I remember the first affiliate payment I received from a website I’d built.
It wasn’t the size of the payment, although it paid for my car to remain on the road after the exhaust fell off; it was the discovery.
I’d made something, people had found it through search, and money had arrived because of it. The internet suddenly looked like somewhere I could build an asset rather than just spend my time on it.
If you worked in SEO ten years ago, you’ll recognise that feeling. The industry was full of stories – affiliate sites, publishing businesses, people who’d turned a website into a serious income. Some of it was exaggerated. But the underlying proposition was real: build a useful site, earn rankings, attract an audience, turn that attention into revenue.
Go looking for SEO content today and you’ll find a rather different mood. Articles about query fanout, citation overlap, structured data, which pages appear in AI answers. A lot of that work is useful. I read it myself. But it’s hard to find the old commercial energy in any of it.
We’ve become very good at discussing how a machine retrieves a page. We’re less comfortable answering the question a business owner asks next:
How does any of this help me sell more?
The economics that made SEO exciting have shifted. A ranking can still be valuable. Websites still generate serious revenue. But the old promise of create more pages, win more traffic, grow from there – is much harder to make with a straight face.
People discover businesses, gather evidence and make choices across search results, AI answers, social platforms, videos, communities, reviews and conversations. A company’s website is part of that journey. It’s rarely the whole journey.
This leaves SEO with an uncomfortable question. If growth now depends on a brand being visible across that wider environment, how much of that growth can we actually optimise into existence?
First, give optimisation its due
I’m an SEO. I’ve no interest in pretending the technical work has suddenly become irrelevant.
Sites still have pages that can’t be crawled properly. Internal links still fail to connect important information. Product pages still omit the details people need to make a decision. Service pages still speak in vague claims instead of explaining who the service is for, what it does and why anyone should trust it.
AI can help identify and fix some of these problems. That doesn’t mean every suggested fix is correct, that every implementation is safe or that somebody no longer needs to decide which work matters. If anything, the ability to change a site quickly makes informed oversight more valuable, not less.
Google’s own guidance for its AI search features continues to stress the basics: pages need to meet Search’s technical requirements, important content should be accessible, and internal links should help make it findable.
So yes, a technical SEO has an important and defensible job. Make the site work. Make its important information accessible. Find the commercial opportunities the site is missing. Check what automated systems change. Protect a valuable asset from mistakes.
The amount of help a business needs will vary. A hairdresser in Wigan doesn’t necessarily need the same search team as a national tile retailer with thousands of product pages. A large ecommerce business may have enough technical and commercial work to keep specialists occupied indefinitely.
This is the plumbing of visibility. I mean that as a compliment. When the plumbing fails, everything built on top of it suffers.
But once the plumbing works, where does the next wave of growth actually come from?
The old growth job was easier to describe
For years, the SEO growth model had a reasonably clear structure. Research the keywords. Build pages that match them. Improve the site. Earn links. Increase rankings. Turn the traffic into customers.
There was nuance, naturally. The best SEOs understood audiences, products and markets. But the website was the centre of the plan, and search demand gave us a practical way to decide what to create next.
Now imagine a potential customer asking an AI assistant:
“Which floor tiles would work best in a busy family kitchen, and which brands should I consider in the US?”
The interesting part isn’t simply whether a tile retailer has a page containing the words busy family kitchen. The assistant may compare materials, weigh durability claims, factor in customer experiences, pull from guides and retailers, and produce a shortlist with reasons.
The business wants to be on that shortlist. Better still, it wants to be recommended for an accurate reason that actually matters to the buyer.
A perfectly functioning website helps. A clear product page helps. Neither, on its own, gets you on the list.
That’s the gap the “it’s just SEO” conversation tends to miss. Technical eligibility and market preference are connected, but they’re different problems.
Start with the decision, not the citation
I’ve spent nearly two years working with prompt mapping. Done properly, it’s a way to investigate the decisions AI systems appear to make in a particular market.
Start with two broad groups of prompts.
Branded prompts reveal how a system describes a business when the user already knows its name. What does it think the company sells? Which strengths or weaknesses does it mention? Does it understand the offer accurately?
Unbranded prompts test whether the business appears when a person describes a need without naming a supplier. These are especially interesting when someone asks for options, comparisons or a recommendation.
If you only test broad informational questions, you’ll learn a great deal about the pages cited for broad informational answers. You may learn much less about the moment when someone actually chooses a provider.
Say you sell car hire with comprehensive insurance. “How does car hire insurance work?” is a useful question. But so is: “Which car hire companies in Spain offer clear cover for a family trip without a large excess?”
The second prompt creates a decision situation. It helps you examine which companies are recommended, which criteria the assistant uses and what evidence appears to support the answer.
I use examine deliberately. We don’t get to see every step inside a model. Prompts produce different responses over time or under different conditions. A source appearing in an answer doesn’t prove it caused a particular brand to be recommended.
The analysis is still worthwhile. You can repeat prompts, compare answers, inspect cited sources, record where competitors appear and look for patterns. You can form a defensible view of the information environment in which customers are making choices.
That’s a more useful commercial exercise than counting citations and declaring victory.
Recent industry writing has rightly explored query fanout – the additional searches a system may perform while building an answer. Worth examining.
But don’t let it become the strategy. The fanout happens in service of the original question. If we optimise endlessly for the searches along the way and forget the recommendation at the end, we’ve mistaken the mechanism for the outcome.
Then ask who is shaping the answer
Once you have a set of relevant prompts, look beyond your own website.
