Every few years our industry discovers a new acronym and declares the old discipline dead. Most of the time the claim is marketing. This time it is partly true, but not in the way most of the commentary suggests. Here is the honest boundary between search engine optimization and generative engine optimization, based on what we measure rather than what we would like to sell.
What genuinely changed
The consumer changed. Classical SEO optimises for a ranking algorithm that orders a list of documents. GEO optimises for a language model that reads a set of retrieved documents and assembles a single answer, choosing which sources to quote and name. Those are different objectives with different winning conditions.
In a ranked list, being fourth is still being visible. In a generated answer, being fourth usually means being absent. There is no scroll, no page two, no long tail of curious clicks. Three or four brands get named and everyone else is edited out of the conversation before the buyer ever knows they existed.
The second change is format. Ranking rewards comprehensiveness, a long page that covers a topic thoroughly. Retrieval rewards extractability: a self-contained passage of forty to sixty words that states something specific, attributably, without requiring the surrounding context to make sense. Those two things are not opposites, but optimising for one does not automatically deliver the other.
What did not change at all
Crawlability. If a model's retriever cannot fetch and parse your page, nothing else matters. Server-rendered HTML, fast responses, clean heading hierarchy and stable URLs are as important as they ever were, arguably more so.
Authority. Models weight sources they have reason to trust, and the signals they use (links, citations, mentions on established domains, consistency across independent sources) look remarkably like the signals search engines have used for twenty years.
And demand. A page nobody has a reason to read will not be retrieved either. The fundamentals of understanding what your buyers actually want to know did not get easier.
Where the work genuinely differs
Entity consolidation is the single biggest practical difference. Search engines tolerate ambiguity about who you are because a human clicks through and resolves it. A model resolving your identity across contradictory sources hedges, and hedging means omission. Making your name, address, founding date, service list and leadership identical everywhere a model looks is unglamorous work that moves the needle more than any content project.
Original data is the second. Models cite what only one source can provide. A study nobody else has run gives an engine a reason to name you specifically, and, usefully, gets you cited in answers about your competitors too.
Measurement is the third, and it is where most programmes fall down. Ranking is a stable, checkable number. Citation share is a sampled statistic that varies between runs. You need a fixed prompt set, repeated sampling and trend reporting, not a screenshot of a single good answer.
The takeaway
Run them together. GEO without SEO fundamentals fails because nothing gets retrieved. SEO without GEO increasingly wins rankings on queries that no longer produce clicks. The programmes share a foundation and diverge at the top, and the brands winning right now are the ones treating that as one discipline with two scoreboards.