US AI Visibility Study 2026

Do AI assistants name your company?

People now ask AI assistants what to buy and who to trust. We took 2,047 US companies and asked four AI assistants 15 everyday questions about each company’s market. Every question was unbranded: none named a company. Then we counted which companies the AI brought up.

US companies

every one tested

AI answers read

15 questions per company

AI assistants

OpenAI, Claude, Perplexity, Gemini

never named

by any of the four

Each dot is one company. Hover one to see which; click to open it.Named oftenNever named

Short on time? Read the six takeaways

Never named

1 of 6

are never named by any AI assistant

Named at least onceNever named

What this means: For about 1 in 4 of these companies, a customer who asks AI for options will not hear their name.

All or nothing

2 of 6

are named by all four AI assistants, or by none

What this means: A company tends to show up on every AI assistant or on none, so there is no single one to focus on.

The scores

3 of 6

is the typical overall score, out of 100

What this means: No company is close to full marks, so every one of them has room to improve.

The groundwork

4 of 6

35%

of the least prepared websites are never used as a source by AI

What this means: Technical setup is the groundwork. The websites with the least of it are the ones AI leaves out most often, so it comes first. Content optimized for the questions people ask AI builds on it.

By sector

5 of 6

of answers name the typical travel company, the most named sector

What this means: How often a company is named depends a lot on its industry, so compare with your own sector first.

The ranking

6 of 6

companies with the highest overall scores

Pinterest leads with a score of 54. The score covers more than how often a company is named, so these are not simply the fifty most named.

See the fifty

The short version

Six things to know

1About 1 in 4 of the US companies we studied are never named by AI assistants (498 of 2,047).
2The four AI assistants mostly agree. 64% of companies are named by all four or by none.
3Technical setup comes first. Of the least prepared websites, 35% are never used as a source by AI, against 23% of the rest. Content optimized for the questions people ask AI builds on that groundwork.
4AI names companies far more often when people are browsing (31.8% of answers) than when they are ready to buy (6.2%).
5AI builds its answers mostly from other websites. A company's own website is 1.5% of the sources AI uses, and 6.0% in answers that name the company.
697 of America's 100 largest retailers publish no way for an AI agent to check out. Small shops on Shopify and BigCommerce nearly all do.

The ranking

The fifty highest overall scores

Companies are ranked by overall score, which covers more than how often they are named. Rows marked rarely named came up fewer than three times, so read their scores with care. Pick a sector to see its top ten.

Browse all 2,047 companiesor search by name, website or stock ticker from the header

What counts as being named

We never name the company. Some answers do.

We asked 15 questions about each company’s market, worded the way a buyer would ask them. None named the company. These are real answers from the study.

Asked of all four AI assistants

What are the best GPS watches for tracking fitness and outdoor activities?

OpenAI•••

Claude•••

Perplexity•••

Gemini•••

Gemini answered
“Garmin and Coros are frequently cited as top brands, offering a range of models suitable for various activities and budgets.…”
AI answer as returned, shortened and not checked for accuracy.

GarminNamed by 4 of 4 AI assistants on this question

What else we found

Seven more findings

1.5%

of the sources AI used are the company's own website

Only 1.5% of the sources AI used were the website of the company the questions were about. 523 companies were never used as a source at all. The share rises to 6.0% in answers that name the company, and is 0.8% in answers that do not.

What this means: AI mostly reads other websites, such as forums, reference websites and news. It uses a company's own website far more when it is already naming that company. Content optimized for the questions people ask AI makes a company more likely to be named, and its website more likely to be used.

Each square is a thousandth of the 1,170,936 sources. The 15 in blue are the companies' own websites.

22%

of websites have a guide written for AI (llms.txt)

Most websites have what search engines need: 85% have a rules file (robots.txt) and 73% have a sitemap. The newer files written for AI tools are rare. 445 companies publish an llms.txt, a short guide that tells AI what is on the website.

What this means: Most websites are set up for search engines. Very few have added the newer files written for AI tools.

3 of 14

technical checks passed by the typical company's website

We ran 14 technical checks on each website to see how ready it is for AI tools. The typical website passes 3. The best in the study passes 10.

