Marketing has a magnificent habit of inventing new names for old shit.
We turned direct mail into CRM. Advertising became content. Sales promotions became growth hacking. Customer service became community. And every few years someone discovers segmentation, gives it a stupid new name and announces that marketing has changed forever.
So I understand the scepticism around Generative Engine Optimisation.
GEO sounds exactly like the kind of acronym a conference organiser invents because SEO has stopped selling tickets.
Except this time something genuinely has changed.
GEO is not SEO with ChatGPT bolted onto it. It isn’t content marketing with an AI dashboard. And it isn’t another channel for the SEO team to quietly absorb while everyone else carries on as before.
It is a new marketing discipline because, for the first time, there is a machine sitting directly between the customer and the market, interpreting what the customer wants, deciding what attributes matter, gathering evidence, comparing alternatives and frequently making the recommendation itself.
Google’s AI Mode can break one customer question into multiple subtopics and issue a series of searches simultaneously. Google calls this “query fan-out”. ChatGPT Search does something similar. So do all the AI platforms.
Think about the implication.
A customer asks one question.
The model asks many.
And you don’t get to see most of them.
That is the beginning of GEO.
The keyword is no longer the unit of competition
SEO gave marketers a wonderfully observable market.
A person searched for “best steakhouse Manchester”. You could see approximately how many people searched for it. You could see who ranked. You could inspect the pages above you. You could estimate the traffic. You could build a better page and watch yourself crawl from position 14 to position six to position three.
There was uncertainty, obviously. But there was also a visible battlefield.
GEO is different.
Imagine someone asking:
“Where should I go for a really good steak dinner in Manchester for my wife’s birthday? There are six of us, one person has a severe shellfish allergy, we want somewhere that feels special but not stuffy, and ideally no more than £80 a head.”
That is not a keyword.
It is a bundle of requirements.
The model can decompose it into location, cuisine, occasion, atmosphere, price, allergy handling, reviews, opening hours and group suitability. It can retrieve information against each of those criteria and then reassemble the evidence into a recommendation.
The restaurant is not competing for a keyword. It is competing to satisfy a recommendation condition.
That’s useful to understand because recommendation conditions will become one of the central objects of GEO.
I say this because this week I’ve read several posts from SEO industry influencers about trying to figure out how one chunk of text in an article becomes part of an AI answer.
Most of these posts get hundreds of likes and comments, and yet one comment in one post summed it up: “How do we use this information?”
The answer was basically “I don’t know”.
And they don’t. Aside from the sheer work of finding the data, it’s basically pissing the VC backers’ funds up the office wall because it’s commercially useless and only interesting to an SEO nerd.
It’s the SEO industry equivalent of panning for gold in a dirty stream, hoping to get rich. And when someone finds a nugget, everyone in the SEO industry gets excited until the reality hits. It’s fool’s gold.
And that’s largely what is happening right now. SEO pros are panning for gold, sharing every time they find something; only when the cheers die down, we’re no further forward.
As I’ve said before, the SEO industry is damaging itself because it is failing to grasp what’s really happening; we are moving from a keyword-based world to something vastly more complex.
“Outdoor kitchen” is a keyword.
“Modular outdoor kitchen for a coastal property that won’t corrode and can stay outside all winter” is a recommendation condition.
“Baby toy” is a keyword.
“Toy for a three-month-old that helps tactile development, is easy to grip and is safe to chew” is a recommendation condition.
“CRM software” is a keyword.
“Simple CRM for a 20-person professional services firm with no dedicated sales operations team and a Salesforce migration we don’t want to repeat” is a recommendation condition.
Traditional search forced complicated human needs into short search strings. Generative search does the opposite. It invites people to put the complication back in.
And that means the market you need to understand has suddenly become much more granular.
The GEO team cannot simply produce a list of 500 keywords ranked by monthly search volume. It needs to understand the circumstances under which a customer should choose the brand.
That’s a strategic problem before it is a technical one.
AI creates an invisible consideration set
There is another difference that I think marketers are underestimating.
For decades we have talked about consideration sets: the relatively small collection of brands a buyer seriously thinks about before purchasing.
Search engines expanded those sets because they made alternatives discoverable. But customers still did a lot of the assembling themselves. Google handed you ten blue links. You opened five tabs. You read reviews. You compared prices. You decided which brands deserved further inspection.
AI search increasingly does the assembly for you.
It creates what I would call an invisible consideration set.
You ask the question and the model decides which brands deserve to appear in the answer.
