Don't worry about it, it's just SEO...isn't it?

Amelia (00:16)
Hi there, I'm Amelia.

Paul Wood (00:17)
I'm Paul.

Amelia (00:19)
And this is Fin the Week. How's everybody doing?

Paul Wood (00:22)
Really good. Glad it's Friday again.

Russell (00:24)
Yeah, sun's out. It's beautiful over here.

Paul Wood (00:27)
I get to rant about Google today, so I always enjoy that.

Amelia (00:30)
Best day ever! Full house, which feels right, because this is one that is firmly a marketing episode. Today we're talking about some news from Google. Paul, fill us in.

Paul Wood (00:43)
In the last couple of weeks, Google has released a guide. They've called it "Optimizing Your Website for Generative AI Features on Google Search." Really rolls off the tongue, that one. On the face of it, that's quite an exciting thing. Generally, people see that as good news that Google is providing some kind of guidance straight from the horse's mouth, because everybody wants to appear in Google. So when Google says, "here's how you do it," that seems like a good thing. A large portion of the industry I've seen online have sort of embraced it, particularly those who have had an issue with the idea of — we'll get into the different acronyms, but AEO, GEO, all about ranking in AI-driven search.

A lot of people have argued that there are snake-oil salespeople who are pushing services that just don't do anything, and that actually this is just SEO rebranded. And Google's kind of saying that in this document. But on the other hand, when you more closely inspect the document, it's not quite as clear-cut as they would maybe want you to believe. So I think it's going to be important for us to dig into it and take a look at what they're saying, and why it might not be as clear as they would suggest.

Amelia (02:08)
We're going to do this in three parts today. Firstly, we're going to explain what the actual guidance says in plain English, hopefully. Two, we're going to explore how helpful this guidance actually is. And then three, work out what FS content teams should actually do with it. So let's start with the document itself. What does it say?

Paul Wood (02:31)
It goes through a few different sections and I'll give a rundown of what they are. But I think it's important to start with a bit of context, because Google as a search engine is still far and away the most widely used search tool going. But when you look at the statistics for AI chatbot usage, such as ChatGPT, it's a less clear picture. ChatGPT, according to StatCounter, has 77% market share at the moment for AI chatbots. Google's Gemini has 9%. Now it's worth saying that ChatGPT last year was at 84%, so it's going down. Gemini was at 2%, it's now at 9, so it's going up. So this isn't a finished battle, but Google doesn't rule the market in this space.

Paul Wood (03:29)
It's worth bearing that in mind, because it changes the way you look at this document — you wonder whether they're just trying to position themselves as still the leader in this space. But anyway, the document goes into, first of all, an introduction. It addresses the fact that users' preferences are evolving. People are increasingly using AI-driven search tools to find the information they need. It says that for website owners, owners are looking for the official best practices for how to succeed with these AI-driven search tools. And then it starts out by saying that SEO is still relevant.

There's going to be a lot of potential jargon here, so it's probably worth quickly clarifying. SEO is search engine optimisation. I would imagine most of our listeners know about that, know roughly what it is. But then you've got GEO and AEO, so Generative Engine Optimisation, Answer Engine Optimisation. Pat, from your perspective as an AI expert, is there any difference between those two terms, or are they just one and the same, just with different names?

Pat (04:43)
SEO and GEO?

Paul Wood (04:45)
GEO and AEO.

Pat (04:46)
Okay, no, I mean, we were sort of on this journey quite early and we coined the term AIO, AI Optimisation. And since then, the industry has settled on GEO, Generative Engine Optimisation. And AEO is also floating around as well, but they're all exactly the same thing, which is just a term to describe the process of managing your online presence so that you appear in the answers provided by these chat tools.

I wouldn't say LLMs, I'd say chat tools, because the LLM is like the base training, and then all of these chat tools have a RAG — Retrieval Augmented Generation — data layer across the top. If you ask a question, the AI tool will query its internal knowledge base, the data that it's been trained on, and then will carry out a web search to augment that knowledge base answer with up-to-date information, and then it will present a combined answer based on both of those.

