subinternal Talk to us

ResearchGEO, SEO

GEO vs SEO: what’s actually different

Most of what gets an AI assistant to recommend you is the same work that gets you found in search. Of the 44 checks we run on every site, 29 serve both. Here’s what’s shared, what weighs more when an assistant answers, and the little that’s only for AI.

An engraving of a harbor at dawn: an old stone lighthouse and a newer iron signal station standing on the same long quay, with two sailing ships coming in.

Is GEO different from SEO?

Less than the name suggests. GEO, generative engine optimization, is the work of getting AI assistants to name and link to you. Most of it is SEO done properly: pages a crawler can reach, words it can read, and a clear account of what each page is. Of the 44 checks in mttr’s rulebook, 29 serve search engines and assistants alike, and they carry 65% of its weight.1

Google says as much for its own AI features: there are no additional requirements to appear in AI Overviews or AI Mode, and no new files or markup to add.2 What changes is how an answer is put together, so a few things weigh more. And a very few are only for AI.

Serve search and AI alike65%Matter more for AI answers21%Only for AI9%Let people act on the answer5%
Share of the weight in mttr’s rulebook of 44 rules, by our grouping.

The rulebook doesn’t mark which rule serves which, so that grouping is our judgment, rule by rule. The figures in it are counted from the rulebook itself.

What do GEO and SEO share?

The foundation, and it’s most of the work. A crawler has to be let in, find the page from a sitemap or a link, land on one address for it, and get the words in the HTML rather than after a script runs. Then it needs a title, one main heading, and structured data that matches what the page shows. None of that is new, and plenty of sites still miss it.

CheckSites with a page failing it
Words in the HTML, not only by script36%
Structured data on every page63%
Structured data that matches the page53%
One canonical address per page59%
A sitemap29%

Among the open sites each check applies to, of the 316,711 mttr read in October 2026.3 Fix these first. They count twice: once for search, once for every assistant that reads the web.

What matters more for AI answers?

Whatever helps an assistant lift a passage and trust it. A search engine ranks a page and lets the person read it. An assistant reads the page for them and repeats a sentence or two, so it needs the answer to sit right under the question, in plain words, with sources it can stand behind. The rulebook has 10 checks of that kind, with 21% of its weight.

Answer right under the question99%Figures in tables97%A heading phrased as the question90%Sources a reader can follow51%Lists or tables an answer can lift48%Profiles linked from the site46%
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 2026.

These are where sites fall furthest behind. 99% of sites with guides have one that doesn’t answer right under its first question, and 90% of sites with content pages never phrase a heading as a question. We wrote up how to fix that in Answer the question in its first sentence.

Research on GEO points the same way. A 2024 study found that rewriting content for generative engines could raise its visibility in their answers by up to 40%, and that what worked varied by subject.4

What is only for AI?

Very little: 3 checks, with 9% of the weight. Two are about letting AI crawlers in. 4.2% of open sites let search engines in but turn AI crawlers away by name in robots.txt, often from a line nobody remembers adding. The third is an llms.txt, a short file that outlines a site for language models.5 78% of sites don’t have one.

Google says its own AI features don’t need one.2 Other assistants may read it, so it’s worth the few minutes once the rest is right. But it carries the least weight of any check, and no llms.txt makes up for pages a crawler can’t get into. Which crawler does what is in Is your site keeping AI out?

What happens after the answer?

The person arrives, and they have to be able to do something. That isn’t search or AI; it’s whether the visit turns into a call or a booking. 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. An assistant tends to link to a specific page, not your home page, so every page needs a next step.

Where should you start?

  1. Fix the shared foundation. Access, one address, the words in the HTML, titles and structured data that matches. Every point here counts for both.
  2. Answer under the question. Headings phrased as questions, the answer in the first sentence, figures in tables.
  3. Show your sources and your date. And link to your profiles, so an assistant can tell you’re the business everyone else talks about.
  4. Check the AI crawlers by name. Then add an llms.txt.
  5. Give every page a next step.

The free AI visibility check shows where your site stands on each.

What can’t this tell you?

Whether an assistant will name you. A score says your pages are well formed for search engines and AI assistants. Whether you’re named also depends on what other sites say about you and on exactly what the person asked. And the grouping above is ours: someone else could file a rule or two differently. The point holds either way. Most of GEO is SEO, done properly.

References

  1. mttr, the rulebook: every rule, its stage, its weight and its fix. mttr.si/rulebook
  2. Google Search Central, “AI features and your website.” developers.google.com
  3. subinternal Research, “How 316,711 websites do with AI search,” October 2026. subinternal.com
  4. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, “GEO: Generative Engine Optimization,” KDD 2024. arxiv.org
  5. “The /llms.txt file,” a proposal for a file that outlines a site for language models. llmstxt.org