The Agentic Web: Future of Ecommerce in the AI Era

AI isn't just changing how people search. It's starting to change whether they search at all.

I'm talking about the Agentic Web - an internet where AI agents browse, compare, and buy on behalf of users - and it's rapidly moving from a concept to a reality. The SEOs who understand how it all works will have a real advantage.

In this webinar, Will Critchlow, CEO of Search Pilot, and Katelyn Geary, SEO Manager at JD Sports, got into what's genuinely shifted in the last twelve months, and what teams should actually be doing about it.

They cover:

• why "page vs. product" and "keyword vs. conversation" are the real strategic pivots

• what a serious AI discovery test backlog looks like right now

• which traditional SEO fundamentals you really can't afford to skip

• why cross-functional collaboration (with your paid team, your merchandising team, your data team) has suddenly become the unsexy blocker on all of this, and

• what the website's job actually becomes when shopping journeys get more agentic.

All that and more!

Watch the webinar

By the way, you can also subscribe to the Sitebulb YouTube channel! And register here to be the first to find out about upcoming webinars. 

Webinar transcript

Patrick Hathaway

Hello, everyone, and welcome to another Sitebulb webinar today where we'll be talking about the future of ecommerce SEO and how it is shifting as we move towards agentic commerce. I'm Patrick, CEO and co-founder of Sitebulb, and I'll be your host today. Sitebulb helps in-house and agency teams get to grips with technical SEO on ecommerce sites of all shapes and sizes, including some of the most well-known retail sites on the planet.

Patrick Hathaway

And we have two amazing guests on our panel today who are very well versed in ecommerce. So please welcome Katelyn Geary and Will Critchlow. So a little bit of housekeeping before we start. Everybody all the time asks if we're recording the session. The answer is always yes. So anyone who signed up will get the recording emailed to them tomorrow, courtesy of Jojo, who you will see already in our chat. And we also always save fifteen minutes at the end of the discussion for Q&A. So please put your questions in the little Q&A box on the left of the chat bit, not in the chat, which, again, everyone always does, and we always have to tell you, go and put it in the Q&A. And then at the end of the session, we will basically be on screen asking the panel the most upvoted question. If you haven't got any questions yourself, still check out the Q&A because there might be some in there that you are also interested in. And if you can upvote them as well, then we tend to work from upvote down. So thank you everybody for joining. So let's get into it and meet our guests. So we have Katelyn Geary. She has direct experience growing multimillion dollar ecommerce sites in her role managing the SEO at JD Finish Line. And Will Critchlow's company, Search Pilot, works with some of the biggest retailers on the Internet, literally proving what works and what doesn't with SEO split testing.

Patrick Hathaway

Will, everyone talks about AI search as if it kills SEO. From an ecommerce perspective, what has actually changed in the last twelve months? And what still has not?

Will Critchlow

Well, firstly, thank you for organising this, Patrick. Great to be here, and thank you for including me. I heard a great comment in one of our feedback forms the other day where somebody described AI search — or GEO, or whatever you want to call it — as basically SEO but with more panic.

And I quite like that description. There's a lot of executive attention and leadership focus, and suddenly the boardroom's interested in Search in a way that hasn't really been true for decades. And I can't speak for everyone, but I'm pretty excited about that. I also think it's great that there's some competition for Google, even though Google seems to be playing pretty well at the moment. And that innovation is exciting.

But we're in the early days. I think the biggest thing I would say is: remember regular search is still bigger. Google is still your biggest channel, your biggest AI channel even — even though we've seen the explosive growth in all of these chat-based platforms.

That's kind of the status quo. We're seeing the explosive growth, so it's all going to change. And so I think the big thing that I'm talking to teams about right now is that forward-looking thing — a 2026 click or a 2028 click is not the same thing as a 2021 or a 2023 click, in the sense that people have probably done a lot more research and been told a lot more information before they ever get to your site. And that's the thing that I'm really focusing on.

Patrick Hathaway

So would you say that not a lot has actually changed in the last twelve months? And it's more of a future-focus thing?

Will Critchlow

I would say that is probably true in the actual, kind of, button-pressing sense — what keys do you hit on the keyboard. Mike King had a great way of framing this, which was: the exact same activity could be, in other industries, PR, outbound sales, link building. Right? You're sending emails to persuade somebody about a story that you're talking about. And in the same way, SEO and GEO can have different purposes, different goals, different objectives.

At a very tactical level, we are now running experiments in what we call multi-metrics mode, for sites where there's enough traffic to independently run experiments on LLM traffic and separate it out from regular search traffic. We're now able to measure the net combined effect as well as breaking it down by regular SEO and GEO. And we have seen our first examples of hypotheses that apply to one and not the other — or are even positive for one and negative for the other. So they're definitely not exactly the same thing, but the majority of the activity is still the same in my experience.

