Website visibility in ChatGPT and AI search: what a study of 68.9 million AI crawler visits showed
In short, BotRank’s research answers a simple question: why are some sites crawled by AI crawlers frequently, while others are barely reached.
Authors BotRank research on 68.9 million AI crawler visits looked at 858,457 sites on the Duda platform. This is not a ranking study or a list of secret factors for ChatGPT. This is a large cross-section of where AI bots generally reach, how often they return, and what signals are more common on sites with high crawl activity.
For the SEO team, this is more useful than it seems. If the bot does not reach the site, does not see the structure well, or cannot quickly understand what kind of company is in front of it, it is also difficult to expect good visibility in AI responses.
What the study showed
The main figure of the study is this: in February 2026, 59% of the sites in the sample received at least one visit from an AI crawler, and the total volume was 68.9 million visits. At the same time, 81% of activity came from OpenAI bots, 16.6% from Anthropic, and the share of Perplexity and Google was noticeably lower.
But what is more important is not the number itself, but what it coincides with. The authors saw several strong correlations:
- Sites with 50 or more blog posts received an average of 1,373.7 visits from AI crawlers, while sites without a blog received 41.6;
- sites with Google Business Profile synchronization had a 92.8% crawl rate versus 58.9% without such a signal;
- pages with local schema showed 72.3% crawl rate versus 55.2% without it;
- Review integrations also markedly coincided with more frequent visits.
It is important not to draw the wrong conclusion here. The study does not prove that it is enough to add schema and there will immediately be more bots. It shows something else: it is easier for AI systems to work with sites that already have a clear structure, formalized business data and a normal layer of content.
What does this mean in practice?
For AI search, simply “being in the index” is no longer enough. The site should be a convenient source: with a clear structure, normal service pages, readable contacts, understandable geography and data that can be quickly verified.
This is especially noticeable in local SEO, B2B and service niches. When an AI system tries to figure out who to show in a response, it checks more than just the text on the page. She needs to quickly gather a picture: who you are, where you work, what you do, do you have a confirmed business profile, do your contacts match, are there any reviews and external references.
There is one more important point. B OpenAI help about ChatGPT search it explicitly states that the system can rewrite the user's original query into more precise search queries. This means that it is not enough for a page to simply repeat one key. It should properly answer related questions: who you are, what you offer, what city or segment you work for, how you differ, and what facts confirm this.
Which signals are really worth amplifying?
1. Website accessibility for the necessary bots
If the site is not accessible to the necessary crawlers, further discussion of visibility in AI search is almost pointless. B OpenAI Help for ChatGPT search It says that to enable a site, it is important to allow OAI-SearchBot to bypass and not block traffic from published IP addresses. B OpenAI crawlers documentation it is specifically explained that GPTBot and OAI-SearchBot - these are different bots and different scenarios. One is responsible for training, the other is for searching and showing sources.
Therefore, the first practical step here is very mundane: check robots.txt, CDN/WAF, server logs and make sure that you don’t accidentally kill the necessary bots.
2. Correct understanding of Google-Extended
Around Google-Extended There is still a lot of confusion. B Google Search Central It is explicitly stated that this is a robots token to control the use of content in Gemini Apps and Vertex AI grounding. It does not affect the site's presence in classic Google Search and is not a ranking signal.
This is useful to remember so as not to waste time on a false task. Open or close Google-Extended - this is a decision about the use of content in Google AI products, and not about the growth of conventional SEO positions.
3. Formalized business signals
The BotRank study highlights what SEO teams already see on hand projects: it’s easier for AI systems to trust sites that don’t have a mess in their business data.
Google recommends filling out business details first: confirm the site in Search Console, declare the official site, monitor the knowledge panel and fill out the Business Profile. For local and service companies it is also important LocalBusiness markup, which helps to more clearly show the address, phone number, opening hours and other data about the point.
If you have several branches or several services, this is where problems usually begin: different phone numbers on the site, outdated addresses, empty cards, incomplete service pages. This is also bad for regular SEO, but in AI search such contradictions especially quickly destroy the credibility of the source.
4. Content that answers real questions
It was not the “50+ articles” number itself that gave rise to the increase in bot visits. Rather, it is a signal that the site has coverage of topics and intents. In other words, the AI bot has somewhere to go when it clarifies a question.
It is not abstract texts “about everything” that work well, but pages that cover understandable scenarios:
- what kind of service or product it is;
- who is it suitable for?
- how much does it cost or what does the price depend on;
- what are the differences between the options?
- how the service works in a specific city, niche or segment;
- what are the cases, limitations and supporting facts.
If a company needs to restore order in precisely this circuit, it is already closer to SEO promotion and technical auditthan to a one-time edit robots.txt.
5. Measuring not only bypass, but also real visibility
Frequent bot visits do not mean that the brand will be quoted in ChatGPT, Gemini or Perplexity. The bot could come, read the page and not use it in the response. Therefore, after basic technical verification, the team needs a second layer of analytics: which pages are actually mentioned, what issues the brand pops up on, and where it is consistently outperformed by competitors.
What should an SEO team do without illusions?
If you put everything into one working plan, you get a rather mundane sequence:
- Check if access is open to
OAI-SearchBotand other necessary crawlers. - Make sure that the CDN, firewall and hosting do not cut off AI bots over the network.
- Organize your Business Profile, contacts, addresses and branch pages.
- Check
Organization,LocalBusiness,Breadcrumband other basic markup where appropriate. - Reassemble key service and information pages for real user questions, and not for one key.
- Compare not only crawl logs, but also actual brand citations in AI responses.
This theme fits well with BI and SEO analytics, because without a separate measurement layer, the team very quickly starts looking in the wrong places. Instead of “they really show us,” all that remains is “the bot came in, that means everything is fine,” and this is not the same thing.
Conclusion
The study about 68.9 million visits by AI crawlers is useful not because of the high-profile number. It is useful because it brings the conversation back to understandable things. Sites are more often bypassed not for a magic tag or a buzzword in llms.txt. They are more often bypassed when the site has open access, a clean structure, clear business data and content that can be parsed without guesswork.
This is good news for the SEO team. To become more visible in AI search, you don’t need to invent a separate religion. First you need to remove the chaos on the website, in the company data and in the content structure. And only then see how this affects the actual appearance of the brand in AI responses.