Enterprise SEO in the Age of LLMs: Testing What Matters Now | Martin MacDonald
Hi, and welcome back to The Search Session!! I’m Gianluca Fiorelli, and today I’m joined by Martin MacDonald to unpack how SEO has shifted—from the scrappy, experimental 2000s to today’s AI-infused search—and what that means for doing impactful work at enterprise scale.
Here’s what you’ll take away:
Climate, culture, and context: A warm (and humid) human intro—because the best SEO brains also have the best backstories.
LLMs as “search wrappers”: How ChatGPT’s answer-building pulls from search results via query fan-out—and why top-10 visibility suddenly matters more than ever.
Structured data’s second-order win: It may not “rank” you directly in LLMs, but it boosts inclusion in the source corpus LLMs pull from.
A plug-and-play experiment framework: Mine GSC for ultra-long queries with impressions but zero clicks to uncover LLM-triggered demand you can actually optimize for.
GSC trust issues (and fixes): What’s real, what’s sampled, and the one place to validate truth: your logs.
Enterprise reality check: Big sites aren’t “more sophisticated”—they’re more constrained. Success = breadth (millions of pages/queries), not trophy keywords.
Enjoy watching and listening!
Video Chapters
Transcript
Gianluca Fiorelli: Hi, and welcome back to The Search Session! I'm Gianluca Fiorelli, and today we have a guest who's been doing SEO for—well, many, many years.
He could be defined as a globetrotter in the SEO world. He’s Scottish, from a very important clan—but spent most of his life, until not so many years ago, right here in Spain, where I live as well, more precisely, in the Costa del Sol in Andalusía. He then moved to London, Great Britain, and later to the US. And now he’s living “la vida loca” on a beautiful Caribbean island.
Our guest is an enterprise SEO, specializing in product-led SEO, AI workflow, and digital growth. With his deep experience and knowledge, I'm sure today’s going to be a fantastic conversation. Please welcome—Martin McDonald! How are you doing, Martin?
Martin MacDonald: I’m good! How are you doing, my friend?
Gianluca Fiorelli: I'm fine, I'm fine. Still very hot here in Spain! Today’s mid-September as we’re recording this, and it’s still very hot. I imagine it's hot in the Caribbean, too?
Martin MacDonald: It’s actually hotter in Spain! To back up what you were saying in the intro—I spent the first 27 years of my life living in Spain, so I’ve got a solid basis for comparison between the two kinds of heat.
I’m currently living on the island of Barbados, which is an absolutely beautiful place—fantastic, really. But the temperature year-round is between a minimum of 28°C at night in the middle of winter and a maximum of 31°C during the daytime in the middle of summer. So the temperature graph basically stays flat all year. It doesn't get any hotter or colder throughout the rest of the year. The only thing is we have so much humidity here versus Spain or that European Mediterranean heat, which is nice and dry. So that takes some adjustment, but it's just a nice temperature.
Gianluca Fiorelli: It's the only constant, yes. I was just thinking about that—the closer you get to the Equator, the more stable the temperature tends to be. It’s not like when you were living in the UK, where you’d have those horribly cold winters, and then suddenly, boom—25 or 28 °C—and everyone would start yelling about climate change!
Martin MacDonald: Oh yes, totally. Totally. I was born in the Highlands of Scotland, but I don’t really remember it. My parents left before I was two years old.
And yes, I distinctly remember, throughout my youth, going back to Scotland for two or three months every summer, since Spanish school holidays ran from July through September. As soon as the temperature hit 23 or 24 degrees, there were ginger people exploding—everyone getting sunburnt. It was ridiculous.
I can’t live like that. I was back in Scotland for half of December, all of January, and part of February this year. I remember looking out the window at snow on the ground and just thinking: “What did I do in my life to end up here at this point in time?” Barbados suits me perfectly. The temperature is just right.
I also lived in San Francisco for about 10 years. And honestly, it’s cold. People assume California is hot, but I lived in the South Bay, in the Santa Cruz mountains, and I saw snow on the ground there. It’s not warm year-round—it definitely has seasons.
So yes, maybe I’m boring, and plenty of people prefer having four seasons. But I’m perfectly happy with 28 to 31 degrees all year, with just some variation in humidity.