You usually have no idea which other brands were evaluated and discarded. You don’t know which source contradicted your claim. You don’t know whether you were eliminated because of price, location, weak evidence, missing product information, an outdated article or simply because another brand was described more clearly.
You see the answer.
You don’t see the audition.
That is a fundamentally different problem from classic SEO.
And it explains why GEO cannot be reduced to “ranking in search to win with AI”
Yes, SEO therefore remains enormously important. I’d argue that as more people build with AI, SEO and its technical aspects become incredibly important to ensure websites can be found and discovered and not buried under code AI struggles to access.
But retrieval is only one stage in a much larger system.
A page can be retrieved, and your brand still not make the answer.
A page can be cited, and your brand still not be named.
A brand can be named but not recommended.
And a brand can be recommended for the wrong reasons, to the wrong customer, which is arguably even worse.
That’s an important warning for the emerging GEO measurement industry.
A citation is not a sale.
Your website and brand need evidence, not just more content
This brings us to the practical bit.
For years, businesses have been told they need “more content”.
More blogs. More landing pages. More FAQs. More articles explaining seven ways to do whatever it is the company sells.
AI search makes that advice look increasingly inadequate.
What brands need is evidence architecture.
Start with the performance attributes that should cause the brand to be recommended.
Not the mission statement.
Not “we believe in quality”.
Not the values HR made everyone vote on.
The concrete reasons why this business is the right answer under a specific recommendation condition.
Go back to a Manchester restaurant.
If severe-allergy handling is one of its performance attributes, don’t imply competence with a tiny allergen icon next to the prawns.
Show the evidence.
Explain the process when a severe allergy is declared during booking. Explain whether separate preparation areas exist. Explain fryer practices. Explain what the restaurant can safely accommodate and what it cannot. Explain staff training. Keep the information current. Put it somewhere both a customer and a retrieval system can actually find.
Then reinforce that evidence elsewhere.
Booking platforms. Local press. Interviews. Reviews. Videos. Third-party articles. Wherever there is a legitimate reason for the claim to appear.
You are trying to create consistency between claim, evidence and corroboration. You are trying to win more business.
That is subtly different from traditional content marketing.
The objective is not just to publish.
The objective is to make a proposition sufficiently explicit and sufficiently evidenced that a generative system can confidently use it when constructing a recommendation.
The same logic applies to a sensory toy company.
If the product was genuinely designed around tactile development, “perfect for little hands” is piss-poor evidence.
What materials were selected?
Why?
For what developmental stage?
What does the texture do?
Who tested it?
What did they find?
What age range was involved?
What makes the product objectively different from another lump of silicone shaped like a giraffe?
Those facts were always commercially valuable.
What GEO changes is the penalty for failing to articulate them. And most brands fail to talk about themselves.
The model has two opportunities to misunderstand you
There is another peculiarity here that makes GEO distinct.
A brand can enter an AI answer from at least two broad directions.
The model may already have associations with the brand based on what it learned previously. And when search or retrieval is involved, it can also gather current information to help construct the answer.
That gives the marketer two jobs.
You need the right broad representation of the brand across the information environment.
And you need current, retrievable evidence that supports the specific recommendation being made now.
One without the other is fragile.
A famous brand with a strong historical reputation but shit current product information can lose.
A technically immaculate website belonging to a brand nobody talks about can also lose.
This is one reason why the emerging discipline cuts so inconveniently across existing organisational boundaries.
The technical SEO person can make the information accessible.
They cannot create the product truth.
The product marketer can identify the performance attributes.
They cannot generate independent corroboration.
PR can generate third-party discussion.
It cannot repair a website that explains nothing.
The brand team can establish what the company wants to stand for.
It cannot simply declare that positioning into existence and expect a model to believe it.
GEO sits across all of them because the model sits across all of them.
That is new.
Prompt cultivation changes the demand side too
And then we get to the most interesting part.
Most GEO discussion focuses on influencing the answer.
I think the bigger opportunity may eventually be influencing the question.
Call it prompt cultivation.
Imagine you manufacture premium outdoor kitchens.
If the prompt is simply:
“What’s the best outdoor kitchen?”
You have a difficult problem.
The largest, most famous brands begin with an enormous advantage. They have distribution, awareness, reviews, coverage and years of accumulated signals.
But perhaps your product has specific advantages that matter enormously once a customer understands the category properly.
Maybe the frame uses a particular grade of stainless steel because coastal salt destroys cheaper materials.
Maybe components are modular because weather damage usually affects one part rather than the whole unit. Maybe ventilation and heat tolerances were designed specifically to reduce risks that cheaper installations handle badly.
