GEO vs SEO: what changes when buyers ask AI assistants

SEO fights for a position in a list of ten links. GEO fights to be one of the two to four names an assistant speaks inside a written answer. There is no position four in an AI answer — a business is named or it does not exist in that conversation.

The mechanic that changed

A search engine returns links and the buyer does the filtering. A generative assistant returns a paragraph, having already done the filtering. It summarises what it can read about the category and names the handful of businesses the underlying evidence supports. The buyer often never sees a link.

What still matters

• Crawlability. If the assistant cannot read the site, it works from stale third-party data.
• Clear facts. Service, area, price band, hours, credentials — stated plainly and identically everywhere.
• Corroboration. The same facts on independent pages. One website is a claim; five agreeing pages are a fact.
• Consistency. Contradictions (old address, wrong hours, five different service lists) get a business dropped from the answer.

What matters less

Backlink volume for its own sake, keyword density, and page-level ranking tricks built for ten-link result pages. Structure and evidence beat volume.

What the work looks like

Three layers: the machine layer (robots.txt, llms.txt, JSON-LD, one canonical entity page), the answer layer (ten to twenty-five answer blocks on real buyer questions, plus rewrites of the pages assistants already try to quote), and the proof layer (a baseline citation matrix on day three and a weekly delta of questions won and lost).

How it is measured

A fixed prompt set — twenty to thirty real buyer questions — run against four or five engines, recorded weekly in a matrix: question, engine, which businesses were named, was the client named. That is the only honest scoreboard, because it is the same question the buyer asks.

How to get named in ChatGPT · Pricing