Paul Wood (05:49)
That was going to be one of the other questions: we throw around the term LLM quite a lot. What's the fundamental definition of a large language model? Is it that base training that an AI tool uses?

Pat (06:03)
Yeah, an AI model is simply a kind of database of weights with an input and an output. You put an input in one end — it's a configuration of artificial neurons in a kind of neural network — and these neurons will fire in a certain order and put an answer out the other end.

I use the analogy of a set of dominoes quite often. Imagine you've got a set of 100 dominoes in 10 rows, and you've got 10 inputs on one side and 10 outputs on the other. Each row has a number from one to 10. So if you put the input of one and three in one end, the dominoes will tumble down in a straight line and you'll get one and three out the other end. That would be your starting model. Say you want to set up a model that will reverse what you put in, so if you put one and three in, you'll get three and one out. There's probably a configuration of dominoes that would allow you to do that, but you don't know what that is. So what you do is you just tweak one domino and test it. If it gets closer to the end result, you keep that change. If it goes further from the end result, you abandon that change.

You keep doing that thousands of times until you've got a model that does what you want it to do, even if you don't understand how the configuration of dominoes does it — you've found it by a sort of pseudo-random approach. That's only got a hundred parameters. If you scale that up to hundreds of billions, that's what a large language model is. They've just trained it to be able to interpret text coming in and then spit text out the other side.

So that's the core base model. Your base models will be trained once every six months or a year and then that'll be fixed, so you can't modify them. So in the context of improving your performance in a base model, it's hard to benchmark and monitor improvement, because you might only see a change once every six months or a year when you go from GPT 5.5 to 5.6 or whatever.

Paul Wood (08:25)
That sort of leads on to the other things they mention at the start, which is that the document suggests this new way of searching is basically the same as what SEO was, except it relies on AI techniques. They mention RAG, retrieval augmented generation, and query fan-out. My basic understanding of both of those is that they'll have a base model — so if you search for something that's a known fact, like what is the size of this country in square miles, it will just use its base knowledge and give you that answer. But if it's something more like "what is the score for this football match that's going on," it will need to rely on search, retrieval augmented generation, to get more current information.

And query fan-out is the concept that if you search for something, it will carry out other searches on your behalf that relate to what you searched for. So there might be 10 additional things, so they can draw together lots of pieces of information. Have I understood that correctly?

Pat (09:49)
Yeah, it's pretty much bang on. And that's what makes the current AI tools so much more powerful than they were a few years ago. When ChatGPT was first released, there was a cutoff of like one year before it was released. People just started working around that — you could ask it about all sorts of stuff that happened in the last six months and it would have no idea. And then they started weaving in search results, and suddenly it became really accurate and really up to date.

It's really interesting because each different chat tool uses a different search engine. Gemini obviously leans on Google, so if you want to rank in Gemini, then you need good SEO on Google. But actually ChatGPT leans on Bing, and that's the leader. So suddenly Bing's a lot more important than traditionally it has been. And then Claude, which is just huge in the business space, I think leans on Brave, which is a really niche search engine.

Paul Wood (10:43)
Yeah, and you can imagine how those different tools suit different audiences. Your average person who's experimenting with this stuff probably goes to ChatGPT. People who are perhaps like us, who use Google's tools internally, might use Gemini day to day. And then you've got companies again like us who in a professional context use Claude to do some of the work we do, and probably then use Claude to carry out searches when we need to as well. And then you've got things like Microsoft's Copilot, which is probably used by a lot of professional services firms. So people in those companies might just by default be using Copilot, which is obviously backed by Bing.

There's a lot of technicalities around the start of the guide, but it's essentially arguing that ranking in AI-driven search is the same as ranking in search, because they're saying that the large language model is using retrieval augmented generation to go to Google search and bring back results. Therefore, if you're ranking in Google, you'll do well in their AI tools.

It then gives some SEO best practices. These are quite vague, but: create valuable non-commodity content for your audience, and build and maintain a clear technical structure. They're both good bits of advice that have existed forever in this digital marketing world, but they're also easily said. First of all, create valuable non-commodity content — that's fine if you're producing content for a professional services website. You would instinctively know you stand more chance of success if you produce a solid bit of opinion and research from a very respected person in your organisation, as opposed to writing a guide to "what is XYZ basic thing" that just draws on information already out there in the world.

