AI Search Is Full of Spam and It's Getting Worse! | The future of SEO with Mark Williams-Cook

Hello and welcome to this week's episode of Search with Candid. I am one of your hosts, Jack Chambers-Ward. We're back in the studio with Mark Williams-Cook.

Hello.

How are you, Mark?

Very, very well. Enjoying the British summer.

I see you've got the shorts on, ready for summer. I have, of course, traditional summer outfit of head to toe in black.

I did think today, I was like, it's a bit cool down. I was like, I wonder if Jack will be in shorts today.

I did think about it, but then it's supposed to rain later on.

So I was like, ah, nah.

Welcome to British summer, right?

You've got to wear the shorts, scare the rain away.

Well, this isn't the British summer weather podcast, unfortunately, as much as some people maybe wish it was. We haven't talked about anything, you and I, since Google I/O.

Wow.

We haven't done an episode in almost 8 weeks together.

Why not? You've been slacking.

Well, you've been flying around doing conferences and stuff and busy things like that.

And yeah, it did occur to me the other week. I was like, I haven't spoken to Well, I should probably do that.

It's funny, people will message me sometimes and be like, I can't believe what Mark said. I'm like, right, he's in a different country. I have no idea what he said. I haven't seen Mark in like 2 weeks. I don't know what you're talking about.

What did I say?

I don't know. Who knows with you? They assume we just like hang out all the time like this. This is our just workday of just chatting about SEO stuff.

That'd be nice, wouldn't it?

Yeah. I mean, sometimes it is.

We should make more time for each other.

We are literally like desk buddies.

You're not in as much now though.

I know. I moved out of the city. Yeah.

WFH.

I try to be in sort of like 2 or 3 times a week.

I appreciate it.

Nice bit of hybrid.

Yeah.

Anyway, we're not talking about that either. We're not talking about work habits or anything like that. We're going to get into some broader kind of spammy LLM discussions. I know it's been a hot topic recently. There's been talk about people manipulating LLM search, shall we say, with a lot of expired domains, kind of parasite SEO type stuff. There's been a lot of stuff going on with basically general fluff and crap that people are using LLMs for. And we're seeing a lot in LLM search and how that kind of relates to what we've seen from traditional search side of things, how we think the browser wars, as I guess we're going to call it. We've got OpenAI obviously have announced their browser. Comet is now available to Perplexity Pro subscribers. Whatever is going on with Chrome, with Google, We're in a fairly tumultuous era of search right now, it feels like, right?

Exactly.

This week's episode is sponsored by Fat Joe. Since 2012, Fat Joe have been the world's highest rated provider of outsourced digital marketing services, and their productized services help SEOs, marketing teams, and agencies not only scale but thrive as well. And they have a workspace designed for outsourcing efficiency Basically, it's the no-hassle way to get your SEO work done. Fat Joe offers over 20 different services ranging from blogger outreach and digital PR campaigns to content writing and even design and video as well. So if you need the hassle taken out of a time-consuming task like blogger outreach, then Fat Joe might be the perfect solution for you. Plus, it's 100% manual outreach. That's no PBNs, by the way, and there are no hidden fees. What you pay for is exactly what you get, and there are no extra charges. Plus, it's all 100% white label. There are no contracts, no minimum spend, no tie-ins. You are fully in charge of what you pay for and the services you use. But don't take my word for it. Fat Joe is rated 4.8 out of 5 stars on Trustpilot and has over 1,500 reviews on Shopper Approved. Plus, they give you 100% satisfaction and money-back guarantee on your first order. If that doesn't say confidence in your product and service from Fat Joe, I don't know what does. So if you're ready to get started, please do sign up using the link down below and create your free account today. Thank you so much to Fat Joe for sponsoring us here at Search with Kander. Now, Back to the episode. There's a lot to talk about. And I know basically you just said, I said to you the other day, like, oh, what should we talk about on the podcast? You're like, spam from LLMs. Like, cool. That's the internet at the moment, Mark. You want to narrow that down slightly?

Yeah, I don't. I would like to talk about it because I don't think enough people are thinking like moves ahead in terms of what's going to happen. And this is the— I think I have an answer for it now, but the most interesting question for me has been, Long-term, I can understand and see people moving to AI search, right? Because it's easier for a lot of things, right? It's less friction to find a company, to compare a thing, to do like a search with multiple parameters. Because with traditional search, you can kind of only search for one thing at once. Whereas with AI search, you could be like, I want to find this and it needs to do this and then rank this and compare this. And it goes off and does all that for you, right? Great. And the thing that I've been thinking about is we know, and it's not discussed a lot, that LLMs have been buying up link graph data from the big SaaS companies, right? We know from the DOJ documents that Gemini uses some traditional Google ranking scores when it comes to weighting their training data so they pay attention more to sites that they know are trustworthy, not spammy, are authoritative, etc., right? And LLMs have tried to get in on that by buying up this link graph data to help weight their initial training, okay? And what really got me thinking was this. So that works at the moment, for now, it's pretty good. And the assumption is, and my assumption has always been, well, technology gets better, things improve, so it's as bad as it will ever be. My worry here is that if people do start using, say, AI search for, um, for say 50% or more of general searches, right, fine. And we've already seen people somewhat surprised at the fact that the traffic they get from LLMs, they're like, oh, it converts higher than search traffic. And it's like, well, Yeah, of course it does, because the whole research consideration phase is done inside there instead.

