Webinar: Brand Visibility - Optimizing for AI Citations

Our first webinar of 2026 was a corker! A panel discussion with two of SEO's heavyweights: Dan Petrovic and Jes Scholz.

Dan brought his trademark AI and technical expertise, while Jes - a self-confessed SEO Futurist - brought the straight marketing talk we all need in this era of AI search.

Webinar recording

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:

Hi everyone. Welcome to 2026 with our first webinar of the year. We've got two guests who we've been genuinely wanting to get on a webinar for ages. We've got Jes Scholz and Dan Petrovic, and I think we've got the perfect topic for them today as we're talking about brand visibility and optimising for AI citations. So hi to everyone watching. Please also welcome both Jes and Dan.

A bit of housekeeping before we start. Yes, we will send that recording out after the webinar is finished. So Americans, you can go back to bed now. And we will also save 15 minutes at the end for Q&A. So if you have any questions for our guests, please put them in the Q&A box. You can also upvote other people's questions in there. So if you don't have any questions of your own, you can go in there and upvote other people's questions. We generally start with the ones at the top who get the most upvotes. So if there's things that you want answered, go and do the upvoting as well. So thanks everyone for joining. Really, really excited about this one.

Now, our guests today are two of the sharpest follows on LinkedIn. If you aren't already following them, just go and do it now. And I will let our guests both introduce themselves and then we can get into the questions. So Dan and Jes, if you can just give me 30 seconds about who you are and what you're currently focused on. Jes, you go first for us.

Jes Scholz:

So hi everyone. I'm a growth marketing consultant. I specialise on big brands looking at smart content distribution and optimization of entity authority.

Patrick Hathaway:

Awesome. And Dan?

Dan Petrovic:

Dan Petrovic. I run an agency called DEJAN and I'm from Brisbane, Australia. I've been doing SEO for a very long time with recent interest in machine learning.

Patrick Hathaway:

Amazing. Okay, so what I want to start off with is some questions that explore what's going on with the search landscape and how it's shifting underneath our feet. So Jes, visibility used to mean ranking on Google. And if we zoom out a bit, what does visibility mean now? And where do you think brands are misunderstanding that shift?

Jes Scholz:

I think visibility, the digital definition is coming closer to the actual marketing definition. So for a long time, I think the online community kind of toy toyed at offline branding, at TV, at radio because it couldn't be tracked and we could be tracked. And now all of a sudden you have these AI platforms where you can't reliably track this is exactly what this user saw, and then they clicked and then they came to the website and then they converted. And we're getting a little bit scared. And so this visibility metric is coming back to our roots as real marketers understanding what users, who of our target audience is actually seeing, interacting and building brand salience with our brands online.

Patrick Hathaway:

So where does the AI search part of that fit into the broader ecosystem?

Jes Scholz:

I think it's just another new channel that we're tacking on. I'm not a big fan of this, "Oh, GEO is replacing SEO story." It is another channel. It's a new surface, but the core of SEO doesn't change. The core of your job doesn't change. What you're trying to achieve doesn't change, but you do need to factor in, yes, they are different algorithms. There are different tactics. There are adjustments that you need to make to your strategies because this is a surface which is high up the funnel, not that I like the definition of a funnel either.

But it's something that hits people very early in their journeys at their category entry points. And if your brand's not there, likely you're going to be out of the race before it even started. So it is important that your brand is visible there. It's important that you're aware of your presence on those platforms because it will impact downstream, whether that's into search or social or another marketing channel.

Patrick Hathaway:

I'm so pleased by the way that everybody nodded when you said that we don't have to start calling it GEO. That's definitely my place in that debate as well.

Okay, so Dan, if you had to explain what happens between a user typing a prompt and an AI citing a source, how would you describe what's actually happening there?

Dan Petrovic:

How many hours do I have? Let me try to come up with a balanced description. Okay. So the user is on, let's say AI mode in Google and they're typing in the prompt. Prompt contains the various dimensions and properties of a variety of facets of what the user actually wants. A little system in the background reformulates that into multiple search queries. These are synthetic search queries that are used for Google to Google itself.

So there might be like two, three, six different queries, and they all bring a unique set of results. Then some level of filtering happens and let's call it re-ranking because people are familiar with that sort of terminology these days. But basically we go from several hundred results, we go down to a few. And then those few results are then observed from their cached version of their page.

And the content is then scored against a query in each passage by using something like, I hope I'm not too geeky now, but like a model called cross-encoder. Think of it as a BERT or an embedder, embedding model that takes both the query and the target chunk, embeds them together in the embedding space, and then does the scoring. So basically plucks bits and pieces from each ranking results cached version and then formulates a grounding snippet for the model. It does that for every competing result.

So the prompt, the queries and the grounding snippets, generated using that scoring method, then go into the model's context. And then the model has an opinion about things, has its preconceptions, it has been influenced by what it sees. And you're, fingers crossed, hoping that the model saw what it needs to see from your page because if your page is thousand words, the model might see a couple of hundreds out of that. And you're hoping for the best that the pre-processing pipeline sent the right things to the model That speaks to your product, that speaks to your brand, that conveys the message properly.

