I’m writing this as a companion piece to a recent talk of the same name that I presented at Performance Marketing World Unlocked in March 2026. You can find the slides here.
When I created the first draft of the content for the talk and realised that I had about an hour’s worth of material that I needed to cram into 12 minutes, I knew I was going to have to make some serious cuts, which is a real challenge when you are passionate about a subject – as I am this one.
Hopefully this is a truer representation of what I would have said if I was allowed on stage for an hour and also serves to answer some of the questions you may have had if you were at the talk and not couldn’t catch me for a chat at our stand.
I realise I may come across as a bit of an Learned Language Model (LLM) hater. I’m not. I studied early versions of similar technologies as part of my degree in the early 00’s. My concern is more that the current iterations of LLM technology weren’t introduced correctly. In a few years, when we start to see young adults coming into the workforce who have actively been trained on how to prompt and critically analyse the responses from these systems, things will be different and I genuinely think we will look back on the current cohort of LLMs and laugh about how basic and unfit for purpose they were.
At the end of this piece, I’ll talk about what I think the future might look like, so please stick around and hear me out.
Scene setting
In 2017, I spoke at a small SEO meeting called Brighton SEO about the concept that “brand” would be the only real ranking factor at some point. At the time, pre- ‘AI everything’, this was relatively new thinking. Me and Tom Capper from Moz shared the stage discussing concepts such as Entities, brand signals, external validation and how other channels can be leveraged to directly impact SEO performance.
Fast forward to 2026 and people are still talking about that 2017 talk. Literally just last night, I was at an SEO event hosted by Majestic SEO where we ended up discussing that talk as well as some of the things I’m about to talk about.
Brand remains a critical component of success within performance marketing and will remain as such for a long time beyond this current phase of LLM technology. The channels will change, the funnel will change, our roles as SEO will change, but the core principles of why a consumer choses brand X over brand Y remain constant. Marketing, in all its guises, still matters.
That’s just a little contextual scene-setter for you. Now, on to the actual content from the talk.
How did we get here?
I actually have a vague memory of the day OpenAI unleashed ChatGPT upon us all. It was a crazy day, which feels odd to say as it was a piece of technology rather than a world event.
Within 5 days, ChatGPT had gone from being something a few people knew about (yes, OpenAI and LLM technology had been around for some time prior to ChatGPT) to having 1 million active users and being talked about everywhere. The fastest growing technology ever. Everyone was dabbling with it to see what it could do, and a few were even trying to understand how it did it.
The SEO space reacted exactly as expected. Loads of people jumped onto LinkedIn, Reddit, blogs and the like claiming to be instant experts, often spouting pure nonsense and racing to be the first to cash in on this new technology. GEO, AIO and many other new things started to appear. Whether you wanted it or not, every tool suddenly had an AI writing assistant. People were spinning up websites populated solely with AI written content and every conference talk became about AI and how it could 100x your productivity.
It was a bit of a Wild West scenario. People were talking about how to spam Reddit with listicles (please don’t) and how to create fake authors (again, please don’t). Even now the dust has settled and we’re starting to really understand how we can utilise this technology and how it works, the subjects of GEO, LLM visibility and the death of SEO has become a frequently discussed topic. That’s not only on LinkedIn but in the monthly meeting, in the pitch and in the boardroom.
Generally, it boils down to 3 common questions:
“Is Google Dead?”
“Is SEO Dead?”
“What is our GEO strategy?”
By answering these questions, I can hopefully give you some confidence in the fact you likely don’t need to tear up your strategy, and you’re not as far behind as some people might have you believe.
“Is Google Dead?”
I genuinely can’t get my head around where this narrative comes from. Yes, ChatGPT was first to market, mentioned in every conversation and LLMs promised the world. However, being talked about and experimented with doesn’t necessarily mean you’ve replaced something else.
Google is more than just a search engine. It’s a data harvesting behemoth. I would argue that Google has more data relating to people and their habits than any other company in the world.
Yes, traffic volumes might be down, but as things stand people are still Googling a hell of a lot and it’s easy for Google to counter – you can read this as fact – anything their competitors are doing backed by an unprecedented volume of data, insight and expertise.
Gemini is a perfect example of this. Until OpenAI exploded onto the scene, Google was likely using a lot of this technology such as LaMDA etc. – they just weren’t talking much about it. Maybe they figured it wasn’t ready to be surfaced, more likely they were worried it would destroy PPC revenue, but whatever the reason they were forced to show their hand in Bard, followed by Gemini in 2023.
Recent Similarweb data shows that Gemini now has a healthy 21%(and growing) market share and this doesn’t take into account things like AI Overviews, which expose Gemini to even more people through the standard Google Search interface.