Patrick Hathaway

Should we be expecting case studies and research papers?

Will Critchlow

100%, you should be expecting case studies. If we haven't published case studies soon, then send somebody round to my house and have a word.

Patrick Hathaway

And then, Katelyn, have you got any — from your side, working actually within a retailer — anything that you've seen change in the last twelve months?

Katelyn Geary

Yeah, absolutely. So I think if I had to put it in a nutshell, there are two things. It's a shift from page to product — speaking about entities versus URLs — and then a shift from keyword to conversation. And I think that's not only in AEO, AI search, GEO — that's across SEO as well. Right? We've been looking at search intent versus just straight keyword. So I think with AI, it's leaning more heavily into that, and joining that conversation.

So it's less — I don't want to say it's not about optimising your pages. By no means. That's one of the caveats I want to set ahead of all of this. This is not — at least for me, and from what I'm getting from Will and Patrick as well — this is not "mission: abandon traditional SEO" and put all your eggs in the GEO basket. That would be a travesty. If that's something that you're trying to do, good for you — drop your name in the comments and I will follow you on LinkedIn. But I think maintaining that foundational traditional SEO — making sure you've got crawlability, indexability, and making sure your sites are fast — those types of things, you can't forsake those. You have to start there.

And then from there is where the process kind of changes: looking at, do you join the conversation? What are people asking? What are the pain points? And then focusing less on page-specific optimisations, but rather entire entity optimisations, and helping AI understand the whole scope of what a product is. Right? Not just "these are red shoes." But answering all of those queries that are going to reduce friction in the customer journey.

Patrick Hathaway

So the Search Pilot study you're talking about — let me just drop that in the comments actually, because everyone watching basically needs to go and read that. There's loads of really valuable stuff in there. The quote you mentioned was "AI systems reason over product entities, not URLs." We're just talking about that. So does that change what you're doing day to day?

Katelyn Geary

It does. I think the — let me start with a little anecdote. My boss had our entire team do one of those StrengthsFinder activities where you fill out a quiz and you find what your strengths are. So I got back my results. I was really surprised. I was thinking something like "competitive" or "activator" or something like that. I was actually "developer." And that's not in the technical sense — that's more in terms of developing relationships. I'm someone who really likes to bring people in on projects. I like to collaborate. I don't like when people feel left out.

So for me, at first, I was like, yikes. That's not really how I feel in the workplace — not that I don't want to include people, but sometimes I can be heads down, super in the zone and focused. So "relationship building" doesn't seem like it's at the forefront for me.

However, now that we're leaning into this whole entity creation and understanding, that cross-collaboration with other teams is extremely important. And so now I'm kind of in the space where I'm like, hey, I'm thriving here. I love being a developer, because we have to reach out to teams that we may not normally work with.

For example, like our paid team and managing our feeds. Right? We need to make sure that all of our attributes are filled in. So you can't just say, "Google Merchant Center, that's on our paid team, that's PLAs." You need to integrate. You need to be part of that vetting process to make sure that each product, every single attribute — do not skip. So in that sense, working with the paid teams, you need to have an ownership of GMC and the feeds that are being produced, a little bit more upstream.

Working with your merchandising team and seeing, you know, what are they entering in the PIM? Is it coordinated between the attributes they're including and the attributes that are being included in Merchant Center?

So overall, I think the place where we need to start — and I'm trying to start — is with that cross-collaboration and relationship building, to say: hey, this is a team effort across the entire organisation, to build this entity that is recognised and trusted within AI search. And so for me, that's really the foundational change — it's less about the actual strategies and tactics, but more about getting the buy-in, and really building to say: you are contributing to how we appear in AI search. You're contributing. You're contributing. How do we all get together and make sure that we're on the same page?

And I think that's probably one of the biggest challenges — getting out of your AEO, GEO, SEO lingo and figuring out how to connect with other teams, because I can talk about building an entity, I can talk about products over pages. What does that mean to a merchandising team? Maybe not a lot. So sometimes it's the type of thing where you take them to ChatGPT, you take them to Claude, whatever it is — you search for whatever it is and show them right there on the screen: here's what's going on, here's what we're trying to achieve, and here's where you can come in to help with that visibility. So yes, the tactics will change, but I think the first thing that teams need to focus on is that cross-functional collaboration.

Patrick Hathaway

And I think as well, because it's obviously the hot topic and everyone's talking about it, you can literally show what's happening — which perhaps allows you to get buy-in that maybe we couldn't get previously, because it was more "oh, it's SEO doing their thing over here."