And then the model synthesises the results and attaches the citation to the grounding source. And then the user sees the thing and then they can go back and forth. And if they do that, the whole process repeats again and again and again, and the multiple cycles of that same process happen over and over. And then we end up with a chat session where user solves the problem, gets the thing, whatever, books the trip and abandons the session. So it's quite a bit of difference in-

Patrick Hathaway:

[inaudible 00:08:15] yeah.

Dan Petrovic:

Yeah. Yeah. So we're all saying, "GEO, boo," but we're not saying things the same. There is a camp in SEO that people are saying, "Oh, you just need to do SEO and that's all you need to do." I'm very much against the rebrand of our industry, but I'm not in a denialist camp saying that there's nothing new to do. There was much to do new when mobile first came onboard and other things. You always have to keep up. That's as concise as I could pack it, but it gives a real grounded, I just said a grounded explanation of the whole process. So hopefully that makes sense. If it didn't, then you can follow up with me and ask me.

Patrick Hathaway:

Well, yeah, there's loads more questions that will help, I think, give a broader picture as well of what's going on. Yeah, I didn't like the sound of crossing your fingers. So we definitely later will get onto things that we can do to maybe help the likelihood that you might end up in there as well.

Okay, so there is like a growing argument now that a click to your website isn't inherently more valuable than someone encountering your brand in an AI response or engaging with you on another platform. So do you agree with this? And if you do, what does that change in terms of how we ascribe value to different channels?

Dan Petrovic:

I like clicks. I like clicks. I think Pedro Dias tweeted or posted on LinkedIn today, he's like SEOs are making themselves believe and really feel like, "Oh, we don't need clicks anymore." We need clicks. But I'm super happy when I get a brand mention and I'll be super happy when next year agent does the transaction for my client in the background and there was no click or page visit. Does that make sense? I'm adapting to the new paradigm, but I appreciate and I want the clicks and I'll continue measuring them.

Patrick Hathaway:

Jes, you were saying before, thinking holistically and the AI search is just another surface, what's your take on this?

Jes Scholz:

I would say clicks is an interesting metric to follow, but clicks is not a KPI of SEO and should never have been a KPI of SEO. At the end of the day, your CMO doesn't care about clicks. They care about market share, they care about revenues, they care about conversions, they care about brand salience. And so SEO needs to tie into those sorts of things if you want to get the resources and budget that you need to be able to tackle all the things that are now under the SEO umbrella. If you are measuring yourself based on clicks, then you're going to be measured as a performance channel, you're in direct competition with CPC. They're going to kick your ass and you're not going to get the budget and resources that you need.

So yes, look at clicks. Yes, they should be going up in an ideal world to the right type of pages with the right intent. But I can get a million clicks to a page by offering something that the market wants, but is that going to help my brand salience? If the answer is no, then I shouldn't be producing that sort of content because the clicks don't help the ultimate goals of the business. And I think we always need to bear that in mind. And that's why for me it's not a KPI.

Patrick Hathaway:

I think that's really interesting, the point about don't anchor yourself to this metric. If we know, we can all see what's coming, clicks are probably going to go down or at least they might stay level because you're doing lots of other activity, but it doesn't make sense to do it that way.

All right. So Dan, I just want to talk about some of the stuff that you've been writing and talking about. And it goes back to some of the stuff you were telling us about earlier and you mentioned obviously the bias of the model. So two of the ideas that keep coming up are primary bias and selection rate. So again, a quick version, but can you explain what they mean and why they matter?

Dan Petrovic:

Yeah. Selection rate is your new click-through rate because the models don't click. There's no mouse, there's no tap. They select. That's the terminology I use. I couldn't come up with anything better. So selection rate is determined by a variety of factors, in fact. There is still position bias. They do care about position. They do care about a number of things. And some of those things are like, oh, brand recognition.

So if you have five results in its grounding context for the model, any model, GPT, Gemini, and so on, the model might pick one and prefer that and make a final recommendation to the user because it's familiar with that brand and it's unfamiliar with the others. Or it's familiar with all five brands, but finds one particularly relevant for the thing that the user is looking for. And so I'm using, not familiarity in like human terms, but models don't have that type of thing. They're just statistical machines and they snap to their probabilities in the setup that they're allowed.

So for example, if we look at parameters like temperature, top-K, top-P, so how wide and how deep and with how much randomness can a model sample in its probability space when determining what next thing to say in the chain of things that it wants to say, token prediction. And in that model geometry, if it finds something, "Ooh, that's nice," it snaps to that. And you want to be the brand that the model snaps to when it's associating things with certain entity product and so on and so on, a problem.

So that stems out of, first of all, model pre-training, where model just learns the language and then fine-tuning to align it with human values, reinforcement learning with human feedback and so on and so on. But eventually the model is baked and released to the world with a worldview. And if you probe the model enough times, you will start to paint the picture of where its biases are and how it aligns.

And so that's what I refer to as primary bias of the model. So this is the raw, naked version of the model without internet plugged into it. Like you could have the model running on your machine and you ask it a question, whatever it answers, that's its raw, unaltered opinion by other factors like search results. So primary bias influences search results selection basically.

