader: Calm7%
- Finance & Insurance 3 markets
- Legal Services 4 markets
- Software 4 markets
- Fashion & Apparel 4 markets
- Beauty & Personal Care 5 markets
- Sports & Outdoors 1 market
- Pets 1 market
02 · Openings
Where it’s easiest
to break in
The lower the top-5 share, the weaker the leaders, and the easier it is
for a new brand to get named.
03 · Sources
Most-cited domains across all 35 markets
The brands change from market to market, the sources barely do.
04 · Concentration
How concentrated each engine’s sources are
On Google AI Mode, 53 sites are half of everything it cites.
| Top domains | ChatGPT | Gemini | Google AI Mode |
|---|---|---|---|
| 1 | 3.3% | 3.7% | 29.5% |
| 2 | 6.3% | 5.6% | 31.6% |
| 3 | 8.5% | 6.9% | 33.2% |
| 5 | 10.8% | 9.1% | 35.6% |
| 10 | 15.3% | 12.9% | 39.5% |
| 20 | 20.4% | 18.3% | 43.4% |
| 30 | 23.4% | 21.7% | 46% |
| 50 | 27.8% | 26.8% | 49.6% |
| 100 | 35.4% | 34.5% | 54.9% |
| 200 | 43.9% | 43.3% | 60.3% |
| 300 | 49.1% | 48.8% | 63.8% |
| 500 | 55.5% | 56.2% | 68.2% |
| 1000 | 63.8% | 67.1% | 74.8% |
| 2000 | 71.5% | 78.1% | 81.5% |
| 3000 | 75.7% | 84.2% | 85.6% |
| 5000 | 80.7% | 91.2% | 90.8% |
| 10000 | 86.8% | 98.9% | 99.4% |
| 10357 | 100% | ||
| 10907 | 100% | ||
| 52463 | 100% |
- ChatGPT324of 52,463 domains
- Gemini326of 10,907 domains
- Google AI Mode53of 10,357 domains
The steeper the curve, the shorter the list you have to be on. Gemini needs 326 domains to reach the same half, so one brand page is a far smaller bet there.
Source: Findrix research. Cumulative share of each engine’s citations by number of domains, United States, 35 markets, measured Oct 2026.
05 · Market context
AI search behaves
differently across countries.
AI engines cite different domains depending on the country.
United States
| # | Domain | Share of citations Share of all ChatGPT citations in this country that go to the domainShare | Citations Citations are counted per pass over the questions, so engines asked a different number of times can be compared. |
|---|---|---|---|
| 1 | ulta.com (opens in a new tab) | 3.3% | 15K |
| 2 | reddit.com (opens in a new tab) | 3.0% | 14K |
| 3 | walmart.com (opens in a new tab) | 2.1% | 9.9K |
| 4 | sephora.com (opens in a new tab) | 1.3% | 5.9K |
| 5 | nih.gov (opens in a new tab) | 1.1% | 5.2K |
| 6 | google.com (opens in a new tab) | 1.0% | 4.8K |
| 7 | nordstrom.com (opens in a new tab) | 1.0% | 4.6K |
| 8 | target.com (opens in a new tab) | 1.0% | 4.4K |
| 9 | allure.com (opens in a new tab) | 0.8% | 3.7K |
| 10 | wsj.com (opens in a new tab) | 0.7% | 3.1K |
Australia
| # | Domain | Share of citations Share of all ChatGPT citations in this country that go to the domainShare | Citations Citations are counted per pass over the questions, so engines asked a different number of times can be compared. |
|---|---|---|---|
| 1 | google.com (opens in a new tab) | 3.7% | 18K |
| 2 | reddit.com (opens in a new tab) | 3.6% | 17K |
| 3 | ndis.gov.au (opens in a new tab) | 2.5% | 12K |
| 4 | nsw.gov.au (opens in a new tab) | 2.1% | 9.9K |
| 5 | adorebeauty.com.au (opens in a new tab) | 1.4% | 6.5K |
| 6 | chemistwarehouse.com.au (opens in a new tab) | 1.1% | 5.4K |
| 7 | chatgpt.com (opens in a new tab) | 0.8% | 3.8K |
| 8 | ndiscommission.gov.au (opens in a new tab) | 0.8% | 3.8K |
| 9 | mecca.com (opens in a new tab) | 0.8% | 3.6K |
| 10 | news.com.au (opens in a new tab) | 0.8% | 3.6K |
Other countries
- Australia39 markets measuredOpen Australia
- United KingdomNext country in the measurement planUpcoming
Frequently asked questions
What does this United States AI Search report measure?
Which brands ChatGPT, Gemini and Google AI Mode name when US buyers ask for recommendations, and which websites they cite in those answers. It covers 35 US markets and 18,400 buyer questions — about 526 per market, each asked once to each engine. Measured Oct 2026.
Which markets are included in the United States dataset?
35 measured markets in 9 categories: Home Services, Healthcare, Finance & Insurance, Legal Services, Software, Fashion & Apparel, Beauty & Personal Care, Sports & Outdoors and Pets.
What does “mention rate” mean?
The share of the engine’s answers to a market’s buyer questions that name the brand. Every rate has a 95% range: when the ranges of two brands overlap, their ranks are not really different.
What counts as a citation?
A link or source the engine shows with its answer. We count each cited domain once per answer, so a site linked three times in one answer counts once.
Why can AI mention a brand without citing its website?
Engines usually build an answer from third-party pages — reviews, guides, forums, official bodies — and name the brands those pages recommend. The brand’s own site is often not among the sources at all.
Why do AI engines use different sources?
Each engine has its own search index, retrieval and ranking. In the United States, Gemini spreads its citations widest — half of them go to 326 domains — while Google AI Mode puts half of its citations on just 53.
Why does AI source behavior differ between countries?
Engines prefer local sources: national regulators, local media and local retailers. The same category draws on different domains in the United States and in Australia, so visibility has to be measured country by country.
Where do the questions in this dataset come from?
They are written from how buyers search in each market — comparisons, “best for” questions, problems and price questions — and asked to every engine the same way.
How often is this data updated?
Each page shows the month it was measured; the numbers here are from Oct 2026. When a market is measured again, its page shows the new month and numbers.