Most sites are bad at this. Only 29% should care.

We checked 57 companies for one thing: whether a machine sent to answer a question about them could read their homepage. Almost all scored badly. That on its own is not a finding, it is the weather. The finding is that only 29% — about a quarter — had the pattern that actually costs money.

What we were actually testing

A low answer-engine score is easy to sell and mostly meaningless. Across these 56 sites the median was 24 and the best anyone managed was 69, so telling a company "you score 24%" describes the entire market, not them.

The thing worth measuring is the mismatch: a company that demonstrably paid for search visibility and got none of that value back from machines. Search readiness 100, answer-engine readiness under 25. That is a specific, checkable loss, and it is what 16 of 56 sites here have.

What would have proved us wrong

Three results would have killed this. We report all three rather than only the flattering one, because a study that could not have failed is not evidence.

By company type

Weaker than the headline, and worth saying so. The differences are real but modest, and the smaller groups are too thin to conclude much from. Treat this as a mild preference, not a targeting rule.

The direct-to-consumer result is the interesting one, and it is not that their gaps are small. It is that they have the lowest median search readiness of any group, so fewer of them clear the bar of having invested in the first place. There is no loss to point at.

Every site we checked

Highlighted rows are the 16 that qualified. Scores are out of 100, measured on the homepage only, on 2026-09-07.

Method

One page per site, the homepage, loaded in a real browser on 2026-09-07. Search readiness is meta completeness, heading structure and structured data. Answer-engine readiness is whether the page states answers a machine can lift: question headings, marked-up Q&A, self-contained claims, and whether the page needs JavaScript to say anything at all.

The same scanner and the same scoring produce every row here, so any of them can be reproduced from the description above.

What this does not show

• Whether any AI assistant actually mentions these companies today. That needs repeated sampling of the models themselves. This is the cheaper question underneath it: if one comes looking, is there anything usable?
• Anything about site quality. Performance, accessibility, security and content were not part of this measurement.
• Anything beyond the homepage. A deeper page could score differently.
• A stable market picture. This is one run, on 2026-09-07, and sites change. Anything here can be re-checked and may now differ.

Check your own

The same measurement, on any address, in about two seconds.

Run the free check