Patrick Hathaway:

Okay. So that helps us understand how AI systems form opinions about brands, I guess. So all right, let's now talk then about how we can help understand what things we can do to help influence those opinions. So right, I've got a question here. Jes, I want you to start with this one. So if a team only had budget to focus on one thing, optimising individual content pieces or building brand distribution presence, which should they prioritise and why?

Jes Scholz:

I would say brand distribution presence because you could put 50 hours into one piece and then it completely flops while if you put 50 hours into distribution, of course you have to have good content in order to distribute. So one requires the other. But you're going to get more impact by taking a selection of really good, focused content that is distribution worthy and focus on getting that in front of as much of your target audience as possible. Rather than focus on producing many pieces of content that live on your website, but you don't have enough people knowing your brand and coming to your website in order to enjoy the benefits of those pieces.

Patrick Hathaway:

So is this like, Dan, where you were saying that the models, obviously they've got their pre-training, then the model gets baked and they've got their biases? Is what Jes is talking about how you can influence that by people are more aware of you, the AI, the training models are more aware of you, then you're more likely to be in that initial bias?

Dan Petrovic:

It's really, really hard to do that. It's a monumental task. It took us six months of hard work. It's not link building, but brand to entity association, let's call it. Six months of hard work and maybe a bit of luck. We had one client based in Germany, which services a worldwide market for sports equipment, and the models were just not going to recommend them because they're German. So they will recommend only when the customer is German. But if the customer is from Australia, they consider that brand, they're like, "Oh no, that's not relevant."

We did two things. We did the hard yakka, let's build up the network of concepts and alliances. Maybe I can bring it a little bit closer to what I actually did. So there was adidas, Nike, Under Armour, and then our client, not even there, it was just like at the bottom of the list. And so we were piggybacking, constantly mentioning our client alongside the big boys and trying to bring them closer in the status. So that's one thing we did.

Another thing is we were Americanizing the client. So we're like, America, US, New York, this, that. So we were seeing content that aligns them with different countries. What else did we do? Oh, brand to entity associations. So we would take the brand and then put the things we want models to pick up on. That took six months and only worked because Google released, we went from Gemini 2.0 to Gemini 2.5. If it wasn't the model release, nothing would have happened. It's like we are in the early days of SEO, we need to wait for the Google update to see the results. So it's like that because-

Patrick Hathaway:

Yeah, that's the Google dance.

Dan Petrovic:

Yeah, model's frozen in time. So you can do all this hard work, you don't see anything until they release a new version or a fine tune of the existing model. Something happened with Anthropic's Claude Opus 4.5 a couple of days ago. It just went a couple of IQs down. They did something to it, but they didn't announce a new version. I think it was just like a little tweak.

I felt like I had something else important to say. Oh, yes. So that's just interpretative layer on top of search. So this is like the biases that shape the selection, but nothing happens without SEO. If you're so arrogant and thinking, "Oh my brand will be the top of mind when the model speaks," you have to be like a top, a major, major brand. If you're a small brand and you're hoping to build up enough clout within the model's training data, that top of the mind, that's not going to happen. That's like really, really difficult.

So what I'm saying is SEO is number one because to be in the mix, to be even in the consideration by the model, you need to be ranking. Without the rankings, nothing's going to happen. You need to be in the consideration. So when the model is ungrounded, we didn't see any results in six months. When the model was grounded, it took us one week, basic localization. We removed the GmbH and we put LLC and we removed the whatever German phone number, and we put +1 number and the local address and this and that.

One thing that's really nice with Google and Gemini and grounding is that they ground with full HTML of the page when you do a URL grounding. And when they do the cache for the ranking purposes, they see markup, bold and this. Not the full, they don't see schema and other things, but they see a lot of formatting. So they actually picked up the stuff we had updated on the client page and then just like that, they was starting to be recommended.

So that's nice. That's a really easy and simple case where it worked really well with normal SEO and it took a long time with new, it's called AI SEO to differentiate. And I think we got lucky. I need more case studies to show that this can be repeated and that it's reproducible. So that's what we're working on at the moment.

Patrick Hathaway:

Yeah. It is a really interesting case study as well. And I suppose it's also a specific issue that they were having that there was a way to solve for.

Dan Petrovic:

It's an example of a bias, exactly. The model was biassed against the brand, but it could be the opposite. Model could be biassed towards your brand when it comes to certain things. I think it's worth understanding both.

Patrick Hathaway:

Yeah. Okay. So then, Jes, you were talking before about different platforms and trying to be as in many places at once. Do you think that then it makes more sense to invest in your presence across platforms than in perfecting your website?

Jes Scholz:

You need to be investing into the relevant platforms. And I think this is where people are going awry, where we're repeating the mistakes we made of the past when we figured out links influenced SEO 15 years ago and everybody went, "Okay, build as many links as possible no matter where. Yay, links." Don't make that same mistake again, learn from the past. It's not about getting as many mentions. It's not about any site corroborating who you are. It's about a relevant site from Google's perspective of their knowledge graph corroborating. It doesn't matter if I say I'm a hula-hoop champion and some random person says the same thing. It would matter if the Olympic committee said that I was a hula-hoop champion. So who and where and relationships, all of that really matters.