Gianluca Fiorelli: Yes! For instance, I would have no problem living in eternal spring—or that end-of-summer kind of weather.
But let’s shift gears a bit. We'll come back later in the episode to ask a few more personal questions, so our listeners, watchers, and spectators can get to know you better as a person.
From Pre-Google SEO to AI Search
Gianluca Fiorelli: For now, let’s dive into the topic. And I’ll start with the classic question of The Search Session: How is SEO treating you lately?
Martin MacDonald: You know what? Really good question. For anyone unfamiliar with my background—I started doing SEO in the late '90s, before Google even existed. Of course, SEO was a very different thing back then.
I was building my own web products and, between 2000 and 2007, I ran an online casino and gambling company. From around 2001—which I’d call the birth of Google—through to maybe 2017 or 2018, it was fantastic. Honestly, it was the best job in the world. There was always something new happening. Everything was constantly changing.
But to be honest, I kind of fell out of love with SEO from around 2018 to 2023. Nothing really changed. It started to feel boring—too predictable.
But over the last—let’s call it three years—since, well… where are we now? September 2025? Over the last three years, with the evolution of SEO into what I’d now call SEO plus AI, I’ve fallen madly and deeply back in love with the industry. There’s something new to learn every day again. It’s got that same energy it used to have.
The only difference, I guess, compared to back in the day—especially when we first met, Gianluca—is that there just weren’t that many people doing SEO professionally. Maybe a couple thousand around the world.
Now? There are thousands just in London. Thousands in San Francisco. The scale of the industry today is completely different from what it was back then. And that definitely has its positives—though, to be honest, it comes with some significant negatives too. But we’ll ignore those for now.
So yes, SEO has been treating me fantastically lately. I’m incredibly grateful—for the lifestyle it’s given me, the countries I’ve been able to live in, and the companies I’ve had the chance to be part of over these past 25, 26 years. Wow, that makes me sound really old! But yes, it’s been over a quarter of a century since I’ve been calling myself an SEO. That’s kind of terrifying.
Gianluca Fiorelli: You're still younger than me!
Martin MacDonald: Well, do you remember when we first met?
Gianluca Fiorelli: I do—it was definitely in London.
Martin MacDonald: Yes.
Gianluca Fiorelli: I think it was at a SearchLove event—maybe one of the first ones. And I have this vivid memory…Sure, there were the talks, but at the classic SEO networking, we have to confess something. I don’t know if you still have this bad habit, but at the time, we were both smokers.
So we were outside a pub, smoking, and that's when the conversation started. You told me, “Hi, I’m from the UK but I live in Spain,” and we started chatting—beer in hand, cigarette in the other—talking about the differences between Spain and the UK, and of course, how SEO was in 2011.
Martin MacDonald: Oh no, it was definitely before 2011. I can tell you exactly when—it came to me last night. It was the original SearchLove, and I believe that would’ve been 2009.
Gianluca Fiorelli: Ah, yes, you’re right. It was 2009!
Martin MacDonald: Yes. There was an event called SEO PRO, which later became SearchLove the following year. SEO PRO was a Moz conference, held on the fourth floor of some venue—freezing cold room, I remember that clearly. I don’t think we actually met that year, though.
Gianluca Fiorelli: Right—we met the year after. My confusion about 2011 was because you were also speaking at the first MozCon.
Martin MacDonald: Yes, yes—absolutely.
Gianluca Fiorelli: And I attended that one.
Martin MacDonald: In 2009, at SearchLove, I think that was maybe the second or third speaking gig I ever did. I also spoke at the first MozCon and again at the second one. After that, I ended up doing 65 or 70 conferences over the next decade or so.
I was everywhere—doing conferences all over the world for about 10 to 12 years. But I’ve kind of stopped since moving to Barbados. Everything feels so far away now, and honestly, I’m perfectly happy just staying here. The only downside is that I don’t get to hang out with all my friends from the industry as much anymore.
Gianluca Fiorelli: Yes, I can imagine. Maybe that’s also why you tried your own adventure in the conference world, with SEO Barbados.
Martin MacDonald: Yes, absolutely. I didn’t put one on last year—I had some personal reasons that brought me back to the UK for a couple of months toward the end of the year. But we’ve held two events so far here in Barbados, and have had 30 or 40 fantastic speakers come over across those two editions.