Google may technically be behind, but it’s far from being dead. I would actually argue that recent attempts by OpenAI to partner with Shopify and Etsy to introduce a browser called Atlas, in an attempt to get ahead of the Universal Commerce Platform initiative, are a sign they’re worried about the future, their increasing costs and their lack of a genuine USP outside of being first to market.
“Is SEO Dead?”
This is actually a tricky one to answer. For a long time now, SEO has been changing to be more holistic, often to pick up the bits no other channel was willing to take responsibility for, such as site speed. We’ve been discussing how you shouldn’t be doing anything solely for the benefit of search engines and often we’ve started to morph into business strategists who just happen to understand websites, search engines and performance marketing.
While I still see people excessively chasing keywords, churning out content based on scraped ‘People Also Asked’ queries and actively building dubious backlinks, a lot more SEO strategies are more akin to overall marketing strategies that span any surface the client may appear on, regularly working outside of the normal search engines.
How are we represented on Reddit? What do LLMs know about us? How are we leveraging our website as a campaign destination? All these questions and more should be common considerations within your SEO strategy, online strategy or whatever you choose to call it. As SEOs, we need to be comfortable flexing well outside of our normal remit – at the end of the day, it all contributes to client success.
Another reason why SEO may seem like it is dying is the erosion of traffic. It’s taken me a long time to not freak out when I see negative year-on-year growth figures for organic traffic. Obviously, the immediate counter is to check revenue, and in the majority of cases, revenue is the same or better. However, it still makes the conversation difficult if you can’t articulate what is going on.
It’s not uncommon for site traffic to be down 20% year on year. If you were heavily reliant on informational content, it could even exceed 60% (in some cases, it can even be closer to 80%). Affiliate marketing is struggling now as the affiliate sites are no longer generating the viewership they once did. News sites are pivoting to counter the desires of LLMs in an effort to capture as much of the informational space as possible.
The funnel is changing. This concept of a linear journey from awareness to consideration to conversion to advocacy needs reworking. For a long time Google has been talking about the messy middle, and this is now pushed even further due to these new research tools at our fingertips.
It’s not all bad though. We are lucky enough to have several clients with very engaged audiences, meaning we can ask directly how they came to their decisions. In a lot of cases, initial discovery was via Tiktok and a growing number via an LLM, but the deeper layers of consideration and eventual conversion still rely heavily on Google usage.
There is still value in a well thought out SEO strategy. Content that meets the needs of your user still has worth even if it doesn’t rank, as long as you bake it into the user journey correctly rather than stuffing it into a never visited blog. For example, the benefits of writing yet another piece about hayfever proofing your home are going to be very small unless you have something genuinely useful to say and new to add to the conversation.
The key is to actually do what people have been pretending they do for so long. Ask your customers, don’t rely on Google Keyword Planner or Answer the Public. Put yourself in their shoes first rather than hitting the tools. Do something different to all the other sites already out there.
If we deconstruct the fundamentals of a genuinely good SEO strategy, we can see how with a few little tweaks and considerations you should already be doing the right things to appease LLMs. We call this SEO Strategy+. You survived ‘Mobilegeddon’ via doing good SEO and considering the mobile user experience. This really isn’t so different and it brings us nicely on to our final question.
“What is our GEO strategy?”
I get it, everyone is talking about GEO or AIO. The LLMs themselves have been exposed to so much propaganda that even they are starting to consider GEO a thing. Like it or not, GEO has become a conversation point that needs to be addressed in an adult conversation.
The first thing I want to discuss is some common misunderstandings about how these systems work. You may have already heard me referring to them as ‘predictive text on steroids’, which isn’t quite true but is still an important statement on how people over hype what current LLMs can do well. As mentioned before, this is the start of something massive and soon we will be looking back and laughing about how we used to use such basic systems, in the same way that I look at my Pixel Watch 4 and think back to my Casio calculator watch.
“It’s not the same as searching”
While you can type exactly the same thing into an LLM that you’d have historically types into Google, the true value comes from being good at prompt engineering. Search was already becoming more about conversations as shown by the average query length growing over time, but we’re now really at a point where everyone is doing something a little bit different.
I’m genuinely hoping that nobody is trying to rank for something like “I am staying in Frankfurt in February and have a bit of spare time between 11am and 3pm on Saturday. Can you recommend something to do that’s exciting, costs less than 20 Euros and doesn’t involve much travelling?”, but maybe not for the reason you think…
The previous example prompt might be what the user types in, but it isn’t really what’s queried within an LLM and really isn’t what would be queried in Google due to RankBrain. My understanding is that an LLM does an initial piece of analysis on what you have prompted it with, it then shatters the prompt and expands upon it, effectively asking “what’s the information I would need to gather to fully answer this query?”