So don't go for quantity, go for relevancy and quality of that relevancy. And you'll be able to identify that by knowing your industry well. You can do samples with prompt tracking to see what types of sites are actually showing up with a lot of share of voice across the AI tools. And then, do you have a presence on those sites? What is that presence and is it reflective of your current brand? Because a lot of the time these models are going back a long time in history. They're surfacing things that traditional search would not or Discover would not.

And so we're going back to a directory listing that someone else, your predecessor's predecessor's predecessor made 10 years ago with the incorrect brand name, with the incorrect logo, with the incorrect positioning. You need to find all of these and you need to update them. And the best way to do that is do these focus group samples of your industry, your business model, your topics that are important, all of these category entry points. Find all of those relevant citation sources and go and target those at the starting point.

Patrick Hathaway:

By the way, so are you actually a hula-hoop champion?

Jes Scholz:

No, I am not an actual hula-hoop champion. I'm very sorry.

Patrick Hathaway:

That reminds me of, remember when we had all the unnatural link penalties and you're dealing with the client sites or your own sites where you had to go back and right, what is the footprint? And this is hurting us. We need to go and either delete it or remove it or change it, whatever, take the unnatural anchor text and make it more natural. It's a little bit like that, I suppose. Yeah, that's really interesting tactical stuff you can do to approach the holistic brand message, you were saying before.

Okay, Dan, I've got one for you here that's again about some of the research that you shared that suggests that shorter, denser content often gets more representation in Gemini's AI grounding than long form pages. So if that's true then, what does that mean for how teams should be thinking about their content strategy?

Dan Petrovic:

I suppose it's maybe a little bit misleading to frame it in such a way. I did publish the work, but a lot of people misinterpreted what I was saying.

Patrick Hathaway:

Including me?

Dan Petrovic:

I mean, it is a question coming from the audience, I suppose, unless it was your question, was it? Look-

Patrick Hathaway:

No, yeah. At the moment, these are all my questions. And then we'll have a few minutes-

Dan Petrovic:

You have a large piece of content, Google has the grounding. I didn't name it this way, but people took what I published and then went with the word budget. I guess it's familiar. So grounding budget, let's call it, let's roll with it. So if you have a 1,000 word article, a larger portion of your article will go into grounding. If you have a 10,000 word article, a smaller portion of that article will go into grounding because Google has a fixed number of tokens it's allowed to ground each ranking result with.

And so if you have a tiny little article, 100% of it will go into grounding. But I'm not advising that people necessarily write ultra short things because Google does have a fairly robust snippet building system. They use extractive summarization to pluck the most relevant bits to the prompt or to the query.

So there's been discussion around this and they said, "If you have a very large content, we'll cluster it all together." That's not really the solution because I've seen the behaviour of Google's pre-processing pipeline. I've seen them pluck from random parts of the page. So one thing is for you to do nothing and just measure what's happening in the AI search to see if you are misrepresented, and act on it only if you have a problem. If Google speaks to your brand and product and represents you well, then there's no problem. If you're happy with the way your text is structured and there's no problem with it, I wouldn't go and change everything and summarise and condense everything.

Now, at the same time, I did a study about how users read, and so I tricked everybody basically. I had a script in the background that records your mouse movements and clicks and scrolls and everything. It was just a simple JavaScript, still up there. And I had a survey on and I said, "Are you a reader or are you a skimmer?" And people would declare themselves and then proceed to read that page that is where the poll itself is. And guess what? They all spend about two minutes.

Somebody would say, "I'm a reader. I read everything word for word." No, they don't. They said they do. I call them aspirational readers. They feel like I read everything, but they don't. And then there's the other people who say, "Oh, well, it depends." So classic SEO saying depends. But it's true, it depends. So the content piece has to sell itself to the reader for the reader to commit their time to read the full thing. That's common sense. And so is your content piece fit for that?

There's a beautiful harmony or not even, symmetry between how AIs and humans land on content and perceive it. They look for cues. They look for attention patterns, quick summaries, top, bottom, middle gets lost a little bit. And so I think there's utility in providing information-rich, condensed pieces of content that convey information. If your content is 90% fluff, 10% ideas, then you've got a problem with both AI and humans.

In fact, I have a model called Substance on Hugging Face, and there's a tool, I think, somewhere on my tools list, basically a content substance classifier. It tells you if your content is largely fluff or substance and which sentences. So that'll be fun to put your articles through for both users and AIs.

Patrick Hathaway:

So whenever I write anything for Sitebulb and if it goes through Jojo editing it, she will delete large swaths, "You're writing too much. It's too verbose." So yeah, she could probably make use of that tool to take all my writing down. Love that.

Okay, so I want to talk about this idea about brand building again, Jes. So there's this longstanding idea that 60% of your marketing effort should go towards long-term brand building and 40% towards short-term activation. Does AI change that balance?