But if you look at it from the perspective of, say, a publisher like The Economist, which is a high-quality publication, Google's saying, "give us all this good stuff and you'll rank well." But then the argument would be, "no, you're just going to repurpose it." So no one's going to come to The Economist to read this expertise; they're going to read it directly on your platform and they'll potentially never know where it came from. So it's easily said, less easily backed up.

Paul Wood (13:50)
And the other thing, about maintaining clear technical structure: in the same breath, the document mentions semantic HTML, but says don't worry too much about it because we'll figure it out. So they're saying: use properly structured code, but if you don't use properly structured code, it's OK because we'll work it out anyway. So there's mixed signals there. But yeah, Russ, you're probably going to say something about…

Amelia (14:11)
Nice and clear.

Paul Wood (14:13)
…the content and user experience, maybe.

Russell (14:15)
Yeah, you've got me thinking about how we're moving to a world where structure is less important. Up until now, the structure of content and the structure of a website has been one of the key factors in producing a website. And content strategy, I think, is going to change a lot. The way I see it, I think design and also content is going to become much more modular.

So instead of designing a fixed layout for a page or a website and being like, "this is the navigation, you're going to have an about page, you're going to have a services dropdown; on the services page you're going to have an image, you're going to have some copy underneath it; then you're going to try to cross-pollinate across different services" — I think we're going to be much more focused on just surfacing blocks of content, less probably thinking about the structure.

And then those modules, say like a piece of thought leadership — we can almost have each module depending on different things. Trust, especially in finance, we could have a module on the website that needs to explain whether you're eligible for an account, like a savings account or a mortgage, and that provides information on whether you're eligible or not. Then we'll have another block for a thought leadership article, maybe related to that kind of product. It can all tie together. And then those modules will be served by search engines, whether that's AI. And then when it comes to generative UI…

Russell (16:09)
…the AI will then decide which modules and which blocks to show depending on that user as well. And it's going to be a much more fluid experience in terms of the end user. We'll have to adapt quite quickly in terms of our processes to start working in that way. Start thinking less about fixed structure, start thinking more about what modules this website needs in order to rank well in these chat tools, rank well on AI search. It's really interesting, I think, as well from a design perspective.

From a design perspective — obviously this is more in search, but the design system will be much more important, because the design system doesn't look at fixed-layout pages. It just looks at how you style any components and patterns of components, regardless of how they're laid out on the page. So then the design system becomes essentially the most important and arguably the only thing that you need to do really, really well.

And then from a search perspective, as well as which modules are going to get surfaced when and where, one thing I'm hearing from clients, even in the last week, is: how do they track performance? I imagine Google Analytics is probably going to change quite a bit over the next year or two, because we can't really be tracking clicks anymore in the traditional way. It might be interesting to hear your thoughts on that, Paul: in this kind of where we're going with it all, how do we show our clients, for example, that their digital presence is improving with zero-click…?

Paul Wood (18:06)
Yeah, on the zero-click thing, that's one of the flaws with this whole document that Google has published, in that it has a sort of "trust us" vibe about it. They're saying, "don't worry about it, we've got these cool AI features, but it's still Google, it's still SEO, just do what you do and you'll rank and the audience will love it, it'll be great."

But in actual fact — we'll probably dig into the flaws in a minute — I did it the other day. I searched for something using Google, and it was something really specific, like "who are the top companies in this space?" And the AI-generated answer was meaningfully different to the list of results that sat below it. I get why that would be, because the AI-driven element has probably used query fan-out to go and look at much more nuanced things, and it's bringing back different answers based on that. But for the company at the other end who are trying to optimise what they do, Google is the largest source of traffic for them, so this matters. So yeah, it's worth thinking about.

Pat (19:36)
Is it that with large language models augmented with search, and their ability to find out loads of such queries all at once and then collect the information together, does that mean that actually what we're seeing here is that best-practice SEO rules are even more important than they were traditionally? Because traditionally you fire out a single search, Google just queries its index and returns a result.

