Apologies to Nik Ranger for me goofing on my video recording! Thank you for carrying this episode on your shoulders, Nik!
In this episode of Jolly SEO’s TL;DR, I sit down with Nik Ranger, Senior SEO Consultant at Dejan Marketing, to unpack a concept that quietly underpins how AI search actually works: query fan-out. What started as a mind-blowing taxi ride conversation after the Shenzhen SEO Conference turns into a practical breakdown of how large language models deconstruct long, conversational prompts into dozens - or even millions - of sub-queries. Nik explains this in a way that’s accessible even if you’re not deeply technical, using real-world analogies to show how AI systems gather, ground, and synthesize information in real time.
From there, we go deeper into what this means for modern SEO strategy. Nik walks through how Dejan prioritizes content creation when faced with massive query surfaces, why traditional keyword tools fall short in an AI-first world, and how techniques like query demand estimation help teams decide what actually deserves attention. We also explore the non-technical realities of client work - balancing data with business priorities - and why one principle remains non-negotiable for AI visibility: if your content isn’t in the initial HTML, you can’t rely on AI systems to use it. If you’re thinking seriously about SEO in 2025 and beyond, this episode reframes how content, infrastructure, and strategy need to work together.
🔗 Links & Resources Nik Ranger (LinkedIn) - Connect with Nik for deep technical SEO and AI visibility insights https://linkedin.com/in/nik-ranger
Dejan Marketing - Enterprise-focused SEO consultancy specializing in advanced technical strategy https://dejanmarketing.com/
One thing that I want to make abundantly clear and I and I will probably repeat this in our conversation. Um, but for AI visibility, the content must be Hey everybody, welcome back to another episode of Jolly SEO's TL;DR. This week I am so excited. Nick Ranger from Dejon Marketing, you are here the senior SEO consultant. Thank you so much for joining us. Oh, it's an absolute pleasure, Greg. Thank you so much.
It's uh it was awesome to meet you at the Shenzhen SEO conference um earlier this year and it was it was such an amazing conference and it was uh one of those rare times where we had the tour uh of all the different places we could be able to see like from a cultural perspective, from a business perspective, from even from like an engineering perspective of seeing how the factories would make all those multitudes of of uh things that just power the world uh with all those wonderful things that are available and then of course the conference itself with some amazing uh keynotes. U it's wonderful to reconnect with you. Um hi everyone uh listening to um Greg at the moment. Please like, please subscribe because he's a great guy and I'm really excited to get into uh this topic a little bit further. Okay.
You are the first guest to ask people to like and subscribe. So thanks so much for that. Appreciate that. So, we're we're here with you, Nick, because you and I took this taxi ride where you completely blew my mind. You launched into this. I think we only had like 25 minutes, but that's perfect because that's all we have for this episode today. I'm hoping to recreate that. You are very technically savvy. I know that's not all you do. You do strategy as well, but today I really wanted to dive into what you taught me about, which is query fan out.
So my first question um you know I'm not hyper technical but for someone like me and in case anyone else in the audience is like me how would you explain query fan out? Well I think like for anyone who's want to get started and think about query fanouts um like I'll I'll use a little analogy. So um imagine you ask a very literal research question like what is the best uh what is the best electric car for a family of four living in a city with cold winters and what are the potential government rebates? So that's a very very long tale kind of question that a lot of um AI bots will get.
So what they'll do is they'll take that very long answer and then break them down into different conceptual components. Um, so an an experienced assistant might search for that exact long sentence and find very little. But a highly efficient assistant, however, would instinctively break that complex request down into a series of similar, smaller, logical questions. And that's exactly the same thing that that an LLM, a large language model, will do with that longtail query. So it might break it down into five-star safety rating, electric SUVs, EV battery performance, and cold weather. um electric vehicle government rebates in Australia or whatever country you are for the for the time of 2025 or 2026.
Um EV charging station availability in the city that you're in. So they would gather like the best information for each of those small questions and then synthesize that into a single comprehensive for you. And that is exactly what an LLM will do from a technical perspective. They're taking that long tail question, breaking them down into those concepts, doing lookups of real time uh results. So like that's called grounding with Google search or with um Bing search.