Yeah.

And it's like, fine. But here was my dilemma. Okay. If this is all rooted in these traditional search metrics or this link graph, in a way, the link graph is a proxy for user behavior on the web in that people like these websites, so they link to them. Right? The same is true of click data in the SERPs, which we know is really important, right? They— people choose this website, they choose this brand, they search for this brand. So people— so, um, people like Google rank those sites better. If people start moving to AI search and doing everything in there, it means that people aren't visiting places like Reddit as much, or forums, or other websites as much. They're staying in the LLM for the answer. They're not visiting them. We're not getting the click data from the SERP showing the preference and how long they stayed there and if they came back. And then people obviously won't be linking to them. They'll probably be sharing links as I've done before to say like a Perplexity summary. I've shared that. So that whole set of data that's worked pretty well for the last 20+ years is going to rot. It's, it's going to kind of reduce in its usefulness. And then we think about how LLMs are working in terms of their— how they're generating answers, right? And the way I think about it is traditional search is built on this link graph, so how the web links together. And the rough idea is you've got topics that all link together that show where authorities on those topics are. You've got an expression of popularity, these sites are commonly linked to, etc. Whereas LLMs are using essentially a language graph, which is language translated into tokens, and then seeing how sequences of these tokens are connected together. And that's— you get super interesting results from that, right? Which is LLMs, and we can make all this conversational stuff. It's fascinating what we can do with it. But within that, there's no kind of importance spam protection rating for who is saying this, which is why they're leaning on the link graph stuff to, to be like, okay, well, we've got all this data, but how important is it versus this other data? And my question was like, well, what's going to happen in 5, 10 years when the link graph kind of dries up? And then you've already said we we're getting to a stage now where these LLMs have said, oh, we're bringing out our browser.

Yeah.

And then it's like, oh, okay, I see, I see what's happening here.

That's how you get the click data.

That's how you get some kind of usage data for, for what's happening here. But I mean, it's still not— I don't think it's going to be as easy as click data from somewhere like SERPs. So click data, user data, as Google engineers have told us before, is incredibly noisy, right? And the Google SERP traditionally has a lot of parity in that it's a simple— it was increasingly complex, but list of links essentially.

Yeah.

And it's, you know, Google knows what they're providing and then you can compare, okay, these users clicked on this, these users clicked on that with Highly personal, very specific prompts, um, and not as clear data because the, the feedback will be more conversational. It won't be, I'm not going to choose to click on this, because that's not what people in the AI, um, in the LLM and the AI search is doing. They're having a conversation. So how clear it will be on the, what was the quality of the answer we provided, I think it's gonna be a lot more murky.

Yeah.

And all of this is compounded by, as you said, um, I saw, uh, was it Netcraft? It's not Netcraft. There's a study, uh, I read this week which looked at just trying to use LLMs to surface the login pages of known brands.

Oh yeah.

So it was, I think, 50 brands they did, and they used things like ChatGPT, and they were just saying, oh, I've lost my bookmark for my Wells Fargo account. Can you give me the login page, and 33— I think it's about a third of the time, 33% of the time, it was giving a domain that was not owned or connected by the brand, that was either dead, it was either owned by someone else, or in the Wells Fargo case, it was a free Google site that was just a scam site as the number one result.

Yeah, I think we see a lot of that Like there's a great article by James Brockbank from Digitaloft, and whenever we featured it in Core Updates, talking about the expired domains and hacked sites, things that are being featured in LLMs. Obviously Google has talked so much about site reputation abuse and parasite SEO over the last 18 months. So that's been a big, big focus for them, talking about how much they're catching spam and we've removed millions of pages per year, more than ever, all this kind of stuff. And I think that ties well into the link graph stuff as well, because Google has decades of data, as you said, right? They've built this network of information of what is good, what is bad, what is trustworthy, what isn't trustworthy. And they literally just don't have that data from the LLM side of things, right? Like you said, they will buy link graph data and source it somewhere, but how quickly that becomes outdated and exactly as you said, as we move forward and forward in time, that then becomes less valuable in general. So cool, you've got a snapshot of what was trustworthy 6 months ago, and we know how fast the internet moves, let alone with the AI-powered stuff that's going on at the moment. So 6 months is an eternity at the moment. Like I said, we haven't spoken in 8 weeks in this format, and so much has happened in that time. You can imagine what happens in 6 months or 18 months or 2 years. And for me, I think there is a real kind of Yeah, it's an interesting thing to have the browser side of things in there as well. I know regular guest on the show, in front of the show, Miriam Yesia, talked a lot about clickstream data and why people aren't thinking about this browser thing in the way that it should be. We all kind of think about, oh, right, yes, to do with DOJ, it's to do with the antitrust stuff that Google is going through. They're going to probably have to diverge Chrome. That's been arguably the biggest conversation, the biggest potential consequence from a lot of this antitrust stuff that's been happening with Google. So you can introduce new competitors that opens up the market. It's like, yes, that is obviously a business case for it. But I think you're spot on. I think Miriam is spot on as well. When you think about when you can get firsthand data without having to purchase it from other places, without having to use outdated sources, you can then adjust things in your own system and build your own little ecosystem. And they've got a lot of ways to catch up with Google in many ways, obviously. But I think you replied to James on LinkedIn, and I know when I wrote up the headline in Core Updates, even you're like, actually, this still is happening in Google a lot as well. We're not saying Google is a perfect system, right? There are plenty of expired domains. I know you've been using examples of there's been casino things, there's been school websites, high authority things that are manipulated by gambling and kind of like semi-legal, questionable, spammy stuff as well. And you can really easily see how that gets out of hand and you end up with literal phishing sites being listed as login pages for big accounts, for big websites. That could be a pretty serious issue if people start losing money and people start losing account logins and legit getting hacked because of the information they provide. Then it comes around back to the legal side of things, like, who is responsible for that? If I get served a hacked login page and I log in and my account gets taken and it's my banking and they clear my account, are OpenAI responsible for that?