Jes Scholz:

Not in the slightest. That's been a well-established marketing principle since the Mad Men days. It's not going to change just because AI came onto the scene, especially because AI in many ways mimics humans because it's learned from us. It's influenced by our biases or at least what we've put out onto the internet. And again, it comes back to if you see SEO as a performance channel, you don't understand SEO. If you think your job is to drive X number of conversions this month, you are shooting yourself in the foot. You had to have been doing that brand building for years. And the brands who have understood this are now reaping the rewards that AI is coming out.

And it's not looking about what you are saying about yourself on your website. It's looking if what you're saying about yourself on your website is agreed to by influential people in the industry. And that's always coming back to, do you have something interesting to say? Are you contributing to the conversation in a meaningful way that other people are going to share, like, comment, engage, interact, recommend? If the answer to that is no, you need to really look at your brand, your product strategy, what do you have to contribute to your industry?

And I understand that that's a bigger conversation than SEO alone. It's even a bigger conversation than marketing alone. But if your brand hasn't been having those conversations, it doesn't matter if you hire an AI savant like Dan, he's not going to be able to make your product better. Product ultimately needs to actually be able to stand on its own two feet. And that then can be supported by a good brand, and that can then be supported by short-term activation campaigns, which are still important. You should have it as part of your mix, but it is that minority. It is that 40%.

Patrick Hathaway:

And this brand building, is this a way, almost like the earliest form of AI optimization in order to affect these biases of these models?

Jes Scholz:

It comes back to what Dan was saying before about topical association, entity association. So in a user's mind, we have category entry points. There are certain triggers that make us think of buying situations that make us then think of brands in those buying situations. So maybe I'm out on the street and I see a cute top and I'm like, "Oh, I haven't been shopping in a while. I want to buy a cute top. Who are my favourite brands? Okay, now I'm going to quickly pull up on my mobile phone and do that."

If I'm asking an AI model that, it's going to behave a similar way, what brands does it know that are associated with an entity of cute tops? And so you need to understand, well, what are these not keywords, not topics in the way we've understood it from a Google Discover point of view, but these entry points. What am I going to be asking the model for that is repeatable across a large swath of humans that I want my brand to then show up for either as a mention or a citation. It doesn't really matter as long as users can follow the intent through.

And that's really what branding is about. It's about understanding what are these category entry points, how are they reflected on various surfaces from your AI, your traditional search, your social search, your video search, your map search? And then making sure that you're monitoring that share of voice and increasing that share of voice at those points.

Patrick Hathaway:

With the difference between B2B and B2C on this stuff, we've got a Claude subscription, and I can see Claude learning more about us all the time, about who we are, what tools we already use, how the team's made up and all of this stuff. And the personalization that you might have at a B2C level, can you see it being just a completely different kettle of fish at B2B because personalization is not going to be specific to an individual anymore, it's going to be potentially more related to the business? Or is that just not how it works?

Jes Scholz:

I think we need to recognise that in both B2B and B2C at any point in time, the vast majority of the total addressable market is out of market. We're talking upwards of 70% are out of market. Even if you think of something, you think you buy toothpaste all the time, you probably buy toothpaste once, maybe twice a year. And so yeah, B2B buying-

Patrick Hathaway:

Not the way my kids get through it. There's a different tooth toothpaste in ... They just carry them around. Why do you need to open another one? What? There's still loads left in that one.

Jes Scholz:

I have the opposite [inaudible 00:38:04] fighting to get the toothpaste on the toothbrush.

Patrick Hathaway:

Oh, yeah.

Jes Scholz:

There's these things, let's just think about Coca-Cola. I buy Coca-Cola maybe once a year because I'm [inaudible 00:38:19] I would still consider Coke one of the most prominent soft drinks. There's not that huge difference between how often we buy B2B and B2C in terms of our consideration sets. When we actually sit down and think, "What do I want to buy? Why do I want to buy it?" And humans dedicate much less time to those discussion points and decision points than marketers think they do.

So we're designing these really intricate websites and perfect messaging and then A/B testing that messaging and making sure we're changing it over every month and A/B testing a new one when the vast majority of your audience won't see your six or seven A/B tests because they come to the website once a year, do their purchase and off they go.

So really take it up to these base marketing principles, understand how humans actually interact with brands, how that then is reflected in their interaction with AI surfaces. And that gives you a little bit more peace because as Dan was saying before, if you're doing all of this work, it might not reflect until the model updates. That might be once or twice a year. That's fine because probably the vast majority of your audience is only going to be looking at it once or twice a year.

Not to say don't focus on it. If you're only starting now, you're already behind the times. You should have really already started on treating AI as an important surface two years ago. But you've got time to catch up, especially because there are a lot of now documented tactics that are relatively proven to move the needle in the right direction.

Patrick Hathaway:

Okay. So that's a really good lead in then. So what are the tactics that would help brands prepare with where the industry's going? And I suppose as well, how urgent is it to start either preparing or getting into action straight away?

Dan Petrovic:

I think a lot of brands that I've encountered and worked with have this notion that AI is going to steal their content and run it for training and it's like they're feeling ripped off and this and that. And I'm like, "Okay, drop everything, go into Cloudflare and switch off that thingy that was toggled on to block all AI bots. Allow training, allow theft, allow content use, get into the model's mind as aggressively as quickly as you can." Of course, there's rogue bots there, like some random labs experimenting, scraping quite aggressively. You want to block all the rogue ones, but like why would you block Google's AI agent or even for training purposes or OpenAI. You want them to learn about you and your brain and to etch into its mind about what to do.