What this means: No company is close to finished, so this is still open ground for everyone.

3 typicalbest 10 of 14

640

retailers, better prepared and less often left out

Retail is the largest sector in the study, and its websites pass more of the technical checks than the rest. 21.6% of retailers are never named, against 25.6% of everyone else.

What this means: Retailers have done more of the technical work than anyone, and fewer of them are missing from AI answers.

Typical AI readiness score

Retail29

Everyone else21

Never named

Retail21.6%

Everyone else25.6%

more likely to be named when people ask what is out there than when people ask how to fix a problem

When people ask what is out there, AI names a company in 31.8% of answers. When people ask how to fix a problem, it is 4.2%. Questions close to a purchase are near the bottom too.

What this means: AI names companies most often before people are ready to choose, and far less when they are about to decide.

16%

of websites have set rules for AI crawlers

A website can tell the automated readers AI companies send, called crawlers, what they may read. Only 284 of the 1,722 websites whose rules we could read name any AI crawler. The rest rely on rules written years ago for search engines.

What this means: Most companies have never decided what AI may read on their website. Old rules written for search engines are deciding for them.

284 of 1,722 readable robots.txt files name any AI crawler at all.

83%

of companies have Reddit among the sources AI used

Reddit is a source in answers about 1,694 of 2,047 companies, more than any other website. Wikipedia is a source for 1,569.

What this means: What people say about a company on public websites shapes AI answers more than the company's own website does.

Reddit is a source for 1,694 of 2,047 companies, Wikipedia for 1,569.

By type of question

AI names companies when nobody is choosing yet

We asked seven types of question about every company’s market, shown here in the order a buyer might ask them. AI names companies most when people are browsing. After that it drops sharply, except on questions about price.

31.8%

11.0%

9.5%

5.8%

12.7%

6.2%

4.2%

Browsing

5,186 questions

Specific need

4,093 questions

Alternatives

4,890 questions

Comparing

4,249 questions

Price

4,025 questions

Ready to buy

4,175 questions

Fixing a problem

4,087 questions

Exploring

Deciding

After buying

Browsing means asking what is out there, such as “best outdoor gear stores”. Fixing a problem means asking how to solve something that went wrong. Comparing and ready to buy are the two questions closest to a purchase, and both sit near the bottom.

By AI assistant

Gemini names companies most often. Perplexity lists the most sources.

How often each AI assistant names the company

Gemini

Flash with search

of 30,705 answers

Perplexity

Sonar

of 30,705 answers

OpenAI

GPT mini with web search

of 30,705 answers

Claude

Haiku with web search

of 30,705 answers

Who lists the most sources

Perplexity71.3%
Gemini12.1%
Claude11.4%
OpenAI5.1%

These shares come from a sample of the sources that we could match to an AI assistant. One assistant lists far more sources than the others, so the most used websites in this study mostly reflect where that one looks.

AI readiness

Most websites cover the basics. Few are ready for AI tools.

AI tools are starting to read websites and act for customers, for example to compare products or place an order. We ran 14 technical checks on each website to see how easy that is. The typical website passes 3. The best passes 10. The detail below is for your technical team.

Companies by checks passed

198

272

692

466

179

Checks passed, of 14

Each check, share of companies passing

Content Accessibility

Token budget (page weight)

95.2%

llms.txt published

21.7%

Markdown content negotiation

8.3%

Discoverability

robots.txt published

85.0%

XML sitemap

73.2%

HTTP Link headers (RFC 8288)

29.0%

Bot Access Control

Explicit AI bot rules

13.0%

API / Auth / MCP

OAuth Authorization Server discovery

10.1%

OAuth Protected Resource discovery

9.3%

WebMCP tools

4.2%

API Catalog (.well-known/api-catalog)

0.2%

MCP Server Card

0.0%

A2A Agent Card

0.0%

Agent Skills declared

0.0%

What AI is allowed to read

Most companies have not told AI what it may read

A website can tell the automated readers AI companies send, called crawlers, what they may read. We looked at the rules file (robots.txt) of 1,722 websites. Only 284 name any AI crawler. The rest rely on rules written years ago for search engines. Companies that do set rules mostly block the crawlers that collect pages to train AI, and allow the ones that fetch a page to answer a question. The table below is for your technical team.