Katelyn Geary

Yeah. And I think it's an exciting time for SEOs too, because as Will alluded to, there's this sense of panic. And I think SEOs can channel that and make it exciting, and turn it into an opportunity. Everyone else in the organisation is looking around for answers — "what are we supposed to do? I'm seeing case studies on this, I'm seeing these posts on LinkedIn." That's the opportunity for SEOs to rise up and say: if you have a question, if you have a problem, come to me. We're actively researching and tracking these things and we can help guide whatever questions you have.

So I hope anyone who's watching takes away — whether you're an SEO or just within your marketing org — that you be the person to stand up and advocate for the direction of GEO, AEO, whatever your strategy is. Make yourself the authority there, because we really need to drive this forward. And all of you are here because you are doing the legwork to lay this foundation. So you should be the expert in this space. Instead of panicking, look at this as an opportunity to be the one that stands up on the soapbox and says: hey, everyone, come with me. I'm tracking. I'm testing. Let's go together.

Patrick Hathaway

Okay, so let's think a bit more technically then. If a retailer could improve just one thing this quarter for AI discovery — and let's pick: product data completeness, the quality or accessibility of content, or brand authority — where would you put the budget? Will, you go.

Will Critchlow

I'm going to pick information completeness — with a little caveat, if I can put an asterisk on it, of crawlability, if that's the biggest block. If it's kind of blocked or completely unavailable — and I'm assuming they've got kind of SEO basics in place — the biggest difference is JavaScript execution. So if you want to be visible in these platforms, particularly the real-time fetching from things like ChatGPT, then your stuff needs to be available even without JavaScript enabled.

But if we're not in that asterisk caveat, I'm going to say more information — and that's the sneak preview hint of where some of the testing that we've been doing has been going. Either summarising information, adding additional information, or repurposing content in a way that's more digestible for an LLM. Those are the kinds of things that I would be experimenting with. And on brand, I'm going to say all these things should be experiments, because it's all just changing so fast and nobody really actually has a clue what's working today or what's going to continue working.

Patrick Hathaway

I feel like you just somehow managed to swerve around saying "it depends" in a very clever way.

Will Critchlow

I mean, that's my whole job. But let's test to find out what the answer is in each individual situation.

Patrick Hathaway

This isn't your first rodeo. Katelyn, do you have a different take? Would you put the budget in a different place?

Katelyn Geary

Will took the words right out of my mouth. I think step zero is just checking to make sure that your content is rendering server-side, or at least is not completely hidden. Obviously, every single page can't fully render server-side, but making sure those key components are visible to AI search — like, do not do anything else. After you get off this call, after you get out of this webinar, go check your website and make sure it's going to be visible to AI bots. Because you can do all kinds of amazing content optimisation, structured data, everything you need — you can have this fully functioning, conversion machine. But if none of that is visible, what's it all for? So start with step zero, which is making sure your content is visible to AI bots.

Patrick Hathaway

Yeah, which is basically the tech SEO bit. And if you need help with that, go and look in the comments because Jojo has just put a couple of resources — our JavaScript SEO training and Tech SEO in the AI era pieces.

Will Critchlow

If you don't mind, Patrick, I've got one result to drop in there as well. So if people want some of those — we're talking about enhancing product pages and adding more or different kinds of content to them — my colleague Demetria actually did a webinar on it recently. I'll drop the link in the chat. It's positioned as being about testing content produced using AI, but actually there are also quite a few hypotheses in there that are designed to be measured for their impact on AI discovery and GEO. So if folks are interested in some specific test examples or hypotheses, they can check that out.

Patrick Hathaway

That's basically what my next question was anyway. I was going to ask you, Will, to put your Search Pilot hat on and let us know what you'd say a serious AI discovery test backlog would look like today.

Will Critchlow

Yeah. With the caveat I mentioned earlier that it's all in flux — so you're going to want to test and retest, and I can't tell you the exact perfect prioritisation because we're finding different things work in different situations. But some things that we've been experimenting with:

One is reformatting content — so turning product specs into a kind of "key features" content piece, for example.

Adding a more authoritative introductory paragraph on PLPs. You might notice all of these could be SEO hypotheses as well.

Buying guide Q&A is quite an interesting one.

A lot of freshness stuff. One of the key things that LLMs go out to the web for is up-to-date information. Think about it: they've got their training data, and then they've got everything that's happened since they were trained — that's from the retrieval augmented generation, the RAG stuff — and that's where they're going to the search results themselves. They're doing that because they don't know up-to-the-minute pricing information, in-stock status, latest reviews, product drops, et cetera. And so showing that your content is fresh in those ways, and having the freshest and latest information both on the page and highlighted on the page, is key.

And then there's a lot of debate and discussion about structured data and schema. It's hard to tease apart what's good for what, because obviously, as I've said, all these things are SEO hypotheses as well. Some of the ways that you can have an impact in AI discovery is by ranking better in regular search and showing up in the RAG. But all of these things put together — tied in with what Katelyn was saying before about entities, not just URLs — one of the things that can connect entities and URLs is structured data. So I'll be testing new content, content freshness, richness, freshness indicators, and structured markup as my kind of top ideas.