Which brands are repeatedly recommended? What reasons are given? Are those reasons visible on the brands’ own product pages? Do comparison articles reinforce them? Are reviewers saying something different? Which questions seem to expose a gap in the market?
The answers may lead you towards better product descriptions. They may point to a missing explanation of how your service works. PR might develop a story that demonstrates expertise. Customers may need better opportunities to describe their experiences. A video might answer a question that written copy handles badly.
This is where it gets both more interesting and more difficult.
Someone has to decide what to do with that evidence.
Nor should we pretend the evidence offers certainty. If a competitor receives a recommendation alongside a particular review, we can’t simply announce that the review produced the recommendation.
What we can say is that the review is part of the visible evidence in a market where that competitor is being chosen. That’s a reasonable starting point for a hypothesis and a test.
This is why the person doing the work needs more than familiarity with an AI visibility dashboard.
They need to understand customers, content, websites, communications – and the difference between an interesting pattern and a proven cause.
Someone still has to create the visibility
Optimisation language tends to hide this part.
If an analysis shows that your brand lacks convincing evidence in the places buyers encounter it, someone has to create that evidence and get it seen.
A useful article published on your website is a new asset. Put it somewhere prominent, send it to your email audience, share it through people who have reason to care, and more humans can encounter it. A strong product demonstration creates another opportunity. So does relevant coverage, a good video, or a customer account of what using the product was actually like.
The point isn’t to flood the internet with mentions and hope an AI assistant picks them up. The point is to give people and systems more accurate, useful material from which to understand the brand.
This is ordinary marketing work with a new way of observing some of its effects.
It’s also why I hesitate when someone tells me GEO is simply SEO with a different acronym. If the recommended response to your findings involves PR, social, video, product messaging and distribution, you’re discussing decisions that extend well beyond the traditional SEO brief.
Some SEOs will be entirely comfortable there. I am.
My career has involved copywriting, content, social and building a newsletter, as well as search. I like connecting those disciplines to a commercial outcome.
Other SEOs won’t have any desire to decide what a PR team should pursue or whether a business needs a TikTok strategy. They may be exceptional at technical search and at identifying valuable ranking opportunities. There’s nothing wrong with choosing that path.
But agencies and businesses should be honest about the job they’re assigning. You can’t hand someone responsibility for AI recommendations, give them a crawl budget and a list of page fixes, and assume you’ve equipped them to influence what the market says about the brand.
The answer is a shared intelligence function
Could PR teams map prompts themselves? Of course. So could social teams, copywriters and product marketers. The mechanics are learnable, and plenty of tools will even suggest prompts for you.
The problem is what happens when every team builds its own small picture of AI visibility.
PR tracks the questions that make coverage look useful. Social tracks prompts where its posts appear. SEO tracks cited pages and rankings. Everyone presents a chart proving their own work matters. Meanwhile, nobody has a coherent account of which decisions customers are making, what the brand is known for or why a competitor keeps entering the shortlist.
I’d put a GEO specialist – or someone with an equivalent remit – at the centre of that intelligence.
Their job is to build and maintain the prompt map, examine recommendation patterns, study the sources and claims shaping answers, and bring the findings into a shared strategy. They work with the other teams on priorities and assess whether the work is moving the brand towards more relevant recommendations.
They should also have a direct role in the brand’s owned commercial pages.
Product and service pages are where a business explains itself. Too many fail at the basics – burying the important distinction, leaving buyer questions unanswered, using language so generic it could describe every competitor.
Someone needs to look closely at those pages, down to what each claim actually means, and improve them with product experts and writers.
On a large site, I’d keep deep technical SEO as a dedicated specialism. The GEO role already contains enough analysis, communication and strategic work.
On a smaller team, one person may cover both for a period. Team design should follow the workload and the skills available, rather than an argument over whose acronym owns the work.
The outward creation of visibility belongs with the people best equipped to do it. Let PR develop and place stories. Let social specialists understand their platforms. Let writers, editors and video teams produce material worth encountering. Give them shared intelligence, involve them in deciding what it means, and ask each team to make the case for its proposed activity.
The GEO specialist doesn’t need to write every article, pitch every journalist or run every account. They need to help the organisation see the same market and make better choices within it.
Measure the decision that matters
There’s a trap in this model. We could replace rankings with a new collection of comforting activity metrics.
We tracked 500 prompts. We published 17 videos. We appeared in 30 more citations. The dashboard moved.
Fine. What changed for the business?
Did the brand begin appearing in relevant shortlists? Did the reasons given for recommending it become more accurate and more persuasive?
Are more buyers arriving with a clearer understanding of its strengths? Are the enquiries better? Is there evidence of growth in consideration, demand or sales alongside the visibility data?
No single metric answers every one of those questions. AI responses vary, attribution is messy, and other marketing activity affects the outcome.
Treat recommendation tracking as one source of evidence, then connect it to customer research, search behaviour and commercial results. Even guidance on AI visibility dashboards acknowledges the limits of measuring a changing answer with any precision.
That may feel less tidy than watching a keyword climb from position seven to position three. It’s also closer to the problem a business has hired us to solve.
I still believe in optimisation. Fix the site. Improve the page. Make the offer clearer. Help a machine and a person understand what the business actually does.
But if nobody creates the work, experiences, explanations and independent evidence that make a brand worth considering, there’s a limit to what optimisation can achieve.
The next growth job is to understand what leads to a recommendation, help the right people create that visibility, and keep testing whether the brand is becoming more likely to be chosen.
You can optimise a page. You have to build a reason to recommend the brand.
Welcome to the new era of AI search.
Andrew Holland
Originally published on LinkedIn on 27 September 2026.