It then attempts to produce a response built from all the component parts that are returned. It makes a judgement call as to whether this is good enough. It is perfectly good in some cases, after all, facts remain facts etc. However, with many responses, the confidence might not be there, this is where Retrieval Augmented Generation is employed.
RAG is where it gets interesting – especially when we’re looking at this from an SEO standpoint. Very simply put, RAG involves using a search engine to capture additional information that’s fed into the response. We learnt quite early that LLMs were useless when it comes to retrieving and presenting the user with up-to-date information. It was a limitation at the time. They need training and this training results in things going out of date. RAG was introduced to counter this and make LLMs better, more accurate, and more useable. If you used GPT-2, you should now see just how far GPT-5 has come from those early days. Imagine what GPT-10 will be able to do.
So, RAG mimics someone searching in a search engine – probably Google – and how do we influence rankings in Google? Through SEO.
In addition, Google is looking at how to involve a RAG like approach within the training activity. This could potentially replace the need for RAG, or at the least should reduce the volume of responses that need to use it, greatly reducing processing costs etc.
Another thing worth noting is that this technology is only going to get more personalised over time. At present, it can be difficult to gauge how personalised responses really are due to the deliberate randomisation of responses. Rand Fishkin from Sparktoro did an excellent piece on the topic of LLM tracking, variation etc. Read it here.
“GEO is just good SEO”
It is, but only if you were actually doing good SEO in the first place. I am yet to see a technique used to influence LLMs that isn’t either short term spammy technique that’s likely to get patched at some point or something you should have already been being doing.
LLMs are just another discovery surface. Yes, they are not a search engine, but they’re pretty close. Also, who else is going to pick up the conversation and run with it? The PPC team? They’ll enter the conversation soon enough but for now it’s an SEO challenge.
If we look at the very key fundamentals for ranking in a search engine compared to being considered in an LLM, you’ll find there are huge similarities and overlaps. For a start, both require access to your website. From a technical SEO point of view, the bots employed by LLMs are often similar to the old Googlebot. JavaScript might not get rendered properly, your site might be too slow and therefore ignored, you might be accidently blocking the bots through Cloudflare. If your pages are really complicated, you might want to consider dynamic serving markdown. All these issues have been seen before and tackled as part of technical SEO.
Your content needs to be easily digestible. Semantics matter. Topical coverage matters. Cut the waffle and get to the point. Consider the best content types for your audience and the information they you are presenting. Again, this should already be part of your SEO strategy. We are just expanding our considerations to accommodate LLMs, their limitations and their abilities to be multi-modal.
Predictive text is based on probability. Could you imagine how terrible these systems would be if they were just uber predictive text? The reasons they aren’t that bad stem from the principle of external validation, co-occurrence, sentiment and repetition. How do we influence these things? Digital PR, integrated cross-channel campaigns, consistent messaging, getting people to recommend us. Again, a staple of any good SEO strategy, and that includes asking how we’re engaged with on Reddit.
“Should we track it?”
Another tricky one. Seeing a line or number go up is reassuring. It’s been a staple when marketing things forever. If you can’t track it, how do you know it is working? However, have you noticed how tracking in general has got a lot worse over the years? Cookies, AI solutions to fill the gaps, vague reporting through paid channels and now the potential promise of vague reporting through ads in ChatGPT. A lot of decisions are already being made based on flawed data and assumptions – and they have been for a very long time.
Personally, I don’t see much value in tracking beyond the limit view of LLM visits, as well as a set of synthetic queries aimed at gaining an understanding of how your brand is represented in these systems. I refer you back to Rand’s post. If tracking works for you, that’s great. But will it stand up to scrutiny? It feels a bit like the Domain Authority debate.
I do think you should regularly ask your newest customers how they found you. You may even just want to do some more general panel work to get a feel for your customers and whether they are engaging with LLMs yet. If they are, ask them how they prompted the LLM they used. They may not remember, after all, expecting someone to recall this kind of thing is a stretch. However, they may – with the right guidance – be able to query their past conversations and get you the exact information.
You should look at prompts such as “tell me about X”, “how does X compare to Y” etc. to get a feel for how your brand and products are represented. You can use a tool like Profound to track this at over time and analyse the sites cited in the responses. You could use this data to draw up a list of regularly cited sources you might want to engage with as part of your DPR strategy in a similar way to how you use the data from “best X” queries to find the external voices. How do you get included in the next version of this content?
