When we talk about AI systems “reading the web,” it is easy to imagine ChatGPT, Claude, Perplexity and other systems exploring roughly the same information.

New crawler data suggests something very different.

Engagemii analyzed 10,190,915 business pages on which at least one verified AI crawler appeared. ClaudeBot reached almost 9.8 million of them. Applebot reached around 1.17 million and GPTBot about 590,000. Further down, Meta AI reached roughly 32,500 pages, while OAI-SearchBot and PerplexityBot appeared on only a few thousand.

The difference becomes even more interesting at page level. According to the study, 85.7% of pages were reached by exactly one AI crawler, while only 0.7% were visited by three or more.

That matters because “AI visibility” is often discussed as though it were one environment. It probably isn't.

A page discovered by ClaudeBot may never be reached by another crawler. A source available to one retrieval system may therefore be effectively absent from another. And crawler access itself still doesn't guarantee retrieval, citation or attribution.

This fits a pattern we've already observed at NetContentSEO. When we asked six AI models the same question about Net Content SEO, three reconstructed the entity reasonably well while three did not. Different systems can build very different representations of the same thing even when the underlying web is theoretically available to all of them.

It also reinforces another distinction that is becoming increasingly important: crawled, retrieved, cited and understood are not the same state.

Engagemii is careful about this limitation too. Its experiment measures crawler behavior on its own network of pages, not the entire web, and a crawler visit should not be interpreted as evidence that the page was later used in an AI answer.

But the scale of the difference is difficult to ignore.

Perhaps there isn't one “web seen by AI.”

There may be several overlapping versions of it.

Source: Engagemii — Which AI Crawlers Are Actually Reading the Web

Related NetContentSEO research: We Asked Six AI Models What Net Content SEO Is. Three Understood It. Three Didn't.