But because that's a more simplistic mechanism, there are opportunities to game it and tweak it through clever technical SEO workarounds. It's heavily reliant on backlinks to this day. And as such, you can have the best site ever with the best content and following all SEO guidance, but a competitor could have a similar site that, because they've done some SEO hacks, ranks better than you. Whereas in a world of large language models, the large language model can see through those hacks and actually prioritises really well-made websites even better than — yeah.

Paul Wood (20:49)
Yeah, I think that's definitely true. Good AI optimisation, whatever you want to call it, is fundamentally good SEO. But then I would always have argued that good SEO is usually good business. Being a good business that has a niche, has an audience, does everything that audience wants, just operates properly, professionally. And then there's a few technical things — if you've got a website, you need to make sure it's accessible and follows certain rules and isn't a mess behind the scenes. But as long as you tick all of those boxes, being a good business — fundamentally any search tool is trying to surface the best business or the best answer, whatever it happens to be, for the person who's searching.

But the challenge that's shifting now is that we spoke about it in the Brighton SEO episode, in that Google has been sitting on this line between being a search tool and becoming a media presence in its own right. They've done that for years — if you search for the weather where you are, you no longer need to go to the source, you just get the weather on Google. And now Google is sort of overreaching potentially by taking the content of the companies who are optimising for their search engine and surfacing it directly. So it's not quite as clean-cut as it used to be.

But it's probably worth going into the other big section of the document, the one that's probably caused the most amount of chat online: what they call the myth-busting section, where they say stuff for you not to worry about. They call out what they call AEO or GEO hacks that you read online.

They say you don't need an LLMs.txt file. Without getting stuck in the weeds of what that is, it's essentially — if you run a website, you might be familiar with a robots.txt file, which lives on your server. When a search engine bot comes to your site, it just tells that bot, "here's where the proper content lives, here's where all the technical files live that you don't need to worry about." It guides a bot around the site. People are pushing this idea of an LLMs.txt file, which guides a large language model around your site. Google is saying, for their search engine, you don't need that. But that does contradict other stuff they do — their Lighthouse tool for auditing sites does actually check for an LLMs.txt file. So for agents, it's definitely important, I would say.

You don't need to produce a markdown version of your website. I think we touched upon this in a previous episode — the idea of having a public site and then a simple text version of your site. Google's saying, "we can see everything, don't worry about doing that." But again, from an agent and agentic AI perspective, that's possibly not the case.

You don't need to chunk content. There is a world online where people say you need to chunk content into little mini chunks that AI bots can just pluck out and use. That's probably fair to say is not necessary, because search for years has been good at understanding nuance in large bits of content.

You don't need to write content in a specific way. Another big one is inauthentic mentions. There's this concept of people now optimising to get mentions of their brand and their company seeded in different places, say like forums or on Reddit and things like that. So that when these LLMs do their training and they hoover up data from something like Reddit, the idea is that you could game the system by being the brand that's mentioned all the time in certain threads. And Google is saying don't do that, because that's gaming the system. It strikes me that inauthentic mentions, as they term it, is now the new link building. Years ago, Google cracked down on spammy link building, people who were paying to buy links on other websites. And they seem to be suggesting now that inauthentic mentions is that, because people are starting to game that system. But they're only gaming that system because it actually works — it's a powerful way to do it. So it's not infallible.

And then the final one is you don't need to over-focus on structured data. Again, without going too technical with it, Schema.org is an organisation that essentially provides a framework for tagging bits of information on a website. So you can tag, "this is the street address, this is our postal code, this is our phone number," and you can go much further with it. Cinemas use it for cinema showtimes. And this is another frustrating bit with Google's document — it says, "it's a good idea to continue using it because it's just good, but we don't use it for AI search, but do it because it's good." It's just not clear.

That's a problem because there's a lot of people online who are saying you have to do schema markup, you have to do structured data, and then there are people saying don't do it, it's a waste of time, AI doesn't use it — and there's this messy bit in the middle. It's difficult for an agency like us, for instance — we generally advise clients, "yeah, do it where it makes sense, because it's clean code, it's well structured." If I were trying to figure out a system, I would want it to be structured in the way that schema encourages you to. But at the same time, we don't go overboard, because we know you could get caught out, because Google would come along and say, "all that time you spent building schema markup is a waste of time." So there's that myth-busting section, and basically half of the industry have hooked onto it and said, "see, we told you it's just SEO," and the other half have said, "but actually it's not as clean-cut as they say."

