GEO and SEO: what changes, what does not

Two stories are going around, and both are wrong.

The first says GEO is just SEO with a fashionable name, that you should carry on, that good practice is enough. The second announces the death of search and sells an entirely new method, priced accordingly.

Reality is duller: part of the work transfers, part of it does not, and you need to know which is which.

What transfers

Anything that makes a page legible to a machine keeps its value.

Content that states explicitly what you do, for whom, and where. A clean page structure. Correct markup. A site robots can reach, that loads, that does not hide its content behind JavaScript only a browser can run. A language model performs no miracle on a page empty of meaning: it can only cite what it was able to read.

Structured data keeps its usefulness too, for a simple reason: it states facts without ambiguity. Opening hours, an address, a type of business, a price. That is exactly the kind of information an engine repeats willingly, because it does not have to interpret it.

Off-site reputation, the old link question, changes shape but not nature. What counts is less the link than the mention: being named on a page that carries weight in your sector. A link is still useful. A mention with no link, worthless in classic SEO terms, now has value.

What does not transfer

Position, first. There is no number one slot inside a written answer. There is being named or not being named, and possibly the order the names come in, which is not stable. Any argument that starts with "we rank first for" has nothing to grip onto here.

Search volume, next. In SEO you pick your battles from monthly volumes. In a conversation with an engine, phrasings are long, personal, often unique. You do not target a query, you cover a subject and an intent.

Landing pages, last. In SEO every keyword gets its page. In an AI answer the user often sees no page at all: they read a paragraph, remember a name, and call you directly. The click, which used to be the measure of everything, becomes optional.

Three reasoning errors I see often

"We rank well, so we will come up." Wrong in both directions. A good rank helps retrieval, it does not guarantee a citation. The reverse happens too: companies ranking poorly that come up constantly, because they sit in the directories and comparison pages the engine consults.

"We just need to write AI-optimised content." There is no writing formula that forces a model to cite you. What exists is clear, factual, dated content, and above all presence on the pages where the lists get made. Text alone does not do it.

"We'll test it by asking ChatGPT." Testing by hand gives you an answer, not a measurement. Ask three times, you get three phrasings. Ask from your usual session and the model takes your history into account. What you observe is not what your customer sees.

How to organise, concretely

The simplest approach keeps SEO as the base and adds two workstreams.

The base does not move: a legible site, content that says things, structured data, fast pages. That is the entry fee, and it serves elsewhere too.

First added workstream: third-party presence. Take inventory of the pages that, in your sector, list several companies. Be on them, with the right information. That work looks more like press relations than like search optimisation, and it is measured in sources cited by engines, not in backlinks.

Second workstream: the measurement itself. A fixed set of questions, asked regularly to the same engines, recorded with their date. Without that you are working blind, and you will never have the answer to the only question that counts: did anything move.

And the budget

A fair question, because no extra budget falls from the sky.

The honest trade-off: if your classic search work is weak, start there, GEO will not make up for it. If it is sound and your sector is already being asked about in plain language, the two workstreams above beat one more blog post.

Which leaves the matter of where you actually stand. That is exactly what the Scan measures, and it costs nothing.