So I want to go back to basics with SEO, don't block the bots. Or structure your content and pages and work your product descriptions, make it obvious. Don't hide things, don't move important things into tabs and accordions, especially if they don't render properly. Allow page content fetching, make sure that it's server side rendered. Just do some SEO and you'll do like a good chunk already good for AI because AI is an interpretive layer on top of search. I think that's a really good start for people to just start preparing by fixing the problems that they have.

Let's say you're a brand, you've got everything perfect, you're not blocking the bots, you've got everything pretty well set up. The things you want to start thinking about now is, are you on top of the latest trends and technologies and protocols that are currently being forged, agent to agent protocols, agent payment protocols, product feeds? Like if OpenAI says, "Okay, we now have a marketplace," are you on top of that? So being ready for the agentic web.

I think it was last year I did a presentation in Amsterdam and said agentic web is 2027. I didn't say '26, because it felt like coming up too soon. But basically next year, we are going to start seeing transactions happening in the background, no visits to the website. Websites will be optional surfaces, interfaces. And so basically we're looking at agent to agent transactions or direct feed of the product.

So what do we do? What does an SEO do in the world where everything happens in the background, nobody visits the page? Well, it's exactly the stuff that we were discussing earlier? If the model is presented with a variety of different options, how is your content represented by the pre-processing pipeline to the bot, to the AI? And how does that AI experience that grounding context based on what it's learned about you? So those are the types of things I would start thinking and being technically prepared for what's coming. Of course, there's an element of having great content and well-structured, but I'm assuming you're already doing that because of people and SEO and stuff.

Patrick Hathaway:

Yeah, I love that being ready. Jes-

Jes Scholz:

I think to-

Patrick Hathaway:

Go ahead.

Jes Scholz:

To build on what Dan's saying, I think don't fall into the trap of, I have a product feed, so I am ready. So many teams that I start consulting with are like, "Yeah, but we've got a product feed already." Like, "Great. How many attributes have you got filled in there? How accurate are those attributes?" And when they start looking at that, that's where things are falling apart. Because I'm not going to AI and I'm asking for, I don't know, what's a common example? I'm not going for a black dress. I'm looking for a black dress, which is in a size six, which is A line and has jewel, I don't know, whatever I'm looking for. If all you have in your product feed is the category of dress and the colour of black, or you call it noir because you want it to be fancy on the website, guess what? You are not going to match in AI search.

So you have to be so much deeper in the categorization of whatever it is that is your content, whether that's products, whether that's listings, whether that's the depth of your article with new ideas, and how you're structuring that so that it can be chopped up into a feed and fed to these platforms effectively and queried by these platforms effectively, multitudes of them.

So really look at enriching your content and enriching it more than your competitors. If they're a property portal and they're covering off, it's a rental, five bed for this price, then you go in and you talk, "Yeah, but it's pet friendly and it's furnished and it's serviced and it's on the fourth floor with northern light." I don't know. Go deep into these details, that's where the competitive advantage is. The competitive advantage is in the database, not the website so much.

Patrick Hathaway:

Yeah, in the data. Yeah, I love that. Awesome. Well, guys, that's a really great way to finish my questions, but we've got loads of questions from the audience. So I'm just going to put them up in the order that they came in. So if there's a question that anyone really desperately wants to get answered, you have to get the upvotes for it. So go in there and upvote other people's questions. I'm just going to refresh it now. And we probably won't get to do all of them, but we'll do our best. So let's just get going. I'm going to show them on stage. So Dan, perhaps if you've got some links to some of your stuff you can maybe drop in the chat, resources that will allow folks to deep dive on how an agent takes prompts, queries and grounding sources.

Dan Petrovic:

Oh, basically my blog. It's all I write about lately. On LinkedIn, one of my recent posts, I actually put together a bit of a workflow, almost like an infographic piece done by Nana Banana, which kind of summarises. I put a large complex tool code and it just gave me an infographic, which was quite nice. Saved me a lot of work because my guys were asking for that diagram as well. So I was like, "Okay, let's do it."

One thing I wanted to add to excellent points that Jes has made about product completeness before I forget, because I think it's really important, looking forward to the future. So it's just going back to her point. Gemini is multimodal and they have agentic vision now that came today. So basically they'll be evaluating the content of your imagery, product imagery as well, and in the future, no doubt, videos as well.

Google has the ability to generate vector embeddings for every frame of the video and send you to the exact moment in the video that the user describes 24 frames per second, every frame has a vector embedding. And then when you say, "Show me the part in the video when she unbuttons the strap on the dress. I want to see how that works." Boom, straight to that point. That's coming.

Patrick Hathaway:

That's outrageous.

Dan Petrovic:

Yeah. So just think about quality of your media as being part of the content, not just text.

Patrick Hathaway:

Awesome. That's kind of scary it can do that.