Patrick Hathaway

Nice. And then, Katelyn, if you were going to take any of those ideas — which ones do you think would survive most inside a large retail organisation? Would you get pushback on any of them, or would any of them work more easily than others?

Katelyn Geary

Yeah. I think the easiest way is to try to pursue initiatives where workflows already exist. Because, again, I find this is going to be a foundational year in terms of that relationship building, cross-functional collaboration, and determining what's the shared language to ensure that we're all on the same page, aiming for the same goals.

So leveraging the workflows that you already have. For me, like Will was saying, structured data — that's my new power word that I'm going to lean on this year, in the same way that server-side rendering was last year. I'm all in for that.

A specific example would be collaborating with our paid team and really making sure that our structured data exactly matches what is in our Merchant Center feeds. I've had experience previously where the feeds and the structured data did not match, and it is a nightmare. It's very technical and nitty-gritty. But you have to do it — it's foundational. And if your on-site structured data varies from what the merchandising team is entering in the PIM, or what is in Google Merchant Center, that's a mismatch. That's going to signal a little bit of mistrust. So testing in terms of what's going to stick — work with the teams that you already have connections with, and unlock opportunities there.

Teams that you may not be as closely connected with — use this year to build those foundations. But if you want some immediate traction, or buy-in that might not get too much pushback and is probably more easily understood, lean into the partnerships and workflows that you already have.

Patrick Hathaway

Cool. Alright. So we've talked a little bit about being discovered by AI. But what about using AI yourselves? What things could retailers be doing using AI in ways that would genuinely help customers, beyond just manufacturing more content?

Katelyn Geary

I'm all about using AI for content creation, and I'm not going to say that we should shy away from it. But if we're thinking about ways to use it to benefit the customer beyond content creation, I think there's a huge opportunity — as I mentioned earlier — to reduce friction. So: what are the pain points that customers are experiencing as they go through the purchase journey? And can you answer those questions either right before or right at the point of purchase?

For example, you can do prompt tracking, see what people are asking about specific products. You can go into Search Console and do some sort of regex for longer-tail queries to find what people are searching for at the product level, and include those directly on your PDP. So it's not just the typical "it's this material, it's this colour, and it's this brand." Maybe you should size down a size — that's generic. That's not going to be helpful most of the time. People do want to know those things, but you're dealing with real people here, and you need to answer the questions that are potentially preventing them from converting on your site. Maybe they're going to go to another site where those specific questions are answered — like "are these good running shoes for someone with flat feet?" Beyond just "are these quality running shoes?" I think there's a huge opportunity to track where those pain points and points of friction are, and to use AI to answer those questions and enhance your product descriptions and product details on those pages.

Will Critchlow

Not a great deal to add. My mind went to two things. One was separating the two use cases that Katelyn's talking about. There's AI in "active mode" — the user is doing something and AI is doing something for them in the moment. Whether that's answering Q&A, customer service-type stuff, something about their journey. That's different, obviously, to producing content, or some kind of offline static batch process with AI. Both have their place.

For the first one, I just remember there's a thing that has come up a lot in the more programming and engineering and company-building space, which is called the bitter lesson. I don't know if everybody's heard about this. This is the idea that the foundation labs — the big models, the Anthropics, the OpenAIs, the Googles — have essentially overtaken all of the specialised point-solution models. So the bitter lesson is that there's so far been limited value in trying to build some kind of super-specialised trained model to do just specifically your thing. Like, what size of this is going to fit my body type, or something? You might think you need to develop some kind of custom model that you train on all of your own proprietary information. And the bitter lesson is most of the people who've tried this — whether it's in legal tech or health tech or whatever — have found that actually the answer is: wait six weeks and the basic model from the frontier lab can just do it. So I think one of the things we've been trying to internalise a lot in our use of AI is to make sure that we can stay up to date with the cutting-edge model — that we're not building so much on top of it that we can't upgrade the model underneath it every time there's a new state-of-the-art model that comes out.

For the second piece, where I think it's particularly interesting is where it's driving off your proprietary information and data. Whether that's reviews that people have given you specifically, product information, testing information that you've built up over time — things that are unique to you. It's back to that "unique content" type of language. But taking that and presenting it in ways that are useful and meaningful for your visitors — that's where the value gets really interesting. Because there are things that are genuinely useful. As a user, for example, I love the summary of reviews you get on Amazon now. Sometimes you can chat with the reviews — you can say "has anybody tried this for a small cat?" — or it can be the more static description of content, which obviously could be indexable. Yeah, those would be a couple of thoughts.