Russell (27:27)
Can you see Google search still being around in a few years? Or do you think they're kind of covering themselves a bit by saying, "you don't really need it, but you should have it because Google search is still there"? If they tell everyone the schema isn't required anymore, then this transition between Google search being phased out, which is probably likely long-term, and being replaced by AI search and more generative kind of search — it sounds like they might just be saying…

Paul Wood (27:40)
Yeah, they're hedging.

Russell (28:02)
It isn't needed for AI search, but you still need it for Google search. We're kind of saying you don't need it, but you still need it at the moment, perhaps.

Paul Wood (28:10)
Yeah, that's one of my main gripes with the whole thing — and you can't really blame Google for this, but it's written in a vacuum. It's basically saying this is specific for Google search, but it's ignorant of the fact that, well, there's also Gemini, which is a standalone tool, which they probably wouldn't class as Google search. There's also all of the other AI tools out there that aren't anything to do with Google. So fair enough, they don't need to address that, but they're ignoring the wider context.

I think Google search — and this is something that again, in the Brighton SEO discussion we had Dr. Pete from Moz, who does a lot of research in this space — his view is that it will become a kind of hybrid search system where Google will essentially form whatever front end you need in the moment. So it will be a hybrid of an AI chat and search tool, and I think that's probably what will happen.

You'd be brave to bet against Google in eventually winning this battle, because they've done it before. The thing I keep thinking of is browsers. Internet Explorer and other browsers before were way ahead of Google. Firefox was probably ahead, but Google came along with Chrome and they just persevered with it until it dominated the market. And I think they'll do the same with search.

But this document, the final thing on this section is just to say that they gloss over agentic AI as well. They literally say at the end of it, "if this is something that's relevant to your business and you have the extra time, check out available agentic experiences and review our guide to that." But agentic is arguably the most important part of all of this, because that's really the direction of travel. And that ties into what you were talking about, Russ, with this idea of modular things where brands could perhaps surface modules that fit into an agentic experience.

Russell (30:09)
Yeah, and I think the most exciting side of it for the consumer is personalisation. If you have access needs, then when you search for something, whether that's in the traditional Google search or this new hybrid search, you'll get results which flag — if you're looking to go to a restaurant, it would show restaurants which have good accessibility, for example. If you're looking for clothes, it would show clothes in the style that suits you. And if you're looking for something sports-related, it would show articles related to the sports that you're passionate about. I think that's where search is going, and that's really interesting, and that's what AI gives us — that level of personalisation, which in the past was quite a static configuration that you'd have to do in a series of fields in your profile. Whereas now it's picking that up from lots of different sources and figuring it out. So that's really exciting for search. I think it's going to give people a much better experience.

Amelia (31:26)
Shall we move on to talk about how helpful this guidance is? And of course we've touched upon this, but ultimately can we trust it?

Paul Wood (31:34)
Well, I'm usually quite cynical with this stuff, but I don't think we can really trust it, because Google really does have a vested interest in everybody sort of lapping up the advice. The example I gave before — if you read the advice as a business and you think, "well, why wouldn't I follow that?" Because yeah, produce great content, have a technically sound website, and hopefully then Google's search systems and its AI systems will surface us as a source of information. And I have to get comfortable with the fact that I might not get the traffic for that, but hopefully they'll mention my company when they give an answer to something. So that's maybe the new world we're in now — you don't get the traffic but you get the mention, fine, you have to accept that.

But the problem I have is, if I was working for The Economist, they're basically saying, "keep your website open, don't put it behind a paywall, produce this amazing content and we'll find it and we'll surface it." And it's like, well, there's a real dilemma there, because you think, "why would we want to give you all of our assets for free, basically, because you're not going to send that traffic to us?" And they don't benefit from them saying "The Economist gave us this answer," because by that time the person's got their answer and off they go. So I think it overlooks that.

