GEO is an e-bike, not a spaceship
Opinion
An e-bike is still a bike. But as it evolved, the market around it was rebuilt due to its impact on couriers and rentals. LLMs are doing exactly that to agencies. The restructure is the strategy, not a consequence to manage afterwards.
Last year, I argued in The Media Leader that search traffic was always a vanity metric, and that its decline should trouble no one who cares about outcomes. Since then, the question I have been asked most is: what does this mean for how an agency is actually built? The industry has offered two answers, and both are wrong.
The conference circuit calls it an extinction event. One side foresees two-thirds of agency tasks automated, the post-agency era, our entire industry model torn up and rebuilt from nothing. The SEO veterans say the opposite. Relax, they tell us: GEO is mostly just good SEO; the fundamentals hold, and nothing structural needs to change.
Each is right about one thing and wrong about the thing that matters
The technology is an evolution. An e-bike is still a bike. Content that is genuinely useful, well-structured, and that quickly serves wins now, just as it did in 2010. But the e-bike did have a significant impact on the trade around it, from courier fleets to rental bikes on urban street corners.
The vehicle evolved, the market around it was rebuilt. Large Language Models (LLMs) are doing exactly that to agencies, and the restructure is the strategy itself, not a consequence to manage afterwards.
We made these changes at Havas Market UK before I had a tidy theory to justify them, which is the order I would recommend to anyone.
The first was to appoint an AI product and partnerships director whose remit is to manage the buy, build or partner decision behind every client deployment. The value of the role is not that it sounds modern. It is that the decision now sits with one accountable person rather than being dispersed across account teams, each acquiring their own overlapping tools and quietly billing the cost to clients. A single owner of that decision is the difference between a coherent stack and a procurement accident.
The second change was to stop treating search optimisation and generative optimisation as separate concerns. Our SEO function became a combined SEO/GEO discipline, with specialists dedicated to generative engine optimisation, because discovery through LLMs had become an essential offering our clients needed.
The veterans who insist that most of GEO is good SEO are correct about the craft. They are wrong to conclude from this that the operating model can stay where it was, because the thing being optimised for has changed even where the techniques have not.
Underneath both sits Converged.AI, the group’s AI operating system, which brings data, planning, activation, and creative production into a single client-facing stack rather than a sequence of handoffs between teams, each guarding their own tools and their own version of the truth.
This is the point at which the comfortable analogy reaches its limit, and it is worth being honest about how.
An e-bike is still a bike, and the craft of optimisation is broadly continuous with what came before. But the force reshaping the agency is not GEO.
GEO is the visible edge, the part clients ask about because it touches their traffic. The deeper change comes from the LLMs themselves altering how work is produced inside the building, not merely how content is discovered outside it.
One is a demand-side shift in how audiences find brands. The other is a supply-side shift in the way the work itself gets done.
Conflating the two is how firms end up optimising their content for AI while leaving the organisation that produces it entirely untouched, which is the most expensive kind of half-measure.
Emerging roles and job titles
That distinction is what makes me confident about the shape of the next eighteen months. The defining move will be from agencies that use AI to agencies that build with it, and the roles that matter will reflect that.
I expect agentic workflow architects, who design how teams of AI agents collaborate across media, measurement and reporting, to become as load-bearing as media planners were a generation ago.
I expect AI-quality leads, with human oversight, prompt governance, and brand safety in generative output, because the cost of an ungoverned model speaking in a client’s voice is not theoretical.
And I expect commerce AI strategists who pair retail media expertise with synthetic audience modelling and predictive decisioning, since commerce is where the machine-mediated economy often becomes a balance-sheet question rather than a slide in a trends deck.
Each of these roles, should they emerge in our industry, would be a deliberate placement of human judgement at the point where machines need supervising, and that is the whole argument.
The agencies that win the next phase will not be the ones holding the longest list of licences or burning the most tokens. They will be the ones whose roles, ratios and incentives have been rebuilt so that people are paid for the judgement only they can supply, and machines are trusted with the volume only they can handle.
AI token usage is the new vanity metric, and like search traffic before it, it flatters the people measuring it while telling them almost nothing about whether they are built to win.
I argued last year that we should stop mistaking traffic for outcomes. The same discipline applies now to the organisation itself. The question is not how many AI capabilities your agency can demonstrate. It is whether anyone has truly redesigned the structure underneath them, or whether it is the same one you have always had, just running different software.
Search was always a vanity metric – why I’m unworried about its decline
Alex Walker is the MD of Havas Market UK