Um, and it is pulling all of those uh those into um an amalgamation where it uses its own understanding of the world, its pre-trained AI brain and those live SEO uh answers that it's getting from search and using what's called rag retrieval augmented generation basically to amalgamate that into the result that it spits out. And it does it really quickly. so quickly and that's really what you're seeing. So that breaking down process is in essence query fan out. So Google's own patents show that their AI systems do this when they receive those really really long conversational prompts in AI mode. Their agentic layer deconstructs that into a series of more traditional keyword style sub queries to run against their web index.
So our what we do at uh at Dejon is like our query final process is a methodology to really just reverse engineer that. We take a core topic, use our own models to generate the entire universe of potential sub questions that an AI would ask and then we build a single authoritative content page designed to answer all of them in one place or we'll look at existing content to see um what things that we can be able to add to actually amalgamate that to create a authorative a single content piece that is designed to really answer a lot of those things um in that one entity location.
So instead of just answering the user's question, we're proactively anticipating and answering the AI's internal questions, making the page the most efficient and logical source for its use. Hey, hope you're enjoying the show. We're going to get right back to it. Real quick, if you're not sure how you're showing up in AI tools like chat, GPT, Perplexity, and the like, click on that first link in the description. If you fill that out for us, we'll send you your own AI visibility report and see how we can help you boost your results. All right, without further ado, remember that link and enjoy the rest of the show.
Okay, I I want to understand more about how you anticipate the questions. You know, you do this query fan out. There's like a huge surface area there. As you said, that was an incredibly long tail scenario as an analogy and then there were so many different I I'll use the word unpack like so many different keywords you could unpack from that. How assuming we've got a limited content creation capacity or linkbuilding uh how do we know where to focus? How do we prioritize one of those terms or a few of them? Yeah. So, you've um asked two really really good questions. The first one was like how does it do that?
Um if if we want to sort of like maybe just take a brief look into the technical process when it's breaking those down, there is like a tokenistic probability that it will assign to different rungs of what it could possibly answer um from that result and it has like a certain confidence um with its own model like each of those different variations. So when we're doing query fan out, we're really wanting to just to do the ones that are sort of like the next door to the most likely results and those are the ones that we're sort of wanting to analyze and look at. Um we also have a way of like looking at the relevance of that to the client.
So we're doing pre-training with their own um queries that have like clicks and impressions um typically across the top 10. um because we work with very large clients, we have a lot of data to be able to do effective modeling in that way. Um but that's a really good way that like the query fan is actually just exponentially creating all these variations. So we can see what they currently rank for versus what they currently not.
And it's really really helpful because it kind of almost speaks to like the market share of um how much visibility within um within what an AI model may produce for this particular topic for this particular service for this particular product. So that's really really helpful. And the second question which was if we're got limited content creation or linkbuilding capacity like uh thinking about them in terms of prioritizing them. I think that's like a really really great critical operational question. Um, and it's really a part of like a multi-stage datadriven filtering process that that I do to be able to get to that point because um, I never just sort of think like you know we just spray and pray like the prioritization framework is really important where we start with like we look at like what they're currently ranked for.
So like all those really important seed queries from like those existing authorities. So we don't necessarily start just brainstorming. We start with the data extracting the clients like top performing queries from search console uh the topics so that they already have proven authority and visibility. That's really helpful to be able to see what the model already has some confidence of what they're known for. Um and that ensures like we're building upon like a strong foundation. They're not necessarily having to start from zero. Um for some clients we um like do query demand estimation.
So we run the query fan outs on all of those really important seed topics. So you know the key services or the key um categories of products that are really you know net positive you know they have the the largest return on investment. Um and we have a massive list of those seed queries. Uh synthetic queries which is what we just say like what happens when you do query fan out on those seed queries. synthetic queries is all the ones that we're we're doing like for the probabilistic outcomes most of which of those have zero search volume in traditional tools.
So that's really really fascinating when we were looking at this of like well we don't have traditional tools to be able to look at like what is the most positive thing that we should prioritize within um query fan art cuz with some clients we get anywhere from like 600,000 to the the multi-million dollar query fan art results just just in terms of what they do or don't have already surfaced. So yeah, to solve that problem like um Dan Petravich who I highly recommend everyone to Google to follow on LinkedIn um as he is so generous with uh sharing actually how we're building and generating all this um but he built the query demand estimation.