Because they have that little thing that says generative AI sometimes is wrong.

Exactly right. That seems to be the go-to thing. And I know they've also slightly off topic, but Even less so with AI overviews. And what we've seen, the introduction of the AI summaries that are barely even labeled in Google Discover as well, is literally it doesn't say AI overview. It has a little summary thing and then just a little, by the way, this is an AI-generated thing to worry about. It might be wrong. I feel like you should be more clear about that. We've had this huge— and it was a big thing in podcasting and YouTube a couple of years ago, the FCC in America really cracking down on Being visible with ads, using the joke of hashtag ad, hashtag sponsored, because you have to be very clear about when you're paid for a product. Like, we're not being paid by OpenAI and Perplexity or Google or whatever. We can just freely talk about whatever we want. Whereas plenty of influencers, plenty of podcasts, plenty of people were very clearly doing a, let's do a featured product discussion. Oh, you spent 10 minutes talking about this one product and how great it is. and then basically didn't feature anything else. And there was a lot of dodgy shit there when it came to that kind of undisclosed stuff. When it comes to that side of things, for the LLM side of things, how do you get out of that little, well, it might be wrong, tough. Well, then maybe you shouldn't be answering like YMYL questions then, if you're just going to give me wrong information. And I know this is a conversation we've had many times before, right? Of like, how does that balance with the liability, the responsibility of technology and things like that to, yeah, maybe you shouldn't be answering medical questions with AI-generated stuff. Might be wrong. That's life and death situations. That's money. That's livelihoods and stuff. Right.

Let me go back to what you mentioned about James and the AI search results. Because I think something that's maybe overlooked is, you know, I pointed out that at the moment, especially in iGaming, there is a problem in Google with people using expired domains and some other tricks that are working really well. But the way they're exploiting Google at the moment is like a very specific entry kind of method that is hard for Google to defend against because those metrics are helpful for genuine sites. The interesting thing I found with the study of the login pages is that it generated an obvious kind of scam, like Google free website, and that even if AI search have the data from browsers, I don't think they will have— you know, Google's got like 20 years experience trying to get rid of these spam sites.

Yeah.

And I don't think even with the data, I think it's a much harder job than people realize.

Yeah.

And there's going to be a lot of problems with this. And while we're talking about Miriam, actually, so this kind of blends into Miriam's been doing some really interesting work recently about what she calls quadrants, which is built on, if people haven't heard of it before, this Johari window, which is this basic knowledge theory of quadrants about, if we're having a conversation, one quadrant would be things we both know, and then things I know that you don't know, things you know that I don't know, and then things we both don't know.

Yeah.

And you work out where knowledge sits. determines how you can communicate and what you need to do. And she's been applying that really smartly in a way for companies in an AI era in terms of what AI models know about you. What occurred to me this week, and I actually had a very brief conversation with her about this, is where does the responsibility sit if This week I saw an example of someone who was making fake GitHub APIs, right? So they're going on GitHub, publishing loads of code for like this API for this company that didn't exist. And they wrote documentation for it. They wrote tutorials on how to use it. And it got hoovered up by an LLM. So when you ask it to help you with code, or can I, have they got an API? Can I do this? It's like, yeah, use this API. And it's basically a, um, just sends— it's like a crypto thing, sends money to their crypto wallet, I believe. It's just a scam, complete scam, right?

Yeah.

And I was like, that's like a next level scam, right? And it's particularly dangerous because it's being passed through a filter of an LLM that you are interacting with that you trust. So it's kind of making it safe. You're not seeing like the dodgy scam stuff, right? But my question here is this, that if this is happening to your brand or you're aware this could happen to your brand, i.e., people can do searches for you or adjacent to your brand in an LLM and it will surface something that might harm your customer or damage your reputation because it's like a fake API or something. Whose responsibility is it to monitor and mitigate that? Is it a security team thing? Because it's— is that a security issue in a way? It's like a phishing thing, but it's not against your own systems, right? Is it an SEO's responsibility? Because it's a search surface and part of our job is to help control, um, what does influence visibility in those platforms.

Yeah.

So should we be monitoring it? Do we report it? Like, Nobody knows at the minute because it's new, right? And I think that's going to get way worse. And I think phishing through LLMs is going to become a big thing. Because if you're just pumping out this false information, it's that thing of, you know, how fast a lie travels around the world before— what is it? Before the truth's got its shoes on.

A lie travels twice around the world before the truth even leaves.

Yeah, he's got his shoes on or something like that, right?

Yeah, exactly.

But the speed at which you can fabricate fake data that looks believable now. So API, I can, you know, make my fake API and then I can just generate documentation for it. I can generate tutorials for it much quicker than you can, you know, make actual real information.

