ResearchGEO, AI search

How to get recommended by ChatGPT, Gemini and Google’s AI

An assistant can only recommend what it can read, understand and trust, and only helps if people can act on it. Here’s each of those, how 316,711 sites do on it, and what to fix first.

What decides whether an AI assistant recommends you?

Four things, in order. The assistant has to be able to read your pages. It has to understand what you do, where, and for whom. It has to trust what it reads. And the person it sends to you has to be able to act: call, book, or buy. Fail an early one and the later ones never get a chance.

mttr checks every page of a site against one rulebook in those four stages: access, understand, trust and act.1 We’ve read 316,711 sites with it. The middle open site scores 78 of 100, which sounds fine until you see where the points go missing.2

Answer right under the question99%A heading phrased as the question90%A tap-to-call link70%Structured data63%Sources a reader can follow51%A way to get in touch45%Words in the HTML, not only by script36%
Share of sites with at least one page failing each check, among the sites it applies to. From mttr’s reads of 316,711 sites, October 30, 2026.

Can AI read your site?

For most sites, yes, but not all. 9.1% of the sites we read turn crawlers away entirely, so no assistant can read them at all. And 36% of the rest have at least one page whose words only appear once a script runs. Many AI crawlers don’t run scripts, so to them that page is close to empty.

The fix is the plainest one: let the crawlers in, and send the words in the page itself. We wrote up crawler access in detail in Is your site keeping AI out?

Does it understand what you do?

This is where most sites lose. 99% of sites with guides or articles have one that doesn’t answer right under its first question, and 90% of sites with content pages never phrase a heading as the question a person would ask. An assistant looks for the sentence under a question. When that sentence is a warm-up, it has nothing to lift.

63% have a page missing structured data, the block of facts that says, in a form a machine can’t misread, what the business is, where it is and what the page is about.3 Both are covered in Answer the question in its first sentence.

Does it trust what it reads?

Trust is quieter. 51% of sites with guides have one with no sources a reader could follow, 37% have one with no date it was last updated, and 46% never link to the profiles that would tie the business together across the web. An assistant weighs a page that shows its work, says when it was true, and is clearly the same business everyone else talks about.

Can people act on the answer?

An answer that sends someone to a page with no next step is wasted. 70% of sites that show a phone number have a page where it can’t be tapped to call, and 45% have a page with no way to get in touch at all. Those are often the exact pages an assistant links to.

Why does an assistant name someone else?

When we work through a business’s lost questions with them, every one comes down to one of four reasons. No page answers it. A page exists but doesn’t answer it plainly. Crawlers can’t read the page that does. Or the assistant answers from other sites: a roundup, a directory, a review site, a forum.

Why the question is lostWhat fixes it
No page answers itWrite the one page that does. Only where nothing on the site already answers.
A page exists but doesn’t answerMake that page answer in its first lines, rather than adding a second page on the same question.
Crawlers can’t read the pageFix access first. Nothing else counts until the page can be read.
It answers from other sitesClaim the listings you can take over, earn the reviews and mentions, and correct what’s wrong.

The first three are on your own site, so they’re yours to fix this week. The fourth takes longer, and it’s the one people forget.

What should you fix first?

In this order, since each one stops the ones after it from counting.

  1. Let the crawlers in. Check robots.txt and your hosting’s bot settings, and send the words in the HTML.
  2. Answer under the question. Phrase the heading as the question; make the first sentence under it the answer.
  3. Say what each page is. Structured data that matches the page: the business, its address and hours, the service, the article and its date.
  4. Show your work. Sources, a date, and links to your profiles.
  5. Make the next step tappable. A number that dials and a way to get in touch on every page someone might land on.

Then go after the questions you lose to other sites. You can see where your site stands on the first five with the free AI visibility check.

What can’t a score tell you?

Whether an assistant actually names you. A score says your pages are well formed for search engines and AI assistants: readable, clear, trustworthy and easy to act on. Whether you’re named also depends on what other sites say about you and on exactly what the person asked. A good score takes away the reasons not to recommend you. It doesn’t guarantee that anyone will.

References

  1. mttr, the rulebook: every rule, its stage, its weight and its fix. mttr.si/rulebook
  2. subinternal Research, “How 316,711 websites do with AI search,” October 2026. subinternal.com
  3. Google Search Central, “Introduction to structured data markup in Google Search.” developers.google.com