The other big issue with it is that, like I said before, it's written in a vacuum. It doesn't really look at agentic AI. It doesn't look at the wider context of AI chatbots versus Google search. It doesn't address the fact that traffic isn't flowing through to content owners' websites anymore. And that's something we're seeing day in, day out. We produce reports every month for different clients, different companies of different sizes, and across the board they're all seeing quite wacky changes in performance. Even yesterday we were looking at a report where the company's clicks from Google search were really going off a cliff in a quite shocking way. And the availability of search ranking positions for general informational things was just disappearing. And that's almost certainly because AI overviews — Google's AI overviews — is just hoovering up all of that user demand. On the flip side, the company was getting far more brand visibility, with people searching for their brand more frequently, which is arguably a good thing. But this document doesn't really address that massive shift in the way things are working at all. So yeah, I struggle to trust it fully. I think it's a really basic guide — it's a simple answer to a complex question, this guide.

Pat (34:49)
It feels like a bit of a wishy-washy guide for optimising your site for Google search as it currently is right now, which is still fundamentally Google search with AI overviews at the top. But actually, personally, I just don't use Google. The only thing I use Google for is if I'm looking for a specific website. I'll just go into the Firefox browser bar and I'll just type in "Merlin Cycles" or "Camilla's Cafe," and that's literally what I use it for, brand searches. As soon as it comes to product research or advice or medical advice, product advice, solving problems, troubleshooting — straight into Claude. And if it's a simple question I'll use Claude Chat.

Pat (35:37)
If it's a complicated question, I'll use Claude Cowork. Cowork is an agentic AI which will go off and explore in great depth many different avenues. It will browse websites itself. It will obviously use Brave. It doesn't even touch Google. And that paradigm of usage is completely different to search.

Paul Wood (35:57)
It's probably worth mentioning that a while ago, Microsoft released a similar guide in relation to how they surface AI search results. I found it much more interesting — not because it gave any major takeaways, but because it focused on just telling you how the system works. It explained, to some extent, the difference between — and actually one of the things I made a note about here is the difference between AI training and AI grounding. So the thing we were talking about before, where you've got the base LLM model that sits at the bottom and can answer factual things, and then you've got the retrieval augmented generation that goes and does the search for you and fills in the more up-to-date information. The Microsoft document focused on making sure that the reader got a sense of how this system works, so that you can then go away and think about it and think, "okay, I can imagine how I would make myself useful for that system." This Google one on the other hand doesn't address that at all. Instead it just — you could probably boil it down to just saying, "ignore all the hype, it's just SEO, do good content and you'll be fine." And I argue you probably won't be fine. It's going to change and they're not admitting it.

Pat (37:12)
It gives the impression that the Google Gemini chat is just purely the grounding. All it does is carry out searches and then summarise those searches into an answer, and ignores all of the fundamental knowledge-based training that that large language model has. Because that's the difference here — the large language model works very differently to a search index. So you've got these two sources of information. If you want to rank well, the large language model needs to be aware of your brand and what you do, so that it can talk about who you are. And then it will augment that with the search results. And if Gemini is just ignoring that base-model knowledge and just compiling search results, then yeah, this absolutely makes sense. But it's not doing that. It's using the core search model as well, and getting yourself into that model's training data — that fundamentally is a summary of all the information that exists about your business online. But indexed in a very different way.

Paul Wood (38:32)
Yeah, I actually made a note about that somewhere that was essentially saying that there's examples where you could think — coming back to an example like The Economist or the FT or any publisher — they might think, "I want to be available to the bots that come to my site to surface my pages in a search result, but I don't want to be available to an AI training bot that comes to my site to hoover up information to add to its core model." And that's a genuine thing that needs to be solved. This would be a great place to address that, a document like this, but they don't — because that's a fundamental, that's different to SEO. That's not SEO. That's figuring out how we make sure that our proprietary content doesn't get basically taken by a training model, but we keep our website open so that a search bot that surfaces things for real people can find us and can use that information. And there isn't an answer to that, and Google hasn't made any efforts to try and answer that, as far as I can see. Not in this document anyway.