Yeah, yeah.

And when you think again about how LLMs are hoovering up training data, volume is important, how many times it's been mentioned. I can spam Reddit and Quora and link to it and talk about it and make it look very believable to an LLM, and it's going to get surfaced.

Yeah.

I think that's going to be a huge problem because there is— it's very hard to quash. The advantage as well for the spammer is You don't— if you're running a scam website, you need to protect— you need to get that spam website visible, right, in a search engine, and you need to keep it visible for your scam to work. You don't need that with an LLM. You just publish on other websites, you publish lots of different domains, and it all gets hoovered up. And even if your site gets shut down, it's probably still going to sit in that training data. And once it's in the training data, It's very hard to, you know, it's like ink in water, right? It's very hard to take that out. So that was just an interesting question I had in terms of, you know, when we're working with especially bigger companies, which will be the targets of these kinds of things. Is that something we need to be talking about as an SEO?

Like, I know you had an example you found in traditional search measures. You found some, I think it was like bank statements or something like that for a particular bank. You just found it on the SERPs. It was like a perfectly normal thing. By the way, guys, there's like personal, very like financial information about a bunch of your customers. And if I remember correctly, they just ignored you for a year.

They didn't respond. Yeah. And then, um, I contacted— who was it? I contacted like the Financial Conduct Authority because I was like, look, this— and then they were like, oh, you have to do all these— wanted me to fill out some detailed report. And I'm like, Why is this my job to sort this out? Like, can't you do something now?

That's it, right? It's going to be SEOs telling people and then those people being like, well, what do you want me to do about it? Like you said, like you and I working with a client and for whatever reason this happens to them, then we have a conversation with whoever our contact is, a marketing manager or whatever it is. Like, okay, we'll need to then escalate that to the security team who who then needs to get authorization from the C-suite, who then needs to discuss it with a branding position, suddenly becomes this massive thing. And as you said, it's going to be the big brands and the big companies that have this kind of stuff. And with the way those things move, exactly as you said there, compared to how quickly you can churn out crap with LLMs, that's infinitely faster than you can slow this stuff down, let alone pick out the bad stuff from training data. And the Obviously, the analogy always uses the Ouroboros thing, right? The snaking its own tail thing where you have LLMs that are training themselves on their own crap, essentially. And it's just this cycle of bad stuff. And I know we've been talking a lot about sort of like the death of the open web and a lot of big discussions about how this is going to shift and how even Sundar Pichai from Google is talking about agentic search and the agentic web being the future rather than the way you and I and the viewers and listeners at home browse as normal people, which we've done for 25 years at this point. I think for me, when it comes to that stuff, it kind of terrifies me. And I always use this example of like, what about the normies? Like the people who aren't having these conversations, who aren't aware of any of this stuff. Like you said, the people who will just trust an LLM or trust what they see in a search, especially when it comes to AI mode. If they've built 20 years of trust with Google, it's mostly given me the right answers. And then now, oh, by the way, it might be wrong most of the time, but you've already built that rapport and reputation and relationship with a company like Google and technology like Google has been doing for so long, then you're stuck in this situation of, well, who do I trust? What can I trust? Where can— take my parents, and I always use my parents as an example because they can barely order stuff on Amazon. They are the polar opposite of you and I working day to day constantly on some sort of screen doing some sort of internet usage. And there is a real kind of worry for me of how easily, obviously, older people have always been the targets of scams and stuff, right? That's been mail scams, it's been phone scams, it's been things for decades. And when it comes to that sort of stuff, I think it could be really dangerous for brands who aren't being aware of this stuff. So take a bank, for example. Usually they should be pretty good with cybersecurity stuff, but if it's not going to the cybersecurity person and you've just got poor little underpaid SEO guy at big brand to— he's the guy that realizes what's going on. What's going to happen? Because things can move so quickly, you can see brand reputation damage, financial damage, legal ramifications, all kinds of stuff that can be massively damaging to companies. But what do you do about it? Right. There's a real kind of situation where I don't know how prepared a lot of these brands, both big and small, are. for the bad side of this thing. A lot of people obviously were all kind of like chasing the new shiny thing and like, oh, you can do so much cool things and you can vibe code and you can do this and do that and build all these things and it's so much quicker and more efficient. Yeah. A friend of mine the other day who is, um, like head of, head of social and content or something like that, uh, um, at a company, and they were talking to their CEO of the company about it the other day. And there was like a company-wide announcement of like, you should all be using AI in your everyday thing because it improves your efficiency. It was like, okay, I see what you mean. It was like, AI is— it was like word for word, AI is my doctor, my lawyer, my da da da, my da da da. And I was like, that's all the incredibly dangerous things you shouldn't be using AI for. If it was like, AI is my copywriter and my email organized, I'm like, fine. But it was doctor, lawyer, all kinds of stuff. I was like, that is terrifying. And the fact that a CEO of a company is thinking that way can lead to very easily stuff going badly very quickly. And I think coming back around to what you're saying about the quadrant kind of things as well, we don't know what we don't know about a lot of this stuff. And I think even what Google have been saying, you know, for a while now, you can no longer just dial back an update and fix things. You can no longer just say, oh, sorry, we didn't mean to completely tank that website in this helpful content update. It is all so integrated and so intertwined and so entangled. You can't just pluck at one thread and undo an issue. You have to, as you said, try and get the ink out of the water. You're going to—

Yeah.