Dan Petrovic:

Yeah, it can do that.

Patrick Hathaway:

Okay, right. So this one's for you, Jes, then. So when doing brand distribution, what content should be focused on? When you say brand distribution is more important, what specifically should be distributed?

Jes Scholz:

So what should be distributed is whatever you have to say that is interesting for your audience that adds to the conversation, this concept of information gain, which doesn't need to be a brand new thing. It could just be an interesting opinion on something that already exists. It can be a well-written piece. Often, I'm quite happy to read an idea or rewatch an idea that I already know just because it's elegantly constructed or beautifully presented and it just refreshes. It's the same reason people like to reread books or rewatch movies. We don't mind ingesting the same information and refreshing it as long as it's done in a beautiful way.

So information gain, yes, it can be, here's brand new information no one's ever seen before. It can also be, here is a really efficient roundup of all of the key points. Here is an opinion on something, my personal opinion, and I'm an expert, and that's why you want to hear from me. Think of a way to contribute to the conversation.

In terms of distribution points, I think it's important to have a wide variety of formats. So yes, you'll have your own articles on your blog and you're going to post those articles to the social networks probably automated and it's probably going to be really ugly. Think about how can I turn that into a story? How can I turn that into a reel? How can I turn that into long form YouTube? How can I get that in quotes on other platforms? How can I get that into audio formats? If you have something interesting and important to say, say it in multiple formats, say it in multiple places.

If you're struggling, if you've written something and then you turn it into a story and that story doesn't go anywhere, it may be that your social team needs a little bit of training on the algorithm. It may be that it just wasn't interesting to start with. So have that honest conversation with yourself. I think we've been spoiled for a long time that average was good enough, and that is still the case on many platforms, sadly, but it's getting harder and harder because it's getting noisier and noisier.

And as traditional SEO clicks are set to continue to decline, there will be more noise and more competition on all of these other places that you can get digital visibility. And so if you're not strong on distribution of your content, if you don't have automations in place, if you don't have a good database that's hooked up to APIs that can transform with the help of AI and distribute those ideas efficiently and effectively, you need to go and look at that problem with your SEO.

Patrick Hathaway:

Okay, that was a really good answer. This is, I think, a bit of a similar question, but see if either of you have got anymore to add here. So what's one thing you think brands should be asking of their digital PR marketing teams now to improve LLM visibility? Who wants to take this?

Dan Petrovic:

We're currently working on this, so maybe I can provide something relevant. Basically, what we're trying to extract as a step zero from our clients every time we start doing this is to give us a list of entities that they care about. And when I say entities, like top level things that they want to be visible for, so we can then explode them into sub queries, into prompts, develop prompts out of that, so we can start experimenting and probing. And it comes in a variety of shapes. Sometimes it's a hierarchy of products, sometimes it's a flat list and so on and so on, but we really need something, that as a base, to develop out other things.

What do we do with that? So once you have the list of things you want to be visible for, then you do model probing at scale, not prompting, but model probing. So I'm actually very much against like you load up 100 prompts or 1,000 prompts or a million prompts for your brand, and then you probe them every day and you're looking for prompt volumes and rankings for those prompts. Talk about meaningless metrics. That's worse than clicks, I think Jes will agree.

So what I do instead is I do associative probing, first of all. So entity, what we were talking earlier. I can see people are making fun of me for using a dress in the comments. I am actively looking for a dress for my wife. We're going to Dubai. We want to do a photo shoot with a flowing dress, something, something. I don't know. And it's a frustration. I didn't know how it works. I needed to spin the dress around and see how it works and this and that.

So let's say you're probing the model and you want to understand whether it will recommend your brand if somebody was looking for the strapless black dress with a flowing thing that's good for desert photography. That's maybe a bit too specific. And what I do, I don't want the model to say anything, blurb. I don't want a full spiel. I want yes or no. And so for every entity, I explode into query fan-outs. For every fan-out, I develop out the prompt. So for each fan-out, I say, "Would you recommend this brand for that product? Yes? No?" And I repeat that 10 or 100 times and I get the statistical value out of that 63%.

So I'd repeat that for every exploded synthetic query that stems out of the core entity that's associated with the brand. So I refer to this process as relevance probing. So first part is associative probing where you say, DEJAN AI, what do you associate with that? And then it lists things that it associates with the brand. And then you say, for that thing that you said you associate with the brand, if somebody was looking for such and such, would you recommend yes or no? And we do that at scale for thousands of different things. So that's small, distilled, precise, accurate counts of models' attitudes. I even forgot, what was the original question? Can you go back to the ...

Patrick Hathaway:

Have you got a tool for that probing?

Dan Petrovic:

Yes. Yes, we have. I think it's part of AI Rank and there is one tool that we use internally for that. Maybe I should make it a public tool. I don't know. But AI Rank definitely has that. You just have to ping me to enable that feature inside it because not everyone knew what to do with that. They were just like poking around, not really knowing how to use it properly. If somebody's really interested to use that, I can give them access. It's a powerful thing to do and gives a lot of useful insight.