Russell (39:51)
Yeah, it sounds like there's going to be a new structure at some point around how this gets handled, to give some protection to the companies providing this information as well. It should really be a choice of the business where that information is given to and how it's stored. You mentioned the robots.txt file, and the document is talking about these LLMs.txt files, but perhaps it lives in some kind of special markdown where you can choose what is stored where in the future, just so you have some control over your own information.

Amelia (40:39)
So ultimately, is this advice at all helpful?

Paul Wood (40:42)
I think the fundamental advice is: produce good content that isn't just the same old rubbish you could read anywhere. Make sure it actually comes from you, you're not just writing for the sake of it. You're writing something that adds to the conversation, and have a technically sound structure behind it. But that's just good advice — that's always been good advice, and that's good advice not just for search but as a business. So yeah, that's fine, but it doesn't add anything new that people needed to know, I don't think. And if anything, it just stirs the pot a bit, because there are a lot of people online already arguing about: is GEO a thing, or is it just SEO? And I don't think this really helps that conversation, because it glosses over so much.

Amelia (41:33)
So we've talked about the guidance, what it says and how helpful it is. Now let's talk about what FS content teams should actually be doing now, moving forward, based on this.

Paul Wood (41:42)
Starting with the content point, this has been the case forever really, but Google — any tool really — wants a unique point of view from content. So there's no point in writing generic guides anymore. The example I wrote down was "seven tips for first-time buyers." That's a list article. It just provides answers that are well written online by other sites already, and provides no new things. So if you're going to put effort into writing content, your best bet is to use genuine expertise that exists within your organisation, and if you've got the bandwidth, do your own research. That type of content will do better — not for any technical reason or for a trick — it makes sense as a person. I want to read something that comes from a real piece of research. I don't want another cobbled-together guide. And that's what this is saying. I suppose it's just saying that approach is more important than ever, because you could argue AI-driven search is going to be better at surfacing the genuine best answer than search has ever been before. So being the best answer is more important than it's ever been before.

Russell (43:13)
Yeah. And also, from the perspective of who your products are for, especially in terms of Consumer Duty — you should make it clear, say if it's a business account, what size business, perhaps where those businesses can be located, what types of businesses. Having that structure of information about clearly explaining the benefits of each product, whether that's savings products and mortgage products or investment products, who it's for, what it does and the benefits, and the criteria for being able to purchase one of these products, is even more important now, because that information is going to be in search and it's going to be highly personalised. So the websites that are doing that right are not only going to have a better experience on the website, like has been quite obvious for years, but also they're going to have a much higher chance of being surfaced by AI. So it's even more important to clearly define who your products are for and the benefits, and have everything in place, I think.

Paul Wood (44:29)
And another thing is, there's a compliance challenge for financial services teams for a couple of reasons. First of all, there's the workflow that potentially happens inside some organisations where the typical journey to creating a piece of content would be that the subject matter expert drafts it, marketing then edits it, legal reviews it, compliance reviews it. By the time it's gone through all of that, it's potentially stripped out a lot of the character and voice that it started with. There's an argument to say that maintaining as much of that personal — written by a real person with real expertise — element is going to be important. I don't think that's a major hurdle to get over. I think it's just a case of companies being more deliberate now about saying, "we're not just going to write a guide to this thing we know really well, we're going to make sure it's backed by our senior partner who knows this thing inside out, and it will have genuine quotes, and we'll back it by a bit of our own research, and make sure that when it goes through legal and compliance, none of that important stuff is stripped out." Making sure there's a balance between sticking your neck out a bit with something new, but maintaining that professional sheen that it needs.

There's also another compliance challenge in that when you publish something online, you have to be comfortable with the fact that certain bits of that content might be clipped and reused in an AI-generated answer. Which means that if you're making a claim about something that would normally be covered at the end of your article saying, "this was written on this date by this person in this context and shouldn't be used outside of this context, it's not advice, blah blah blah," you almost have to start adding your disclaimers in line with claims. Which is a challenge, because you want to make sure that if it does get clipped and used elsewhere, you don't want Google's Gemini saying "XYZ firm said this piece of advice" but they miss out the disclaimer, because that's an issue. So yeah, you almost need to wrap your disclaimers into whatever you're saying. That's a challenge, and I'm not sure how you overcome that.