Get out one of those spectrometers.

Put more dyes in, see if you can change it back to a color you like.

Can you just get it back to clear or slightly blue or get one of the mass spectrometer and spin it out and separate the liquids into different— Yeah, cool. Great. You have one of them at home, do you? That's not how it works. And I think a lot of the technology we're seeing is running away from us, for want of a better phrase. Right. Like a lot of us, I know you've been evangelizing a 3.5-hour video about how LLMs work and stuff like that. And there are some very good explainers on there. But the vast majority of people, not only the vast majority of people in general, the vast majority of people who use LLMs for this kind of stuff fundamentally don't know how this technology works. And even the people empowering it, investing in it, funding it, seem to kind of not know how a lot of it works either, which is terrifying as well. And I think when it comes to expired domains, and it seems like a small kind of thing, right? And something we've been fighting in Google for years and years and years. You can get very easily stuck in a place where that will be the SERPs for a meta phrase in AI search. And Well, you can't get rid of that. Like, what are you going to— you're going to noindex your spam pages one by one by one faster than people can generate them? Good luck. Like, there's— yeah, I'm— I feel we're heading towards a layer of enshittification here. It feels like you seem to be— before we started recording, you seemed equally pessimistic. Yeah.

Yeah, I think things are going to get significantly worse. before they get better. And by that, I mean for normies, as you put them, for just general people.

Yeah.

I think there's going to be more easier ways to harm yourself through trusting LLMs and them hoovering kind of stuff up. And as professionals working in that space, I think there's going to be more work than ever in terms of trying to not close Pandora's box, but kind of run around trying to stop the insects getting in the fridge and—

Putting the bandages over the leaks.

Yeah, no, there's gonna— I think there's gonna be a lot of, um, there's gonna be a lot of damage control.

Yeah.

For— and mitigation work because the nature of gen AI is that it's messy. And I was describing it last night to someone because they were talking about— I was describing gen AI as a bit like a leaky bucket, right? In that you're carrying stuff around and it's kind of leaking on the floor, and that's acceptable because of the speed at which you can move. And this agent—

And just copying it with your hands.

Well, no, so that, you know, they were like, oh, but you can You can make another agent that checks the output for this. And I was like, yeah. And it's essentially then you're putting a leaky bucket under a leaky bucket.

Yeah, exactly.

So you are losing less water, but you're increasing complexity. And it's like, you will never have a— And then with this method, you'll never have a non-leaky bucket. You'll never have a non-leaky bucket. But that approach and the nature of GenAI means that People will want to continue to move fast because, like you said about your CEO guy, it's profit sufficient, it's profitable to do so, right? And that's how capitalism works, like more for less. But the stuff that's leaking can be pretty serious at times.

Yeah.

So I think there's going to be a lot of work in firefighting around that to allow people to kind of run around doing this, but minimizing the damage to do so. Because I— that, you know, and that's why, um, like Dixon's tool, again, not paid to talk about it, um, WayK.

Yep.

Um, What AI Knows About You. One of the features of that tool is it presents facts that are known about your brand from models, because you're essentially doing the firefighting check of where have we leaked. Oh, it says also us does this. We Definitely don't do that.

That's the quadrant thing as well, right? Like, yeah, what does it know about you?

What it doesn't know about you. So that's already showing that we're having to build tools to make visible the errors inherent in GenAI before it gets to customers and try and fix them. And again, going back to another analogy, it's not a simple matter of if an LLM is giving false information about your brand. You can't just go in and update it. Like, if you're very lucky, it might be doing like a RAG type thing where you can update a source, but more often than not, it's just a token thing in the model.

Yeah.

That's our ink in the water thing trying to fix. You just have to dump more of your correct color in the water and hope it overrides what is already there. And that can take a while. If it's not doing, for instance, if it's not doing retrieval augmented generation, so it's just generating something from the model, You kind of got to wait for it to be updated, which doesn't happen.

Yeah. So, um, yeah, we talked about that with Google a lot as well, right? Like that happens with brand reputation. And obviously we talk about how important brand is at the moment. I've got an episode coming out in a couple of weeks talking about how important brand is at the moment and how a lot of people think about it wrong. Obviously regular conversations with people like Mordy Oberstein talking about this kind of thing. And I wonder like how many people, like you said, factor in the efficiency and the more for less and wow, you can make so much money and it's so much easier. Don't think about that risk factor. Don't think about, like you said, oh, if something can leak and it's like, yeah, cool. Until that's a really important thing that like breaks your website, puts you out of business, puts you in jail. Like there are so many very bad things that can happen for that kind of stuff. Yeah. And the opposite side of that, coming back around to what you're saying with the quality of LLM traffic, I use this example on the podcast probably a couple of weeks ago at this point, where there was a client I was working with here, still a client of ours, but no longer a client of mine, where each lead is worth like tens, if not hundreds of thousands of pounds. They're big industrial projects. And I remember having, when I first onboarded with this client, when I joined Kanda 4 years ago, Having this conversation of how are you tracking your leads? What's the process here? And the numbers were never big. They're not a big, like, tens of thousands, hundreds of thousands of clicks a month kind of company. But one click can turn into a conversion that can be worth half a million pounds very quickly. So the value of a quick click literally translates higher. So you're thinking about conversion rates in a different way. The difference between return on investment of an SEO retainer with an agency like Kanda to getting a lead like that is like, cool, one lead can pay for 2 or 3 or 10 years of essentially an SEO retainer, right?