Training AI crawlers

Collect pages to train AI

websites block each one, on average

GPTBotnamed 198 · blocked 70
ClaudeBotnamed 168 · blocked 54
Google-Extendednamed 154 · blocked 55
CCBotnamed 124 · blocked 76
Meta-ExternalAgentnamed 110 · blocked 61
Bytespidernamed 87 · blocked 70
anthropic-ainamed 84 · blocked 36
Applebot-Extendednamed 83 · blocked 47
FacebookBotnamed 67 · blocked 34
Omgilibotnamed 41 · blocked 45
Diffbotnamed 40 · blocked 39

AI search crawlers

Index pages for AI search

websites block each one, on average

PerplexityBotnamed 171 · blocked 29
OAI-SearchBotnamed 147 · blocked 17
Amazonbotnamed 88 · blocked 49
Claude-SearchBotnamed 80 · blocked 20
Applebotnamed 62 · blocked 14
YouBotnamed 61 · blocked 30

Answering questions crawlers

Fetch a page when a person asks

websites block each one, on average

ChatGPT-Usernamed 156 · blocked 18
Claude-Usernamed 69 · blocked 20
Perplexity-Usernamed 65 · blocked 25
DuckAssistBotnamed 36 · blocked 24
MistralAI-Usernamed 15 · blocked 24

AI agents

Can an AI agent buy from these companies?

Being named is the first step. The next is an AI agent sent to buy. We scanned all 2,047 of these companies for a published way to check out. Small shops on Shopify and BigCommerce nearly all have one. Almost nobody else does.

97 of 100

of the largest US retailers publish no way for an agent to check out

145 of 154

shops on Shopify and BigCommerce already do

99%

of companies stop at the web basics, level 0 or 1 of 5

See who an AI agent can buy from

The sources

Where AI gets its answers

The websites AI used as sources most often. The more of the 2,047 companies a website was a source for, the bigger it is shown.

Used as a source for the most companies

reddit.com

83%

en.wikipedia.org

77%

youtube.com

74%

justdial.com

61%

forbes.com

61%

facebook.com

57%

linkedin.com

56%

finance.yahoo.com

55%

61 companies

3.3%

of answers name the typical company

typical overall score

10overall score, line at the study’s typical 3260

Read the summary

AI assistants name the typical company in this sector in 3.3% of answers, compared with 6.7% for the typical company in the whole study. The assistants name Upwork, The RealReal, and Funko most often in this sector.

See all 61 →

By type of company

NRF Top 100 retailers are named most often

The companies came from six public lists. The typical company among the NRF Top 100 retailers is named in 23.3% of answers, against 6.7% across the study. AI readiness is low on every list, averaging between 24 and 32 out of 100.

Type of company	Companies	Answers naming the typical company	Never named	Average AI readiness, out of 100	Typical overall score
NRF Top 100 retailers	100	23.3%	5.0%	24	37
Large private companies	60	10.0%	15.0%	25	33.5
S&P 500 companies	500	8.3%	19.6%	24	32
Consumer brands	362	8.3%	22.4%	27	34
Other US retailers	442	5.0%	25.3%	32	33
Other US-listed companies	642	4.2%	30.1%	26	31
All companies	2,047	6.7%	24.3%	27	32

Each company is counted on the first list it appeared on. The NRF Top 100 is the exception: it counts every retailer on that list. All 100 retailers on the NRF Top 100 are in the study. 59 of them are also on another list (28 consumer brands, 26 S&P 500 companies and 5 other US-listed companies) and are counted on both, so the lists add up to more than 2,047. Popular websites we added by hand, such as Facebook, YouTube and Wikipedia, are counted with the consumer brands and the large private companies. See all 100 retailers on the NRF Top 100.

How to read this

What the study can and cannot tell you

Things to keep in mind

858 of 2,047 companies were named fewer than three times and are marked "rarely named". Read their scores as a sign that AI seldom brings them up, not as an exact comparison with their rivals.

Read how we measured this

2,047 companies

Where does your company stand?

Every company in the study has its own report: how often each AI assistant names it, who gets named instead, and how ready its website is for AI tools.