Patrick Hathaway:

Well, as I said before, everyone needs to be following these two on LinkedIn, so then you can hassle Dan then about that. So right, this was the question that we-

Dan Petrovic:

Oh, yeah. So it is tangentially relevant. So things you care about, before you speak to your PR and marketing teams, that's what I'm getting at actually. You need to understand where are the weak spots in the model's perception of your brand, because why would you be flogging things that you're already doing well for? So you need to identify the weaknesses in the model's perception when it comes to your entities. And so once you map those out, then that becomes guidance for your PR team. Do stuff around these areas, focus around these products and services. The other ones are good. Or prioritise, start with the weakest ones first.

Actually I made this advice a couple of months ago and the client was like, "Actually, we don't care about that." So yeah, you then have to factor in both, is this important to our business or is it just like a thing we don't really care about? So that's how we direct the digital PR teams or PR teams in general. In fact, everybody, PPC, just email marketing, everyone should be aligned on that front to bolster the things that we need to strengthen, but are currently really doing really poorly when it comes to model's perception.

Patrick Hathaway:

Would you also go the other way and ask questions what we don't want to be associated with?

Dan Petrovic:

Yeah, I want to do that, but that requires me to open up bigger outputs. I suppose we do that. We do that in, there's a tool called Reviews, reviews.dejan.ai, also free. So what that one does is it does is artificial reviews sentiment mining. Basically I say, write 10 reviews for this brand and it'll give positive, negative, whatever. And then you look for things that it says about your brand that you don't want it to say, especially in the negative sentiment. So that helps, but it's not exactly what you were saying. I think it would be useful to do maybe named entity recognition.

Oh, I'm glad you brought this up. There's a model called GLiNER, G-L-I-N-E-R, N-E-R stands for named entity recognition. So this model, it's on Hugging Face, it's free. You can get your geek to build you a little app with it, or you can vibe code it yourself. So what it'll do, it'll process the text, and let's say text is the output of the model, and you can give it arbitrary classification labels. You can pretty much put any label and description and the model will fit to that rather than being a classifier on a named entity extraction like on arbitrary nature, like money, person, company, blah, blah, blah.

So basically you can do that to extract any entities that you don't want to be associated with using natural language processing. So this is a deep learning model, but it's very light and it'll work on your laptop. So yeah, that would be amazing for extraction of those things. But I don't have a tool that does what you said it should do. And I think somebody should build that. That was really useful.

Patrick Hathaway:

Someone in the audience, you can build that. Guys, we're just at time, but I've got one more question that I wanted to put in. So hopefully people can hang around for a few more minutes and Jojo won't tell me off. But anyway, we're going to do it anyway.

So Jes, finish it off on this one then. A few times today, you've said, "Don't track that metric, don't care about this." So I've got an idea of where you're going to go with this, but I think obviously the question's come up. So what metrics do you think are the most beneficial for reporting LLM brand visibility specifically to senior leadership C-suite?

Jes Scholz:

So I report on share of voice and I report that across all channels, whether that is your traditional search, whether that is your Google Discover, whether that is your AI surfaces, whether that is your social media services. Share of voice is one that unifies and your traditional marketers, if you're looking at Les Binet, for example, he's proof that there's a correlation between share of search and market share. That share of search is a leading indicator for market share. I don't see that not translating to AI surfaces in a few years as well.

So that share of voice is highly likely to become the leading indicator for market share because it is or will become a key category entry point. So when I'm coming new into the category and I don't know what I want to buy, who I want to buy from, the first thing I'm going to do is go and ask my Gemini or my ChatGPT. And if you're not recommended there, you're already out of the consideration set. So make sure that you are mentioned in those queries.

And I agree with Dan, taking big prompts and running them and then believing the AI output on that particular day is the truth and everybody gets the same, like no, that's not how this works. But I do use LLM prompting similar to how we use focus groups. I take a sample of attempted at as unbiased as possible and I see what that sample says. And then I go off and I action that data and then I go back and re-ask a new sample, okay, now what has changed?

Patrick Hathaway:

Love it. I love that answer. And what a great way to end. It's a shame we do have to end because honestly could talk to you guys for ... I mean, it is nighttime for you, right? It's probably like midnight. So thank you both for fitting into a British schedule. We really, really appreciate it. I think there's so much value from this. So yeah, really, really huge thanks to you both and to everyone who's watching and the amazing questions as always. We will, again, be emailing out the recording. So if you missed some of this, you can get it either today or tomorrow that will come out.

I very, very quickly, before we go, just want to talk about the next webinar. That's on 18th of Feb. It will be back at the normal time. We have a 4:00 PM GMT and we have Andrew Shotland telling us all about Reddit for SEO. And specifically on the topic of Reddit, Jojo has just recently started a sort of JavaScript SEO subreddit. She's running an AMA tomorrow there with the JavaScript legend, Sam Torres, who's done all our training along with Tory Grey. So that's on the subject of JavaScript SEO. It's a pretty new subreddit and it's our first AMA there. So please come along with your questions for Sam and that would really help us out and you can get a load of value obviously if you've done the course. So yeah, go ahead and see us there and hopefully see you on the next webinar as well. Thank you everybody for watching. Good night, good today to all of you.