What 34 Million Scored Websites Reveal About the Web AI Engines Actually See

Ask someone how big the web is and you get a shrug: billions of pages, endless sites, too much to count. We wanted a sharper answer. Through our own crawling and analysis we have scored more than 34 million websites, enough to say something concrete about what the web is actually made of, and about the much smaller slice of it that AI engines can treat as a real source.

The web AI engines actually see, from 34M+ scored websites. Source: Engagemii, August 2026.

One in five domains is nobody home

Here is the composition. About 73 percent of domains are live with real content: an actual business, organization, or publication saying something. The rest is parked pages, placeholders, empty responses, and shells that render nothing without JavaScript. Put together, just over 1 in 5 of the web answers the door with nobody home.

That number matters for anyone thinking about AI search. When an AI engine assembles an answer about a business, a product, or a service, its pool of usable sources is far smaller than raw domain counts suggest. The billions-of-pages web and the web an AI engine can actually draw on are two different places, and the second one is where visibility is won or lost.

The answer layer is heavily American

Geography tells a similar story of concentration. In our hosting-location analysis, which has resolved more than 1.2 million domains so far, 60 percent of the domains that resolve are hosted in the United States. Nearly 300,000 more returned no DNS answer at all: the domain exists on paper, but the site behind it is gone.

Who AI engines actually reference

So who, inside that smaller real web, do AI engines actually reference? We analyzed 1,606 real citations from AI engine answers and matched them against our scoring data. Only 7 percent of the referenced companies were household-name giants. 72 percent were mid-sized businesses.

The pattern holds from the other direction too. 63 percent of the companies AI engines referenced score 6 or higher on our scale, against 8 percent of businesses overall, roughly an eight-to-one difference. We are not claiming a score causes a citation. We are reporting what the data shows: the companies AI engines reference tend to be the ones that score well, and scoring well has almost nothing to do with being big.

What separates the high scorers

What separates the high scorers is work, not fame. Their sites state plainly what the business is, who runs it, and what it offers. Their information is machine-readable and consistent wherever the business appears. Their pages contain real written answers to the questions customers actually ask. None of that requires a famous name or an enterprise budget, which is exactly why mid-sized businesses dominate the referenced set.

The takeaway from 34 million scored websites is not that the web is dying. It is that the real web, the part with something to say, is a smaller and more winnable arena than the raw numbers imply. A fifth of the field is empty. The giants are, on average, sloppier than their reputations. The businesses AI engines reference earned it on the page.

The honest limits

The composition figures come from a large random sample of our records, accurate to about a percentage point. The hosting analysis is ongoing, so the 60 percent figure describes the first 1.2 million domains resolved, not a final census. And the citation findings are correlational: they describe who gets referenced, not a mechanism. We publish the numbers as we measure them.

See where you stand

We keep scoring the web every day, including more than 15 million verified US businesses. Your free score shows how your own site reads to the machines, measured on the same scale as everyone else in your category. Check yours free at engagemii.com/aeo.

If you want to cite this article, the URL is engagemii.com/blog/what-34-million-scored-websites-reveal.