Patrick Hathaway

Yeah. I really like the summarising of reviews. Although I wish they would also do it on the negatives as well, because you kind of have to go and hunt. You're like, well actually, what did people say that was negative? Because sometimes it's like, "the product didn't arrive on time" — and you're like, you've given it a one star for that? You could helpfully summarise the negatives as well, but no one seems to do that.

Will Critchlow

Yeah. But if they did, a bit of user experience tracking on those pages would show the first thing people do is go and find the one-star reviews and just check out what they were.

Patrick Hathaway

Okay. So then just in terms of measurement — traffic is already becoming a weaker signal. What do we show leadership? What are good early indicators and real business outcomes? Katelyn, I'm going to throw this one to you first.

Katelyn Geary

Okay. Well, I think if anyone has the locked-down answer to that, reach out to me, because I would also like to know.

Still kind of developing that. But I think one of the most important things — and I'm a huge advocate for this — is expectation setting. Start having conversations now with leadership and people inside your organisation about this exact thing. Traffic is an eroding metric. I think it kind of has been since AI overviews. And just let people know things are changing. If you don't have the answers yet for what you or your organisation needs to track, that's fine. Maybe that's an opportunity to collaborate with leadership and say: okay, have them explain to you what they're interested in seeing out of AI search, and then you can use your skills to deduce what possible metric you can put to that.

So for us, it's citation rate, share of citation — that's definitely one. And then looking at not just how frequently our brand is mentioned, but how accurate is the information that is being presented with our brand. Those are kind of, at least for now, a little bit intangible. But if you set those expectations — I've seen a lot of charts of "this was the metric in 2023, and this is what it's going to be in 2027 or whatever" — communicating those types of things is super important.

In terms of ultimate business goals in ecommerce, those things are going to remain unchanged. Revenue, margin, and CAC — those things will go unchanged. So I don't think there needs to be that concern. It's more about how are you going to share with leadership the new metrics that you're testing. And if you're not sure, ask your leaders — because it's a collaboration, and then you don't have to guess. You can say: okay, they're interested in seeing how frequently we show up for our branded searches. Great — we can pull that information, we can articulate what's going on behind it, and then we can refine reports as we go.

The exciting thing is when you can hone and refine some of those metrics that you're tracking until they're really tailored to what you're trying to see.

Patrick Hathaway

Do you expect that you might have metrics like revenue for humans versus revenue from agents?

Katelyn Geary

That's an interesting one. It's something I would track, but I'm not sure that at the organisation level it's something that's ready to be integrated. I think it's one of those things you can hint at and kind of talk to your engineering or data team about: is there a way that we can segment this data so we can see specifically what's coming from agents and what's coming from humans? Not something that we're actively working on. But I think for SEOs and marketing practitioners, that would be extremely valuable.

Patrick Hathaway

Cool. And then, Will, what sort of signals are you helping your customers present to show — if you've got tests, for example — that this is a positive test?

Will Critchlow

Yeah. It's actually quite interesting that the answer to that is often still traffic. And I'm going to loop back around to explain why.

But the first thing, just to zoom out a second — if we think about how far into the future we can imagine: a year or two. Five years, who knows. But a year or two into the future, humans are still the ones who want stuff. I don't think we're going to see a reduction in humans wanting stuff or buying stuff. Shopping's still a thing. And it's relevant here that we're having this conversation about retailers and ecommerce — it would be a very different conversation in, say, media, or the affiliate space. But speaking specifically in ecommerce and retail: ChatGPT isn't going to ship you a pair of sneakers.

So yes, there are all kinds of questions about which surface somebody checks out on, what checkout even means exactly. But there's still a retailer and a brand in that dynamic. Somebody's still shipping the thing. Somebody's still making the thing. And I think that's going to continue increasing. I'm bullish on the GDP of the Internet, as Stripe would say. So let's assume more people are buying more stuff, even if they're AI-assisted.

How are they going to find that stuff? Some combination of things that look like organic discovery, one form or another. And obviously there are paid channels in the mix as well. But as people go hunting and looking and searching — at some point that becomes buying. And because the buying has to happen with you — the retailer — the question isn't "are these people going to buy stuff?" The question is "are they going to buy it from you?"

I've always liked the old name for SEO — "competitive webmastering." We're trying to make it so that people buy the stuff from us instead of from somebody else.

When you look at it over multiple years, the number of clicks could do anything. Clicks is basically a vanity metric in that world — we could get ten times fewer clicks and twice as many conversions. That's why I say you can't compare a 2023 click to a 2026 or 2028 click.

But for an experiment, you can. So our hypothesis is typically: by making this change, we will get a measurable, statistically significant improvement in the number of clicks we're getting. Because over the timeline of a test, that dynamic isn't shifting in the same way. We're measuring most of them in clicks, and then we're converting that into revenue numbers to get close to the business — typically working with our customers' finance teams on that.

