The question is put as a choice: do we carry on with conventional search or switch to optimising for AI engines? In that form it makes no sense. A generative engine feeds on indexes built by ordinary crawlers.
A page a conventional engine cannot reach is no more read by a generative one. The groundwork is shared, and it is not optional.
What both require equally
- crawlable pages, served quickly, with no crawler blocking and no script-dependent content
- a heading structure that reflects the reasoning, not the layout
- accurate structured data — author, date, organisation, content type
- internal linking that connects pages on the same subject
- established authority: being cited elsewhere remains the signal that is costliest to fake
A company whose site fails on one of these points has no trade-off to make. It has groundwork to repair, and any day spent elsewhere is lost.
First divergence: what wins
In conventional search, the unit being contested is the page, on a query. In a generative engine it is the statement: a sentence repeated, attributed to a source.
The practical consequence: long, complete content keeps its ranking advantage, while a clear answer placed high on the page has the citation advantage. Both fit in the same page, provided you answer before you argue.
Second divergence: what can be measured
Conventional search is measured in positions, impressions and clicks — data the tools supply. Visibility inside generative engines measures badly: answers vary from one user to another, from one session to another, and nothing keeps a public record of them.
What can be tracked: the share of referral traffic coming from conversational interfaces, and brand mentions recorded by hand against a stable set of questions, queried at regular intervals. It is handmade, and it is the best measure available.
Visibility that cannot be measured cannot be steered. It can only be observed — which is itself a reason to start early.
Third divergence: the value of coming second
On a results page, second position still receives a notable share of traffic. In a written answer, the second source cited is read, but the fourth does not exist — the answer does not reach that far down.
The distribution is therefore more concentrated. That argues for a narrower choice of subjects: better to be the reference on five precise questions than the ninth source on fifty.
How to split the effort, in practice
- Technically faulty site. One hundred per cent on the groundwork. The GEO question does not arise yet.
- Sound site, thin content. The effort goes on what you have to say, not on how to mark it up. Both approaches benefit equally.
- Sound site, solid content, rarely cited. This is the only case where work aimed specifically at citation is justified: rewriting the answers, attributing the figures, dating the page.
What remains is knowing where to start in practice. Three moves are enough, and the first costs only an hour.
In all three cases, the share of work genuinely specific to generative engines stays in the minority. What is presented as a new discipline is, in the main, an editorial standard that conventional search allowed you to do without.
Common questions
Do you have to choose between GEO and SEO?
No. A generative engine draws on indexes built by ordinary crawlers: the technical foundation is shared. What needs deciding is the marginal effort — fixing a faulty foundation, or making already-readable content citable.
What are the differences between GEO and SEO?
Three real differences: the unit that wins (the page for SEO, the cited statement for GEO), measurement (rankings and clicks on one side, mentions recorded by hand on the other) and the value of coming second, far lower in a written answer.
How do you measure visibility in AI engines?
Imperfectly: answers vary between users and sessions. You can track the share of traffic coming from conversational interfaces, and record brand mentions against a stable set of questions, asked at regular intervals.