Yeah, right.

But then on the other side of it, like, are you willing to also risk that on the other side of things if the wrong piece of data leaks? It's kind of that balance of, yeah, cool, you can gain a lot of stuff and there is potential there for fewer clicks but higher conversion rates and the chance of getting those leads. But if you are doing that kind of stuff and you are susceptible to that kind of thing, there is a risk of doing that on the other side as well. And I think a lot of people are not being aware of that balance and the—

We've had it on All 2 Arms before. We've had someone that's purchased and then asked, can it do this? And I was like, no, it can't do that. And they're like, oh, I read it could.

Interesting.

And it turned out they'd had a conversation with ChatGPT and it told them we could— I can't even remember what it was now. We could do some— we had some—

Research for them on the question. Yeah, right.

I was like, no, we couldn't do it. And they were like, oh, it said it could count my money back because that's what I wanted to do. And I was kind of like, well, yeah. And then we get charged a fee to do that, obviously. And I was kind of a bit annoyed because I was like, nobody's done anything wrong here except OpenAI.

Yeah.

And I've literally now had to—

you're out of pocket.

This person's, yeah, wasted their time. I'm out of pocket just because they're making stuff up.

Yeah.

I'm like, come on.

Yeah. Yeah. I do wonder how often we're going to see that kind of thing where Like even a conversation I had with an e-commerce client here a few years ago, like they wanted to be ranking for a fairly broad sort of industry term, like SEO agency as an example, right? I was like, do you talk about that on your homepage at all? Like, do you mention that phrase? Like, we're a digital agency in Norwich, we're a marketing agency, like, right? You don't say SEO agency. And I know, I know SEO is no longer as simple as just put the keyword into the homepage and stuff. But if you're not doing that already, that's the bare bloody minimum, right? And we had this conversation. I was like, literally, let's add these phrases to the page title. Let's add a bit of copy to the homepage. Suddenly the homepage was ranking. It went from like the 60s or 70s for this phrase to like top 3 in the next couple of days. It was like really, really quick and a big like shift. And I remember having a conversation with the boss of that company. It was like, Yeah, we probably should have been doing that already. I'm like, yeah, that is exactly what you should be doing. If you're thinking about as soon as you as a user, as a customer, as a user of the internet land on a page, you should get an idea of what that company does and why you should care about it.

It sounds so simple now.

I know, right? Right. Yeah, exactly. But there's the element of it, right? There's a— I can't remember. I might have even been the aforementioned Chris Green I was talking to the other day. where there was this, this nature of come around, you were saying earlier, of like the research is done in the conversation stage. There's less informational content that people are literally going to browse on your site. So things like internal links and you trying to kind of build that customer journey within your site is going to be less relevant as more and more of it happens externally, essentially on the SERP of AI search, right? So the actual pages people are going to click on Because they're more likely to convert and there's that clearer intent at the end of it. It's going to be those deeper pages of the products, of the specific services. It's not going to be how does this product work or a blog post or a homepage.

I think that's great because a lot of websites I feel are by design anti-consumer.

Yes.

In the same way when you go into a supermarket, all your daily stuff you need is at the back of the store normally. So you have to walk past everything else. And websites pull the same shit, right? They're trying to upsell you, they're trying to do this. And this is now the equivalent of, you know, you get to sit in your car outside the supermarket and have your concierge go in and be like, yeah, bring me the 3 loaves of bread that people like. And this one's cheap and it's not expensive, and this one has seeds and is healthy for you. And you're like, I want that one.

Yep.

That's essentially the websites being skipped out.

Yeah.

Which I think is good for the user in many ways, but Yeah, the robot sometimes just gets the wrong stuff, which is kind of problematic.

I'll pick up on that robot thing in a second with the agentic side of things. But what I was talking to Chris about was that means you no longer just need to do that on the homepage. Traditionally, you would say like, right, if somebody lands on your homepage, that's going to be the most linked to page. If people are searching for your brand, you need to explain what you do, right? If somebody lands on a lesser-known product page that isn't even one of your bestsellers, isn't particularly featured or whatever it is, but they found it in a search, and it's just the product with no context about what you are and why they should buy that product from you rather than say you're a wholesaler from another person, right?

Okay.