We're basically trying to report in revenue, but the statistical significance is typically done on traffic. Because if you go too far into rankings, you're not measuring everything — you don't get click-through rate, you're not measuring the whole universe of things people search for. If you go too far into conversions and revenue, there's so much other statistical noise around discounts, promotions, special offers, competitors' pricing, the economy, interest rates — all kinds of things can affect those. And so we're typically doing the hardcore statistics on traffic. But as I mentioned earlier, we look at that in a multi-metric world, so we can pull in all of those bits of data.

Patrick Hathaway

Yeah. I think as well, I'm not really sold that people will change to this method of "I'm just going to let an agent buy stuff for me." Like, retail therapy is a thing. Part of what people like about retail is going and looking and exploring on your own. This idea that you could just — oh, my agent has gone and bought this stuff and I had no real input in the decision other than giving it a prompt. I don't know. Not 100% sure that's definitely the future.

Anyway, moving on, because we're running out of time before we get to the Q&A from the audience, and I've got a few more for you. I wanted to talk about something that ecommerce teams should stop doing, or do less of, because the environment has changed.

Katelyn Geary

Stop producing mass content that is thin and offers nothing for the user. Produce quality content inspired by what your customers are searching. Think about the user. Yes, of course we are trying to figure out how AI looks at our content, but we know thin content that is just produced to offer multiple entry points for customers is not the way. That's out. What's in is dense, informative, helpful, quality content that really resonates and speaks to your users' pain points, their needs — if they're looking for buying guides, if they're looking for product comparisons, you need to speak to them. Don't just churn out content. I'm not saying don't focus on content. I'm just saying don't use AI to churn out mass quantities of content that provide no value. It's going to hurt you in the long run.

Will Critchlow

Also, we'll think badly of you. We won't be mad. We'll just be disappointed.

I was going to say: stop guessing. I see all of these different ideas about what is or isn't an "AI rank factor." And that isn't even a particularly coherent concept, because they're not doing ranking. Right? It's not information retrieval, really. They may be grounded in or augmenting their data with information retrieval, but that's traditional search information retrieval. It was hard enough trying to figure out what the ranking factors were in a pre-AI, ten blue links world.

Throw AI in the mix, throw in the fact that a single prompt might be doing dozens of fan-out queries in the background — so you're looking at where you rank across dozens of different queries that you don't even know what they are. And you probably don't even know what the prompt was. And maybe that prompt is unique and has never been typed before in the history of the universe. This is just an impossible game. You've got to play the science game. You've got to measure and see what makes the net impact.

Katelyn Geary

Yeah. And I was going to say the same thing, Will, about — we're all the experts in our specific business, and don't forget that. You can see these case studies — people did amazing research into how ChatGPT surfaced certain queries and things like that, and that's great. But if it's really not resonating with your business goals or your SEO/GEO goals, you need to know that and make the pivots from there. Don't just go out looking at every solution that someone has talked about, because too, that could be completely different tomorrow. Stay grounded in your business and your business goals.

Patrick Hathaway

Yep. 100%. Okay, right. This is the question I'm most interested to hear you respond to. As shopping journeys become more agented, what becomes the website's job? And does this future tilt towards bigger retailers, or can smaller brands still win?

Will Critchlow

I think people like shopping, and I think people like buying things. And I think people like PDPs. I like PDPs. I like URLs. I like websites. I like PDPs.

And I think we've seen some of this movie before — with things like Google Checkout. Remember the idea that you could check out from the Google search results without even visiting the website? You can't anymore. And I think that's because people didn't want to. There's also a question of the retailers not wanting people to, right? Because that's a whole question of who owns the user, where does the data flow, et cetera.

I think agentic research is a huge thing. But people want to see the thing before they buy it. And the best place to see the thing should be the retailer's or the brand's own website.

What I think agentic commerce most looks like in the short to medium term is fancy Apple Pay. Right? I don't want to be filling in forms. I don't need to create an account. I don't need to do any of that stuff. Just: yes, please, buy this one. And if there are any questions, my agent can handle those things. So everything after the PDP can become agentic as far as I'm concerned. But I think the flow is: agentic research → PDP → agentic checkout.

On the question of big retailers versus small: I'm actually most worried about the medium. I think there's a place for the massive retailers who can offer something differentiated and sticky — the whole experience is something they're renowned for. And there are certain classes of product I would buy from certain places because I know the brand and I know the experience and that's built up over time.

At the other end, I think there's space for the indie. Whether they're using Shopify or not, they're still indie if they're the craftsperson, if there's something really human-scale about what they're selling. Maybe they only have a couple of SKUs. Maybe they sell one thing. That end is differentiated because the actual thing they're selling is maybe only available from them, and there's something human about it.