So it's a basically a deep learning model that we trained millions of search console queries and their corresponding impression data. Um, so it's a purpose-built to predict the potential impression volume for all those new longtail and conversational queries. So that allows us to score and rank the semantic clusters generated by the fan out um focusing on our limited resources of the topics with the highest predicted demand. To say this really simply, query demand estimation gives us what the traditional tools do for just normal queries. It gives us like a search volume.
It gives us an estimation of what we think that search volume may be for all these new synthetic queries that the query fan out gives us. And that's really really helpful to be able to to um have that as a further filter to look at prioritization. Um the other way we prioritize is also looking at strategic brand gaps. So um cross referencing this prioritize list with insights from AI rank and tree walker analysis. Uh we also have um citation mining which is something that we didn't talk about in Shenzen but is now um something that is a part of our toolbox.
Um, so if a high demand topic cluster also happens to address a weak spot or a negative bias in the model's perception of your brand, that gets elevated to top tier priority because um it might just well be that um you know, let's use a a brand at random. Let's say Patagonia um an amazing um sustainable apparel company um with really specialist um quality gear. It's not necessarily known for um women's marino um underwear and bras. Say for example, it's just not known for that. So, it's finding that like when people are searching for this, Patagonia isn't the one isn't the brand that's used as part of um you know, its top results.
So we can be able to see that it's like well if that has like a a good margin, it's something that is really important as far as like it's a it's a project that they're wanting to really reinforce um you know having adventure gear and hiking gear that is you know high peak performance um for the throughhiker for the for the for the climber. Um, even for folks that are traveling who want to have minimize on luggage and really just focus on just having a couple of key pieces that they don't necessarily have to replace constantly. Um, we want Patagonia to be there.
So, if that is found as a weak spot or even a negative bias in the model's perception, that's what we're going to prioritize um as far as um like thinking about creating content or even doing an outreach campaign to really strengthen some of those signals. So, that's really really helpful just to you know have like a three tiered idea of like how how we solve those kinds of problems. Okay, Nick, you you are not just technical, you're very strategic. And you you know what I'm thinking next is there's an emotional element to client work where you can you can take a databased approach. You can show them the strategic value of what you're proposing and sometimes at least in my experience clients and say that's nice but here's our real problem and this is what we want the focus to be.
So, you know, when the rubber hits the road, so to speak, and you've presented all of this, how do you involve clients in this? Do you do you typically say, "All right, then if that's your priority, that's what we do, or do you, you know, try to try to get them to understand that this is what all the data says we we need to do?" You know, what what's the client involvement process here? Yeah, this is a really great question and I think for anyone else that is in in my shoes, uh whether they're inhouse wanting to influence key stakeholders to prioritize um this as part of like projects that have been scoped um in the near or distant future.
Um or it's uh on a consultation or an agency basis where you're wanting to liaz with a client um and really want to prioritize some of these key actions and maybe the internal team is sympathetic but again they have to be the internal advocacy for getting these key actions done. Um it's a very very tough thing and I think it always requires a lot of patience and creativity and empathy but I think the default model in my mind is to always lead with the data and then collaborate to refine that.
So um I'm thinking like you know having that initial datadriven analysis where we complete the first three stages of our prioritization framework and whether they want to you know go through that or not is irrespective cuz we need to understand the problem inside out um to be able to deliver that. So in my mind I do all the search consults um seed extraction and when I say seed to folks out there I just mean like looking at all the keywords that are actually driving traffic that are really net commercially valuable um and really helpful and maybe even not just commercially valuable but also look at this across the entire um customer journey funnel from discovery all the way through to um consideration and maybe even um you know brand heroing of
repeat customer type things. Um, I want to be able to look at all of those keywords and extract those to understand them. Um, I love doing the query fan outs because that's really really helpful for us to be able to see as far as market share how much ground that they're currently covering and how much that they're not. Um, doing the AI rank gap analysis.
So before we even have a strategic session with a client um I think it's really our job as some as a person with all the tools and um the experience to come to the table with a clear objective a databacked point of view because if the decision is to say for example decommission a whole subdirectory or subdomain based on an assumption and that assumption isn't correct that could be a multi-million dollar decision that goes ahead without um really having the due diligence. So I try my best to be as well informed across um across their their key verticals as much as possible in the case of questions like that or at least try and um invite room to to be able to do that to to be able to do that due diligence. And that just prevents the entire strategy from being immediately anchored to the highest paid person's opinion.