You maybe— this is again a whole kind of theoretical conversation here of like be including basic brand information on every single page that people are going to land on. So including things where you say, oh yeah, we are this, we have been around for this long. This is why all those trust signals you would typically do on an about page or a homepage or, you know, Trustpilot reviews, Facebook reviews, Google reviews, all that kind of stuff, including that stuff on the deeper pages that you maybe wouldn't necessarily include that on because they are the new landing pages. Right, for sure. People might not ever see your homepage because unless they're searching for a branded search or an informational thing and all the informational stuff is going to be dealt within the chat itself, So if you're coming straight in with commercial intent, but you don't get that extra information of like, cool, I've landed on this page, you might have that intent to purchase. But the classic thing, and I know again, we had a conversation with a client a few, couple of months ago at this point of like marketing 101 stuff of why buy from brand A rather than brand B, right? What differentiates each one from each other? And it's an interesting thing. Like that's a whole restructure his website, but rethinking about what are going to be the key landing pages and those like touchpoints and access points for users. Traditionally, it would be a homepage and then, or a blog post, as you know, we talked about often with that kind of internal linking strategy I talk a lot about with, it was even my SEMIN 2024 recommendation last year was informational content for e-commerce and using CTAs and internal links and all this kind of stuff, there is a right way to do that. But with informational searches being pulled away from the site itself, I think you might need to include a bit more of that in those commercial pages and have them as a more branded, more kind of customer-focused kind of thing. I don't know, I'm thinking about it from, from an agentic side as well. I know we had the example, the first kind of demonstration of ChatGPT, the agentic search, it was, get me a roast dinner from Tesco. Because you used the example of sitting in the car park and doing it there, right? And I think something as vague as that— sorry, non-British listeners, but roast dinner is a big deal here in the UK. Classic Sunday roast type stuff. And as you said earlier, that is essentially doing 10, 15 product searches at once and also customizing each one of like, oh, but Jack's vegetarian, so he wants a nut roast. Whereas, well, if Jack and his wife are going to visit Mark and his wife, then we should, you know, we need a vegetarian option, we need a meaty option, we need a gluten-free option for this person as well. Oh, they're hosting other people, blah, blah, blah. All these different things. You're then doing 15, 30, 45 different searches, comparing different things. It's comparing prices, it's comparing Oh, by the way, I have this particular dietary requirement. Okay, that eliminates those ones. Then I want something under 600 calories. So that's eliminating those ones as well. That for me is, like you said, the experience of browsing around in a supermarket. As a lifelong vegetarian, people often ask me like, oh, do you struggle to find this kind of stuff? Like, well, no, I just don't go to the meat aisles. Like, I don't struggle to find the veggie stuff because I just go straight to it. I know what I want.

Yeah.

And I think the same is true. You know what you want. You know you want a product. You go straight to that product. For me, when it comes to almost balancing the supermarket, kind of like you said, the upsell kind of stuff, I wonder how much of that will become practice again in websitey kind of stuff. When you've got people who are not landing on your informational pages, when they're going straight to your commercial pages, do you then try and really hammer the upsell? Or will that piss people off? Because usually for me, being a very cynical shopper, that tends to piss me off. If somebody is like, I've landed on this specific thing, I clearly have a search intent for this product.

Yeah.

And you're also trying to sell me 3 other things at once that aren't even necessarily related. I'm like, yeah, fine, whatever. That will often put me off a brand being a bit too pushy and a bit too much. we always say the same thing with display ads and newspaper websites and stuff like that. I have stopped using websites and stopped reading news because ads, ads, ads, ads, ads, ads, ads, and it's just literally unreadable. I wonder with the agentic web, with the way LLMs serve information, how important those commercial pages then come as the primary source of click data, the primary source of all that kind of stuff.

I think eventually you won't need to visit the page, will you?

Well, true.

Yeah. The point of the agentic stuff is it will do the transaction for you so you don't get exposed to any of that.

But what I mean to that is like, you're still, as us, as I guess creators of web pages, you are still serving that information in order for the conversion to happen.

I think the actual like product Data is more important. I saw an experiment of someone asking for these like heavyweight t-shirts, and then they did a Google search for the same thing, and the Google search was returning quite a few big brands, but they weren't being returned in the LLM. And when they asked why, it was because the big brand didn't include the, uh, GSM, the weight, the material weight.

180 grams per square meter. Yeah, in the listing.

So it couldn't confirm it was a heavyweight t-shirt.

Interesting.

So it didn't kind of— it had— it decided there was a threshold of, what, 250 or whatever GSM to include it. And all the sites that didn't have that in the product listing, it didn't include. So I think there's probably a lot more in terms of in-depth and maybe, like you said, some information to swing it in terms of brand. But I I feel that the website, the LLM to website thing is, is a stepping stone.

Yeah.

And we're very quickly, if the tech, if they get the technology working and they fix the leaky bucket thing, will be the, you just tell it you want this, it'll do the transaction for you. I mean, that would be really easy, right? I, then you don't have to make those different accounts on websites. It just fixes a whole bunch of problems. But again, Probably a good place to end it there without, um, going into a whole other thing. But essentially, LM's become almost like an Amazon gateway to commerce, which again is going to be terrible for consumers in terms of monopolies and prices, and it's gonna be terrible probably in terms of businesses.

But Ape and I have confirmed, essentially confirmed, that they will be taking commissions for any of their Shopify partnership stuff as well.

So they're going to get jacked up. They're going to be Prediction for you. It'll be super cheap. And it's going to be the best thing to do to begin with when you have a Shopify store. Yeah, everyone's gonna jump on it. People are gonna make loads of money. And then once people are comfortable using it, once consumers are comfortable doing it, by the way, here's a point of view coming out.

Yeah.

And it's going to do exactly what Google Ads has done and just aggressively rinse brands of every possible percentage point they can get.