I think it's the undifferentiated middle where you've really got to worry — where maybe you're too big for the Shopify store, but you're not big enough to have built out the agentic checkout yourself.

Patrick Hathaway

Or it forces all of those to just go to Amazon. Or similar.

Will Critchlow

Yeah. And again, we've seen variations of that movie before. Toys R Us moving onto the Amazon ecommerce platform back in the day. My bet would be a kind of barbell distribution — the big do well and the small do well.

Katelyn Geary

Yeah. So I'm going to have the hot take here. I do like shopping, but I hate comparing products. I hate having to search multiple sites, I hate having to have tabs up and see the price here and the discount here and if I sign up for the loyalty programme I get 10% off here. That's not it for me. And so I love ChatGPT Shopping — everything's right in there. I can give the prompt. And Patrick, you mentioned "oh it's just a prompt," but I'd argue: how good is your prompt engineering? You could get it pretty tailored to show you the exact results that you want.

So I think the website then becomes like this layer of trust. If I'm thinking through my point of view: I see all the products, maybe from brands I don't recognise, and I want to click through to the site. I'm going to use the website as a gauge — do I trust this brand? Do I trust this site? And this kind of goes back to some of those foundational SEO things, like: is your site fast? At that point, if someone is reaching your site after doing all their exploration within whatever AI platform, you really have maybe even a few seconds to shine.

So I think the website shifts from being really this platform to build your journey and help you along your purchase journey — to now you have this glimpse of: are you a trustworthy site? Do we trust this brand? Are we going to purchase? Or am I going to hop back over to ChatGPT, choose a different product, and go through the same process on someone else's website?

That's where I think the job of the website changes. And in terms of smaller organisations — as Will was saying, they're so niche that they are the experts on their products. And that's where they can shine, because AI platforms do reward that type of content. Whereas some of the bigger or even medium-size organisations might lean towards more boilerplate content, those small businesses can really get into the nitty-gritty details. I think they'll be heavily rewarded for that.

Patrick Hathaway

Yeah. I think I probably agree with all of that. Although the only part about trust — for it to be truly agentic, we'd have to trust that when you go to ChatGPT, it can go and buy you that thing. The trust is in the AI platform rather than the website, if you're not going to the website anymore. Which, I mean, I completely agree with you — I don't like the comparison bit either. And I agree with you, Will, that I like to go on the pages and I like to look at the stuff. But I do like the AI coming through and doing that consideration-set bit for me. But if we're going to get to the point where it's doing your shopping for you, the trust would have to be within the platform, and trust takes forever to develop.

Katelyn Geary

Yeah, absolutely. And I think that's evidenced by the fact that they rolled back their integration to purchase directly within the platform. And as Will mentioned, within Google Shopping, those abilities to purchase right there have been withdrawn — because I think there's something like under 50% confidence in that AI bot checkout, which again is changing human behaviour and trust beyond just structural. That's emotional at that point.

Will Critchlow

Something interesting — I don't know if either of you listen to the Hard Fork podcast. So this is two tech journalists getting very excited about the latest developments in all kinds of technology, but generally AI these days. And I was listening to the latest episode this morning, and they were talking about human taste. And whether — and when — these models are going to be good at writing, not necessarily in a "can you write me my business email" sense, but in a "can you write an amazing poem or novel or whatever else" sense. And interestingly, it's the one thing I've ever heard Sam Altman be underambitious about. He was like, you know, maybe in a few years it'll be able to make a poem that's like nearly as good as the worst poem in a good poet's body of work — but he's not going, it's going to crush everything, like he normally does.

And they were comparing that to some of the crazy stuff you used to get from the early GPTs — like GPT-2. You could ask it something like "I think I might be being scammed, what should I do about this email I've received?" and it would hallucinate some kind of wild, otherworldly journey that you might go on. And we don't get that anymore. All this creativity has been crushed out of the models.

There's something similar they were talking about with music. You can now actually get good answers from LLMs about what different songs are like. And it's interesting because the LLM hasn't listened to the song — it's only really consumed things people have said about the song.

And there's something very similar with products. Right? These models — they haven't been for a run in those shoes, or used that umbrella in the rain, or whatever it might be. So there's all kinds of stuff about eventually, you could imagine robots doing the testing or something. But that's a long way off. We're going to get to a very strange place, I think, where the model is superhuman in cognitive ability and still has no sensory perception. And so it's just consuming what like — again — reviews, things people have said. But how does it know that it's a person who said it rather than another robot? We're entering a very strange timeline. I don't have answers to these things, but that's been on my mind today.

Patrick Hathaway

I like it. And also — I don't want to live in a world where I'm reading poems written by bloody ChatGPT. Right. Let's save a bit of time for questions.