The real challenge of AI
It’s time to pivot. I want you to think beyond the current conversations around GEO etc. Above all else, LLMs are an enabler and an equaliser. They raise the bar for everyone. Anyone can make an OK website, anyone can create fake images, anyone can create fake reviews, anyone can get started on Google Ads, Meta or TikTok. The bottom end suddenly got a lot closer to the middle. People are starting to wonder “is this AI?” and it’s only going to get worse.
While we’re worrying about the nuances of tracking LLM visibility, our niche is being flooded with new competitors – all with near identical products, all cheaper than yours.
This isn’t new. Over my 23-year tenure doing SEO, I’ve worked on numerous projects, deconstruction competitor risks and building strategies to counter them. I was working with Waterstones when Amazon released the Kindle. I was working with PrettyLittleThing when Shein arrived in the UK. I was working with Ann Summers when LoveHoney exploded on the scene.
In each of these cases, the businesses suffered. It was unavoidable. A true disruptor had arrived and was targeting a cornerstone part of their audience. Wherever you stand on the physical book vs. e-book debate, many people migrated. If you were simply buying cheap clothes, suddenly there was a cheaper option. If you simply wanted discretion, a rival was good enough to provide that too.
Waterstone’s still exists. PrettyLittleThing still exists. Ann Summers still exists. In each case they had done everything they could to build a brand and to be the default destination for what they did at the time. Some had bricks and mortar as a USP, some a huge amount of brand loyalty and pedigree as a UK brand. Yes, there were difficult board room conversations, but they survived.
I was chatting yesterday and discussed how there are three brackets in most niches. The power brands – your Louis Vuitton, your Prada. Safe, respected, aspirational. On the other end of the spectrum there are your discount brands. Your Primark, your George at Asda. They do the job and they do it very well and very cheaply. Then there are the others. The brands that aren’t that well known, aren’t the cheapest and often have no genuine USPs.
It’s important that we have an answer to the question “why would someone choose you?” It can be because you are the cheapest. Whatever it is, there has to be a reason, especially when the product that you’re selling isn’t so different from the generic unbranded version.
Rather than obsessing over how AI can improve things, take a moment to consider the things AI can’t do. This is how you build a moat around your business. How you differentiate yourself from the low effort businesses that are coming for your customers using Shopify, AI and drop shipping.
UK support, easy returns and repairs, a physical location to try products, genuine UGC and reviews, the people behind the business, TV advertising. All these things can be used to differentiate your business. It pains me to say it, but over-reliance on digital marketing is going to become a problem moving forwards. CPCs will continue to rise, organic traffic will continue to erode, new business will continue to find it harder.
It’s time to be honest and be more holistic. Companies like JD.com and Joybuy are coming to shake things up. Same day delivery is going to return to the UK. We need to build customers and advocates – not just clicks.
The potential future
As alluded to earlier, current LLM technology may seem like extreme sci-fi, but it’s just the tip of the iceberg. Imagine in the future where internet access is stable and always accessible at a very low cost through lampposts, light bulbs or some other everyday mechanism, or where delivery is fully sorted via Amazon, Uber, drones etc. LLMs have moved from being general to smaller, purpose driven models. True digital personal assistance.
You step out of your car and feel your jeans rip.
“OK Google, I need some new jeans”
“Sure, I can help with that. The last 2 pairs you purchased were both Levi 501s. The pair before that was from the M&S Japanese Denim range. Would you like to reorder from either of those or try something new. If it helps, Jane and Ashley both commented on how good the M&S ones looked on you.”
“I think I’ll grab another pair of Levi 501s”
“Great, you can get these from Levis directly at £70 and they will be here tomorrow by 10am. If you want them sooner, JeansJeansJeans, a retailer you’ve used before has them at £73 and they will be delivered by drone in 2 hours”
“Let’s go with JeansJeansJeans”
“Will you be using the usual debit card ending in 1094?”
“Yes, please”
“Great, these are ordered and we already have a response say they’ll arrive at 11:24 to your collection box”
I can genuinely see how this will become a thing. True seamless assistance. No search engine, no phone (unless you need to use a screen) and no need for human interaction. Even in this scenario, marketing and brand matters.
There’s a reason why Levis, M&S and JeansJeansJeans are in the consideration set. They’re recognised brands. They’ve done the heavy lifting that builds awareness and confidence in their offer, product and service. Social media and email can be used to knit together a deeper picture of the customer, reminding them of other people’s opinions and options. Yes, the channels will evolve and change. Your roles will evolve and change. However, ultimately, marketing will still be a fundamental pillar of success.
As long as you start building that moat, that brand recognition and that brand advocacy now.
