Research · October 2026

ChatGPT skips 2 in 3 businesses: what 731 website checks show

Name: OperStack AI visibility checks, 15 September to 3 October 2026
Creator: Maksim Shchegolev

By Maksim Shchegolev, founder of OperStack · Published 3 October 2026 · Data from 15 September to 3 October 2026

66%of businesses were not named by ChatGPT in any of 3 buyer questions.

11other companies were named instead, at the median, when a business was left out.

28%of sites have good SEO but weak AI readiness: ranking in Google does not guarantee a place in AI answers.

Key figures

731website checks731 checks of 720 websites: 172 in English and 559 in Russian. 11 sites were checked in both languages.
56/100median AI readiness scoreAcross the 696 sites that got a score. 263 of them (38%) scored below 50.
28%good SEO, weak for AI194 of 696 sites score 80 or more for classic SEO but below 60 for AI readiness.
66%never named by ChatGPT133 of 202 businesses were not named in any of 3 buyer questions.
11rivals named insteadMedian number of other companies ChatGPT named when it left the business out.
38%of top-graded sites named8 of 21 sites with grade A were named at least once. A small sample.

How the sites scored

72 A

173 B

290 C

186 D

8 E

Number of sites by grade, from A (best) to E. Most land in the middle: C is the largest group.

How often ChatGPT named the business

Named in 0 of 3 answers 133
Named in 1 of 3 33
Named in 2 of 3 21
Named in 3 of 3 15

202 businesses, 3 buyer questions each (606 questions). 133 were never named (66%), 15 were named every time (7%).

What actually made a difference

We compared the 202 sites that got ChatGPT answers by a few things we could measure on each site: the AI readiness score, the letter grade, whether AI bots are allowed in robots.txt and whether the site has an llms.txt file. Four findings stand out, and two of them go against popular advice.

1. AI readiness helps, but less than people think

The share of sites that ChatGPT named at least once rises with the AI readiness score, from 29% to 43%. Yet even the best-prepared sites are named less than half the time. Technical work on the site is not enough on its own: what other sites write about the business matters at least as much.

AI readiness below 50 (78 sites) 29%
AI readiness 50 to 69 (66 sites) 32%
AI readiness 70 and above (58 sites) 43%

Share of sites ChatGPT named at least once in 3 buyer questions, by AI readiness score.

2. The top grade does not guarantee a mention

ChatGPT named 38% of the sites graded A (8 of 21), 46% of those graded B (27 of 59), 25% graded C (17 of 67) and 28% graded D (15 of 53). Grade B did better than grade A. The A group is small, 21 sites, so treat this as a signal rather than a rule.

3. Sites that block AI bots were named about as often as open ones

Where AI crawlers are blocked, ChatGPT still named 32% of the sites (22 sites). Where they are open, it named 34% (180 sites). The likely reason: ChatGPT answers mostly from what other sites say about a company, not only from the company's own pages. Opening robots.txt to AI bots is still worth doing, but it is not enough. The blocked group is small, 22 sites.

4. llms.txt makes no noticeable difference yet

Sites with an llms.txt file were named in 32% of cases (103 sites), sites without it in 36% (99 sites). llms.txt keeps a site tidy for machines, but by our data it does not affect whether ChatGPT names a business today. This runs against a lot of current advice, so we report it as we found it.

English vs Russian sites

ChatGPT did not name 71% of the English-language businesses (80 of 113) and 60% of the Russian-language ones (53 of 89). English-language niches likely have more competitors, so ChatGPT picks from a longer list; this is our assumption, not something the data proves.

AI bots blocked in robots.txt: 20% of English-language sites (35 of 172) and 12% of Russian-language sites (68 of 559).
llms.txt present: 49% of English-language sites (84 of 172) and 54% of Russian-language sites (304 of 559).

Why this happens

AI bots are blocked. Some sites close their pages to AI crawlers in robots.txt or at the firewall, often by accident, so the assistant has nothing fresh to read.
The text is not visible without scripts. When the page content only appears after JavaScript runs, many AI crawlers see an almost empty page.
Thin pages. A home page with a slogan and a contact form does not answer the questions a buyer asks: what you do, for whom, where and at what price.
No structured data. Without markup for the organisation, its services and its questions and answers, a machine has to guess what the page is about.
Few mentions elsewhere. Assistants lean on what other sites write about a company. A business that is rarely mentioned in directories, reviews and articles is easy to leave out.

The 28% figure shows how these causes hide: a site can rank well in Google and still be hard for an assistant to read and quote.

What to do: a 30-minute check in 5 steps

Open your robots.txt and make sure GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot are not blocked.
Look at your main page without JavaScript (view the page source) and check that the text a buyer needs is there.
Answer the buyer's questions on the page itself: what you do, for whom, where you work and how to start.
Add structured data for your organisation, services and frequently asked questions.
Ask ChatGPT the questions your buyers ask, note who it names instead of you, and work on being mentioned on the sites it relies on.

Run the free check on your site →

To see what a full report looks like, open the sample report.

Method and limitations

Who is in the sample. Websites whose owners or marketers came to OperStack and ran the free check. This is a self-selected sample, not a random one: people who worry about AI visibility are more likely to be in it.
Period. Checks run from 15 September to 3 October 2026. Our own sites are excluded, and repeated checks of one site count once; a site checked in two languages counts once per language, so 731 checks cover 720 websites.
Scores. 696 of the 731 sites received an AI readiness score out of 100. The same sites have a classic SEO score, with a median of 83. In English the AI median is 55 (65 of 157 sites below 50); in Russian it is 56 (198 of 539 below 50); the SEO median is 83 in both.
ChatGPT questions. For 202 sites (113 in English, 89 in Russian) ChatGPT was asked 3 questions a buyer of that business would ask, 606 questions in all. We recorded whether the business was named and which other companies were named.
Answers change. ChatGPT can answer the same question differently from one run to the next, so a single check is a snapshot, not a ranking.
Two languages together. English and Russian sites are combined in the headline figures; the split by language is given above.
No site names. We do not publish which sites were checked.

Questions

Does good SEO mean ChatGPT will recommend me?

No. 194 of the 696 scored sites (28%) have an SEO score of 80 or more but AI readiness below 60, and even among sites with AI readiness of 70 and above, ChatGPT named only 43% at least once.

Does llms.txt help?

Not noticeably, by our data. Sites with llms.txt were named in 32% of cases and sites without it in 36%. The file keeps a site tidy for machines, but it does not decide whether ChatGPT names you today.

How was the data collected?

From the free OperStack checks run between 15 September and 3 October 2026 by site owners and marketers who came to check their own sites. For 202 sites ChatGPT was asked 3 buyer questions each. Our own sites are excluded, and the sample is not random.

Can I check my own site?

Yes. The free OperStack check reads your site, gives an AI readiness score and asks ChatGPT buyer questions about your business, without signup: oper-stack.com/ai-visibility/.

How to cite

Source: OperStack, AI visibility study of 731 websites, October 2026, https://oper-stack.com/research/ai-visibility-2026/