Patrick Hathaway

Okay, so let's move on to audience questions. So: love to hear your take on WebMCP and the universal ecommerce protocol. Who wants to take this one?

Will Critchlow

My take is that it's not the top priority for most sites right now. I would say focus on discovery, not on the checkout bit. There's a lot to be done with smoothing the checkout process — as I said, Apple Pay plus plus, making it so that whatever checkout somebody's using, it's very slick.

But I'm still a fan of the PDP, still a fan of the URL. And it's probably the biggest thing that I think the frontier models are doing wrong right now, actually. If you go into a search-like query in any of the frontier models — whether it's Gemini, ChatGPT, or Claude — you don't really get links back. And I think most of the time when you do a search-like thing, you want links. It's a bit of a mystery why none of them have realised this, particularly Google. You can get Gemini — or AI Mode, which is basically the same thing — to give you links if you ask for them. But the example I remember is recipes: if you ask any of them for a simple tomato soup recipe, it will give you some kind of average slot recipe that's the average of all the recipes it's read about. Which isn't what you want. The way to get a good result is to say: give me a list of links and sell me on why I should go and visit those links. And it'd be like: okay, if you're looking for something spicy, go check this one out — the reviews say this will blow your mind. If you're looking for something quick and easy to make, check out this one. That is what an AI search experience should look like, in my opinion. Links, personalised to me, with AI-curated reasons why this is the right link for my specific prompt.

Patrick Hathaway

I think you've both answered that you don't have a strong take on WebMCP at the moment — it's more of a wait-and-see, and focus your efforts elsewhere. Okay, let's go on to the next one.

So: are there any examples you can share where optimising for GEO had a positive impact on one channel, but a negative impact on another?

Will Critchlow

I can, kind of. I'm going to talk around it a little bit with some hints of where we're going. As you alluded to earlier, Patrick, we're going to have some case studies on this very soon. The best examples are still under NDA, and we're working on getting permission to talk about them publicly. Obviously, the people who are doing the most cutting-edge stuff here are not always wanting us to go on podcasts and roadshows and tell people all their secrets.

The kinds of things that are giving differentiated impact — positive in one channel, negative in the other — things like adding a lot more content to a page. Where you can imagine that the LLM is just like, yes, greedy, give me all that stuff. And Google is like, I've seen this before — there's too much on here now, maybe it's too duplicative of other pages. So that's just the kind of direction that you can imagine these things playing out in.

I think it's really interesting that LLMs don't really have a concept of duplicate content. They don't really have a concept of canonical URLs. And they have that in common — as you were saying earlier, Katelyn — with shopping, where all the product variations can all be one canonical URL. I saw a product feed recently where 85% of the entries in the feed were one single product that came in a huge number of variations. And only 15% of it was all the other products in the category. And that doesn't seem to be a problem for paid or other areas. But, yeah — fan of URLs, fan of canonical URLs. And I think the others should get on board.

Patrick Hathaway

I've got one here, Katelyn, that relates to something you were talking about earlier — cross-functional collaboration. What's actually working in practice, and how are you systematising customer insights from other teams into actionable personas for content and getting buy-in without adding operational overhead?

Katelyn Geary

Yeah, so I think — this is an interesting one, and there are a lot of different ways you can go. In terms of the cross-functional collaboration, it's early days there. In terms of leveraging information that other teams can contribute, we're still in the conceptualising phase. But I think one thing that we're going to look to use strategically is zero-party data — finding ways to get customers to tell us what they want, so that we can figure out what we need to share back.

On the technical side, I'm not quite sure what that implementation would look like. But in terms of getting buy-in, lean on the relationships that you already have. And I think a lot can be uncovered just by having conversations — you might make assumptions about other departments, or only know what they do based on how you interact with them. Open up conversations. They may have data points and information that you didn't even know they had. So it's about expectation setting, having those broader conversations to see where there can be shared data, and then actioning from there.

I don't have specific examples quite yet, but it's definitely something — as I've been talking about — that is one of those foundational things we'll be working on this year as we continue to build out our strategy.

Patrick Hathaway

Amazing. Well, I think that's a nice place to finish up — circular almost, because that's where we started as well. That's all we've got time for, unfortunately. But thank you so much to everyone watching, and for those fantastic questions, and huge thanks of course to both Will and Katelyn for so generously giving up their time and expertise. We will be emailing out the recording tomorrow, so anyone who's registered — if you missed the start, don't worry, you can catch up tomorrow. Before we go, let me just give you the quick lowdown on Sitebulb's next webinar, which is only two short weeks away on April 8. That will be the first of a three-part series on website migrations, so keep an eye out on your inboxes for details. Thanks again for watching, and see you on the next one.