And then the final bit of advice I wrote down was that you need to surface the people and surface the working. So you need to write in the voice of the people who are the experts. We've done this ourselves recently, actually, as an agency. We've completely shifted the way our website is structured, mainly to surface the people behind it. So every claim we make is usually attached to a person, and we're trying to — because that's how it works. We're speaking in our own voices and we want to present it. And I think firms are going to need to do that as well. They're going to need to make sure if they write a guide to something, it has to be attributed, otherwise these systems are potentially going to overlook…

Amelia (47:54)
So Pat, Russ, have you got anything else to add when it comes to what teams should be doing now based on this?

Russell (47:59)
Well, researchers — especially user research — is something I'm quite passionate about. And I think that point you made, Paul, about doing your own research, to make you stand out and give your… it's so important to a thought process now, and that's where you add a lot of value, I think, in terms of producing something. It's your own unique perspective in that gathering of research for that final output, I think, which makes you different from every other organisation.

Especially now when scraping the web and compiling all the research that everyone has done before is so easy with AI, there's going to be a lot of end results that are quite similar or a little bit generic. So doubling down on speaking to your audience and testing your websites, getting feedback — I think will really make your work, whether that's written work or visual work, really stand out moving forwards. I think that's really key, and hearing you explain it there, that really stood out for me.

Pat (49:15)
I think going back to a previous comment you made, Paul, which is around just running a good business. With AI, these tools will just become incredibly good at matching the right business to the right customer. So as a business, you need to be providing good services at good value. You need to be focused, you need to find the niche, you need to get all your marketing nailed and just generally run a really tight ship. Understand who your target customers are, like you were saying, Russ — because if you understand who your target clients are and you're marketing really well to them and you're providing great services to them, then these tools will just find you. There'll be no way to fake it. So really, it just comes down to just being a good business, I think. If you do everything in a really solid way, then these tools will find you. There'll be fewer shortcuts to driving traffic. You won't be able to fake it as much. That's my view of where it's all going.

Russell (50:24)
You've just reminded me, Pat — I've been on the web probably since like 97, 98 or something. I remember back in the day, I used to get something called a link farm, which is literally just a page of links. Do they still exist? I don't think they will for much longer.

Pat (50:40)
They exist, believe it or not. And they still work. They're just much more sophisticated than they used to be.

Paul Wood (50:50)
There'll be citation farms now.

Amelia (50:53)
So, before we go, shall we have a quick game of Jargon Busters?

Paul Wood (50:57)
Yeah, yeah, lighten the tone.

Amelia (50:59)
So if you've not caught this before, I've got a list of these industry terms, and I'm going to be putting the guys to the test to see if they know what today's word is. So today's word is: churning. What do we reckon?

Pat (51:13)
There's obviously customer churn, isn't there? You're churning customers because you're doing something wrong, or there's some economic thing that's happened that's resulted in customers just leaving you in droves. That's kind of what I'd say.

Paul Wood (51:28)
Yeah, in a finance context, I don't actually know for sure. If I had to guess, I'd say it's something to do with investments, perhaps. Is it an element of your portfolio that you write off for one reason or another? Like you expect every year to churn a certain amount of your portfolio, so it just disappears.

Russell (51:59)
Yeah, in an FS context, I guess, churning means when you're just going through something that's not being replaced normally. So yeah, churning through — maybe just churning through money, maybe. I'm not sure.

Amelia (52:21)
So churning means excessive trading by a broker in a client's account to generate commissions.

Paul Wood (52:27)
Okay.

Pat (52:29)
That's probably the first one that none of us have nailed, right? Yeah.

Amelia (52:34)
Yeah, to be fair.

Russell (52:36)
Yeah, so actually it isn't churning without anything in return, it's churning and generating commission. Interesting.

Amelia (52:43)
Yeah, exactly, exactly. So we will obviously have more of this next week in our next episode. We will be back next Friday. Thank you for listening, and we'll see you next week.