Analysis
GEO and SEO: What Actually Changes, and What Does Not
Most of technical SEO still applies to AI search, and any agency selling a clean break is selling something that does not exist. Three things genuinely change: the unit of content that competes, where authority comes from, and whether a click happens at all.
Most of technical SEO still applies to AI search. Any agency selling you a clean break is selling something that does not exist, usually because a clean break is easier to charge for than an adjustment. Three things genuinely change: the unit of content that competes, where authority comes from, and whether a click happens at all.
Start with what does not change
This is the longer list, and it is the one nobody writes about because it makes for a duller pitch.
An assistant cannot cite a page it cannot reach, so crawlability still decides everything upstream of it. Canonical structure still determines which version of a page is treated as authoritative. Site speed still affects whether a crawler gets a complete render. Information architecture still decides whether related pages reinforce each other or compete. Clean, accurate markup still tells a parser what it is looking at.
If your site is technically broken, generative engine optimisation is not a separate project you can run alongside fixing it. It is the same project. In practice, most of what we do in the first month of a GEO engagement would have been recognisable as SEO work a decade ago.
What changes: the unit is a passage, not a page
A ranked result is a page. An answer is assembled from passages, and the passage is what competes.
The practical consequence is uncomfortable for most brand writing. A piece that takes three paragraphs to reach its point has a first passage that says nothing, and the first passage is the one most likely to be lifted. The claim has to arrive first and the evidence after it — closer to how a wire service writes than how a brand blog does.
It also means a long page can succeed in parts. A single well-constructed section answering one question precisely can be extracted repeatedly while the rest of the page is ignored entirely. Structure has stopped being navigation and become the thing itself.
What changes: corroboration replaces backlinks
Backlinks were always a proxy. The theory was that a link represented a vote, and enough votes represented consensus. It worked well enough for twenty years and it was gamed for nineteen of them.
An answer engine does not need the proxy, because it can read the consensus directly. If five independent sources describe your brand in compatible terms, that agreement is legible to the model without anyone having to link to anything. If the only place a claim appears is your own website, it carries close to nothing, however well optimised the page is.
This is the single biggest strategic shift, and it moves budget. It means public relations, review platforms, retailer copy and creator coverage are now search infrastructure rather than a separate discipline with a separate agency and a separate reporting line. We have written separately about how influencer seeding produces exactly this kind of independent corroboration, which is a use for it most brands have not considered.
What changes: the click may not happen
The third change is the one with the sharpest commercial edge. A searcher who gets a complete answer has no reason to visit the source, and increasingly does not.
That has two consequences worth planning for. The first is that traffic becomes a worse proxy for influence than it used to be. You can be named in answers all day and watch sessions stay flat, which looks like failure on a dashboard built for the old model.
The second is more subtle: if a click is less likely, the value of being named correctly goes up. A brand mentioned accurately, in the right category, alongside the right peers, has done most of the commercial work even without the visit. A brand mentioned vaguely or wrongly has spent the same attention badly and has no page view in which to correct the record.
What that does to measurement
| Search optimisation | Answer optimisation | |
|---|---|---|
| Primary metric | Rank and organic sessions | Share of answers naming you |
| How it is captured | Continuous, automated | Sampled, repeated on a schedule |
| Precision | High | Low, and honest about it |
| Secondary signal | Click-through rate | Referral traffic from assistant domains |
| Leading indicator | Impressions | Branded search volume |
Branded search deserves particular attention. It is the cleanest available read on whether AI visibility is doing anything commercial, because someone who hears your name in an answer and wants to buy will usually search for you directly. If assistants are naming you more and branded search is flat, the mentions are not landing.
What we would actually change on Monday
Rewrite the openings of your best-performing pages so the answer arrives in the first sentence. It is the cheapest change on this list and it costs nothing but editorial discipline.
Audit which crawlers you are blocking, because the training crawler and the search crawler are different bots and a great many brands blocked the wrong one in 2024.
Then stop treating press and search as separate budgets. On the evidence of how these systems actually assemble answers, they have not been separate for a while.
Want this run over your own brand?
We work with beauty, fashion and aesthetics brands across the UK, USA, Spain, France and the UAE.