The customer’s first prompt may contain none of that.
Because the customer doesn’t know those things matter.
Marketing can change that.
An article explaining what five winters do to an incorrectly specified outdoor kitchen changes the customer’s knowledge.
A video showing corrosion differences between material grades changes it again.
An architect explaining what they specify for coastal properties changes it further.
A piece of PR covering outdoor kitchen fire risk introduces another criterion.
Now look at the prompt:
“What’s the best modular outdoor kitchen for a coastal house that can stay outside all winter and uses materials that won’t corrode?”
That’s a completely different market.
The customer has acquired vocabulary.
The vocabulary has created criteria.
The criteria have changed the prompt.
And the prompt has changed the brands that deserve recommendation.
That is prompt cultivation.
Advertising has always influenced what people want. Branding has always influenced what people remember. Category education has always taught customers what matters.
But generative search creates a new mechanism through which that learning is expressed: the customer literally gives the learned criteria back to the machine as instructions.
That feedback loop is worth understanding.
Your marketing teaches customers how to describe the problem.
Their description changes what the AI looks for.
What the AI looks for changes which brands it recommends.
That is not conventional SEO.
And it is potentially very powerful.
This also means GEO begins with market orientation
There is an obvious danger here.
Marketers will start inventing hundreds of theoretical prompts in a workshop.
Then someone will call them “prompt personas”.
Then somebody else will draw them on a hexagon.
And before long we will have created an entirely new branch of professional masturbation.
Don’t.
Good GEO must start where good marketing starts: with the market.
What are customers actually trying to accomplish?
What trade-offs matter?
What anxieties enter the decision?
What language do people use before they understand the category?
What language do they use after?
Which performance attributes actually alter choice?
Which are irrelevant internal obsessions?
You are not the consumer. And your imagined prompts are not consumer data.
The discipline therefore requires a new kind of research.
Keyword research still matters, but it is no longer sufficient.
Marketers need prompt-space research: identifying recurring recommendation conditions, the criteria embedded within them and the different ways a need becomes articulated as customers add context.
The question is no longer simply, “What are people searching for?”
It becomes:
Under what circumstances should we be recommended?
That is a much harder question.
It is also a much better marketing question.
GEO changes what the marketing department has to produce
Once you see the discipline this way, the organisational implications become clearer.
The output of GEO is not a blog calendar.
It is a connected system.
First comes the prompt map: the situations, customers, needs and constraints under which the brand deserves to appear.
Then come the performance attributes: the specific reasons it deserves that recommendation.
Then comes the evidence architecture: the facts, proof, pages, product information, third-party validation and current data required to establish those attributes.
Then comes distribution and corroboration: getting those ideas represented credibly beyond the company’s own website.
Then comes prompt cultivation: teaching the category enough about the buying criteria that better customer questions start to emerge.
Finally comes measurement: monitoring whether models retrieve, cite, mention, describe and recommend the brand for the recommendation conditions that actually matter commercially.
That is a discipline.
And it requires capabilities currently scattered across SEO, brand, product, content, PR, analytics and customer research.
The obvious organisational mistake is therefore to hand the whole thing to whichever department discovered the acronym first.
Usually SEO.
That isn’t a criticism of SEO people. In many businesses, they understand AI search better than anyone else.
But giving GEO entirely to SEO would be like giving ecommerce entirely to the web development team because they built the checkout.
Technical competence is essential.
It is not strategic ownership.
And the commercial impact is arriving quickly
None of this would matter very much if consumers weren’t actually using these systems to make choices.
But they are. You are. More people are using AI daily.
Tactics will change. The underlying problem will not.
A machine is becoming an active participant in customer choice.
Your job is to understand when your brand deserves recommendation, make the evidence for that recommendation machine-readable and human-convincing, build corroboration around it, and influence the market so customers ask questions that expose the differences you are actually good at.
That requires SEO.
But it also requires brand strategy, product knowledge, PR, customer research, content and analytics.
Not because GEO is secretly all those old disciplines wearing a new hat.
Because a genuinely new discipline has appeared between them.
For years we optimised webpages to be found.
Now we have to optimise brands to be understood, evaluated and recommended.
That’s a much bigger job.
And marketing departments that dump it on the SEO team’s desk haven’t understood the brief. This is a new challenge.
It will require unique marketers with unique skills, and that doesn’t automatically mean it is the SEO team.
Find the people, find the budget…you’re going to need it.
Andrew Holland
Originally published on LinkedIn on 13 September 2026.