As I'm sure all of us listening to this know exactly what that's like. Um I think collaboration is essential because I think when we're presenting our findings in a workshop like it's you know you can put um so much time and effort and energy into um a report but you know and we can say like you know based on our analysis of the data the model's perception the predicted demand uh these five topical um topical classes represent the highest leverage opportunities for you to build authority in the next six months.
um if we're not really helping to to really understand this where their internal priorities are um like which of those align with their upcoming product launches, marketing campaigns or strategic business goals or even like a content calendar um then it's going to be very abrasive to their own internal workflow. So that's really really important that we understand this right from the get-go. Um, and there's been times where like I've I've done reports um and we've sort of had like um a two-month time frame to build the entire strategy and then towards the the quite literally the six week six week mark in an 8week project uh they've said oh we're going to basically uh migrate these pages to this other page type.
Um, and that is consolidating four different templates into one. And they're going to have so many different things um in tabbed content which is going to be hidden. So that's really really important to also understand like what the you know what the the net impact is going to be. And I think that a lot of technical SEOs and I think that a lot of really good strategic um uh SEO consultants are able to at least triage those kinds of um decisions without necessarily needing to do all of the requisite um impact analysis before that happens.
But I think it's really important going into those kinds of conversations knowing exactly what's going to happen. and um the that final prioritization that that collaborative process allows us to really just layer the client's crucial business context on top of our datadriven foundation. So we might re we might slightly or even significantly in some cases reorder the priorities based on their input but the conversation is always grounded in the opportunities that our analysis has already surfaced. Um, one thing that I want to make abundantly clear and I and I will probably repeat this in our conversation. Um, but for AI visibility, the content must be in the initial HTML. It must be in the initial HTML.
This is why serverside rendering is excruciatingly important um part of the project at every single turn. And so many of our clients, particularly our um our lovely Chinese clients who uh work with us um have a very JavaScript client sideheavy um infrastructure that they're trying to win the same kinds of goalposts um against competitors that don't have uh this issue or they've got a dynamic rendering process with really really heavy firewall um and W rules in place and that can really limit the bots from being able to actually crawl and be able to see things.
Um, I think it's a trust but verify model that respects both the data um and the deep domain knowledge but that ensures that we're working on things that are not only algorithmically valuable but also commercially relevant. Yeah, I mean you you said quite a few things there, but first I just want to say that speaks volumes of the environment inside Dejon that you all can think critically. We're not we're not just yes people here uh you know do doing whatever the client wants where we're trying really to solve business problems. Uh you you said um not just doing whatever the highest paid person says to do.
Uh I completely agree with but I I can only imagine when you're working with enterprise clients and you're dealing as you said with trying to win the same goalposts uh but with a completely different set of challenges. It it must be well at least it's a lot of opportunity let's say uh a lot a lot of work to be done. Um so you sounded extremely confident uh that the content must be in the HTML right uh that that's that's database I'm sure. I want to ask you to to stick your neck out there a little bit here and I'm the one asking the question, so I appreciate I'm putting you in a hard spot, but I'm looking for a prediction, Nick. We're heading into 2026.
Uh what what what can this you you said goalpost, but these are moving goalposts really the ranking algorithms for both ChatGpt and Google right there. They're always changing. I would say ChatGPT, the LLM that that is most people's priority right now. uh probably evolving even faster just it's not a very mature platform. Um is there like a single dominant ranking factor you'd put your money on for example Reddit uh in the past year? Is there is there anything you'd predict as Yeah. This is where we we put our focus. Yeah. Um that's a that's a that's a really really interesting question.
Um, I'm actually just trying to think here of what to sort of say. I mean, I I feel like making sure that your your key critical information that you're wanting to present to the end user is in the initial HTML. um I think is really quite critical considering the amount of um the amount of sites that are that are leveraging JavaScript to be able to inform their pages and uh in their their templates and the way that they're structured. Um to me that this is just a really important hygiene um issue.