Yeah, because obviously we briefly mentioned the antitrust stuff with Google. They're also in a separate one about the paid side of things as well, which is separate from the organic search side of things. And I think you're totally right. You can get into that first taste is free kind of mentality where they will tempt you in with, oh, for the first year, no fees or 5% fees or whatever it would be. And then they slowly, slowly start cranking up. Suddenly it's 10%, then 15%, then 20%, then 25%. And like, oh, what, you're going to stop now after you've invested all this time and all this money and stuff in building your products and services to be served by this system? Good luck, mate. What are you going to do now? And you're stuck and kind of trapped in that system. And I think for the agentic side of things, that's really interesting. I know I used the example of when it was first announced, when we last spoke, Google I/O stuff of like, cool ticket comparisons and stuff like that. How do you get around scalping and dynamic pricing and all this kind of stuff? There are so many ways to manipulate that stuff that we're already seeing in so many different parts of the web that I don't trust. Like we said, I don't trust this information to get this stuff right. And this— sorry, this technology to get the information right when we can so easily manipulate it with like even going back as like eBay sniping and stuff. We've been doing this stuff for decades, right? Like having like automated clickbots.

Yeah.

Do it one last extra bid with 2 seconds left and all that kind of stuff. That stuff's been around for ages. And I can totally see scalpers using agents to do this stuff. And then it's just gonna be a battle of scammy robots across the internet. Well, hopefully you're not too depressed, folks. I feel like a lot of the episodes end with existential dread at the moment. It's just like, well, Everything is going down the toilet.

So good.

Faster than we can control. But yeah, I will put links in the show notes if you want to go and check out Miriam's post talking about clickstream data, the Netcraft study we mentioned, and also talk about the hacked and expired domains study that James from Digital Loft did as well. So go and check those out in the show notes, folks. Of course, please go and follow Mark Williams-Cook. I assume you probably do already. Uh, you're doing the usual Unsolicited SEO Tips, you're doing talks and stuff. Have you got any talks coming up?

Uh, the next is Digital Olympus in Amsterdam in September. Really great lineup there. And then beginning of October, that Search and Stuff in Antalya, which I would highly recommend you go to. I went to the inaugural event last year. Absolutely fantastic. Everything down to live translations. So there's English talks, yeah, Turkish talks, but the professionalism of the live translation is amazing. Yeah, yeah. And even all the technical terms, it was just straight through. So that was really cool. Um, got to enjoy a lot more talks than I normally would have. I think it's over 3 days, like beautiful workshops.

There's all kinds of stuff like— yeah, absolutely. Um, I'm gonna be talking Brighton SEO, which is— you are? Which is crazy. Yes, I need to finish my slides.

Yes.

Yeah, that's a terrifying thing. So that's 23rd and 24th of October coming up. I don't know what date or what time or anything yet. I'm also going to be moderating again. So I'm going to be a busy boy. But I know they've done the— to current ticket holders, you do like the vote to see, oh, I definitely wouldn't miss this track or whatever. And that kind of positions you on what stage you're going to be on. So fingers crossed. Oh, really?

Is that how they do it?

Yeah, apparently so.

I didn't know that.

It's borderline a popularity contest kind of thing. Obviously it's a topical thing as well. But yeah. I was talking to a lot of the other speakers I know who were going, I mean, like, oh, just voted for you. I was like, just voted for you. Like, I'm trying to work out engagement pods. I know, right? Yeah, yeah. Pretty much what it turns into.

Have you got an agent that can do this?

I know. Yeah, it's called Jordan Dimitrov. Shout out to you, Jordan, because Jordan's speaking at Brighton as well. And we've been like, first time speaking at Brighton. Okay. Yeah, this is kind of scary. But yeah, so funny enough, my talk will be semi-relevant to this conversation. It's about the future of content creation and publishing with zero clicks, monetization, communities, subscription models, all that kind of stuff. Even talk about what Google has recently launched with things like Offerwall and things like that. So yeah, lots to talk about. Big topic. And I know you have said it to me and a couple of other people said like, big topic, mate. Good luck with that. I was like, I hadn't really thought about that, but yeah, pitched the idea and then Hadn't thought of the consequences of my actions, but I'm sure I'll be fine.

Sure to be fine.

My only worry is that I will be too long. I'm comfortable public speaking. I'm comfortable being in front of an audience like that. I just need to get that timing nailed.

Yeah, we'll work on it.

We'll work on it. We'll tighten up. Yeah. Anyway, folks, thank you so much for watching. Thanks so much for listening. Mark and I'll be back hopefully quicker than 8 weeks' time. We'll be back next month is usually the cadence of these things. I've got another In-studio episode coming up with the one and only Seb Atkinson, I think probably in a couple of weeks or so, sort of, sort of end of August sort of time. So go and check that out as well. That is very interesting talking about, funnily enough, Seb's content creation process he goes through with clients and talking about a lot of the ways he's adapted, you know, revenue tracking and really kind of demonstrating the value of SEO to your clients and to your stakeholders.

Smart guy.

Super smart guy. I think off the top of my head—

No, he is.

No, he definitely is. I think Seb is the only person who has been on the 3 main sort of like iterations of Search with Kander. So with you as a guest.

Yeah, he's one of the first guests I had. Yeah.

With me in audio form.

Yeah.

And then with me in video form here in the studio. And all 3 of them were recorded in the Kander office. I think the other one was previous office. But he's the only person who's done that, as far as I know, which is kind of a little claim to fame.

Let's give him a little trophy.

Exactly. Yeah. Yeah. But yeah, stay tuned for that. I'll have lots more cool conversations and stuff coming up in the next couple of weeks. Thanks for listening. Thanks for watching.