Uh, a lot of folks think that because Google bot has evolved and has a two-wave index um indexing rendering and indexing uh pipeline that because Google bot can be able to render um some of the JavaScript that therefore that is good enough. However, that's always been very very specific to Google bot and Google bot solely that there are of the 144 different user agents that I do with my testing for AI bots and um and SEO bots alike um accessibil data accessibility is the number one uh thing that we can be able to do to just allow um allow them to be able to see our contents to be able to allow to see our structures um and to be able to see in the way that we're trying to put forward a prioritization of um pages and content and uh assets that we're wanting to be
used so that users can be able to find us. Um, I see carninal sins all the time where the site is essentially relying on a search engine to be able to index pages uniquely for specific queries, but there is absolutely no internal way that a user would be able to find those pages natively. And that's something I think is really important to consider when just just uh on your own sites um you know creating these different um challenges of like well if I didn't use the search bar how would I find this page? What internal links are pointing to these pages um that would be able to ensure that I'm having this um be put forward.
I think at the moment even though we're seeing um definitely a lot of changes within um within our LLMs I think that still it's very much at the time where we can be able to do some really interesting experiments and be quite literal in the way that we're to present content um to be extracted and used. So I think if there is one thing I would 1,000% say with that will be true now that will be true tomorrow and will be true in a year's time that if your key assets are not presented in the initial HTML there is zero confidence that I have that your information will be picked up by a model and used. I'll say that right now.
So everything else that comes apart of um of like wanting to optimize a page is really secondary to that initial point. And that's I think uh something that is really important to to be considered. That's awesome, Nick. Uh, listen, I'm really grateful for your time so far. And as we're wrapping up here, you just have a wealth of knowledge that I feel like my questions couldn't even hope to scratch the surface here. This is my favorite question to ask at the end of every episode. Is there something I didn't ask you that you're thinking, I I don't understand why we're not sharing this with the audience.
Let me just dive deep into this for our final few minutes here. Um, yeah, that's that's a really great question and one I actually don't get a lot. Um, I think the one thing I haven't even really touched on is the evolution beyond like what we understand as like an answer engine to like an action engine. Um, right now I think we're all focused on optimizing for an AI that answers the question like what is the best hiking gear? The next front like where the next frontier is optimizing for an AI that can act on the commands.
Um, you know, like uh find and book a a a find and book a uh a time with a hiking guide to show you the best uh uh routes in Victoria, for example. I don't know. Um I think this is the true agentic future. I think it's um it's an AI that can execute multi-step tasks on behalf of a user. Um for us technical SEOs and um even for people who are um you know developer liaison this is a profound implication because it means our job will expand to include optimizing for machine executable actions. Uh this requires flawless API integration machine readable forms structured data that doesn't just describe a service but programmatically explains how to interact with it.
Um, it can even also uh go further to describe all the attributes of maybe certain products and the way that they can be used or even modified or customized. So, the final thought I would leave your audience with is this. Our job isn't just to be a trusted source of information. Um, I think it's to make our clients businesses the most efficient and reliable and frictionless endpoint for an AI agent to complete a task. Um, and that's a true endgame of um, AI visibility engineering. That's awesome. I love it. I think honestly I know you weren't staking it as your prediction, but I I I'd put my money on that. Uh, we're going to see that in the near future.
I Nick, thank you so much. Uh, I know you know Dejon Marketing, but where could people find you if they want to look you up and ask you more follow-up questions about what we talked about? Yeah. Um, I'm on LinkedIn and I'll just say that I will eventually get to your questions, but if you're asking me quite specifically for something, um, I usually like, you know, find those a lot more quicker than, um, people just saying hello. So, just, uh, send me questions in mind that I get a lot of linkbuilding requests that I completely ignore. Um, otherwise, I'm also on Twitterx.
Um or you can find me at SEO Collective as I'm a co-host uh with Peter Machinkovich uh who just did an amazing talk at Shang Mai SEO conference um here in Melbourne and we um have events every month um and they're fantastic. So if you're if you're a local Melbourneite, please come along to SEO Collective where you can meet me in person, ask AMA, ask me anything um and we go through the news and um keynotes. But um I I highly encourage you guys to of course to subscribe to Greg um Greg Helier's amazing podcast. Um check out all of his other episodes. He's uh a very thoughtful person and ask great questions.
So it's always a really great time. And thank you so much, Greg, for your uh for your time. Like, subscribe, and uh let's keep the the good times rolling. Oh, I need you I need you back on, Nick. Be be my hype lady here. Thank Thank you so much for coming on. We'll see you next time.
We turn the ideas in these videos into AI visibility and Google traffic for our clients. Book a free call and we'll map out what's possible for your brand.