The first one was held at Clifton Hall Great House, and the second at the O2 Hotel, which is on the south coast of Barbados—both incredible venues.
I’m hoping to run the third one in November 2026, so toward the end of next year. But watch this space—there’s a lot of organizing that needs to happen ahead of time. Honestly, unless you’ve actually organized and run a conference yourself, you have no idea how much work it takes.
Oh my God—I remember posting my sleep tracker stats at the end of that week. I averaged three hours of sleep a night for the ten days leading up to the event. I was completely dead by the end of it. But I mean, you know exactly what that’s like.
Gianluca Fiorelli: Yes, when I organized the Inbounder, I was going to say the exact same thing. I mean, I remember after the first Inbounder—after the speaker dinner of the last day—I invited all the speakers to a party and then went home and slept for 24 hours straight. I was totally destroyed.
But anyway—coming back to what you were saying before, about the big differences in SEO. When you started doing SEO—before Google was even a thing—of course, it was a very different landscape. But there was also something quite similar to what we’re seeing now: you had to work on dozens of different surfaces.
Even though they were very different types of searches, they still required substantial work. A lot of it was based on directory-style search, rather than the more classic, Google-style search. But one constant, even back then, was that you really had to consider how different search engines worked. And interestingly, that’s something we’re seeing again today.
Now, of course, we have Google and Bing—let’s leave aside Yahoo and other search engines like Baidu—but we’re still dealing with what we could call “classic” search engines, even though they’re now infused with AI components in search. But in addition to that, we now have all the LLMs.
What’s surprising, especially for people who entered the industry during Google’s dominance—when no one really considered Bing as a serious alternative. Back then, optimizing for Bing was just a byproduct of optimizing for Google. Today, the approach is completely different. We have to think about optimizing for Google, and then separately consider how we approach LLMs. And importantly, not all LLMs are created equal. This is something SEOs really need to start understanding. Claude and Anthropic, for instance, generate synthetic answers in a very different way compared to ChatGPT or Gemini.
And if you’re a more advanced SEO, maybe this shift feels familiar—because we had already started to understand that this was where things were heading. We had to learn how to do SEO for YouTube, SEO for news, and SEO for visual search, as well as other different surfaces. Even SEO for TikTok, which just a few years ago was the new frontier everyone was talking about.
So in a way, this is something the early days of SEO and the current era have in common. But as you were saying earlier, one of the big differences is that back in the day, we were just a few thousand people in SEO. Now, we’re in the hundreds of thousands.
Gianluca Fiorelli: So what's the positive aspect of that kind of scale? And what’s the negative one that you were hinting at?
Martin MacDonald: I think—and this actually ties into both sides of that question—the biggest difference between SEO 20 years ago and SEO today is that, back then, none of us really knew what we were doing.
Everyone had to learn by doing—by testing things, building their own frameworks, measuring results, and adjusting strategies based on what worked. And honestly, that’s exactly where we are now with optimizing for LLMs.
Because frankly, as of today—September 17, 2025—anyone out there on social media claiming to fully understand how each individual LLM ranks its answers or generates responses… they’re talking rubbish. They do know at this point because everything’s still changing—weekly, even monthly.
Each and every model you start tracking and looking at responses—it's different every couple of weeks. So consequently, there’s no way anyone can have a full understanding of this at the moment—because it simply isn’t set in stone yet.
All of these companies are still optimizing their information retrieval systems. You could call it “sort ordering,” but honestly, it’s not quite the same thing anymore. And that’s exactly why it’s become interesting again.
We’ve returned to this environment where the key skill is being able to search, interpret the results, draw conclusions from them, and then optimize toward that understanding. But there were no solid, established teachings.
Now, by the time we’ve made it to the mid-2010s, there was very little difference in the way Google sorted its results from one year to the next. Which, as I said earlier, is why I started to find it boring. Once you understood what worked, it was all about scaling that process—not discovering anything new.
And it was the “discovering new things” part that really attracted me to this industry in the first place. You’ll probably remember—if we go back to the early days of SEO—the only crawler we had was Xenu's Link Sleuth. I mean, it was the best tool we had at the time.
Gianluca Fiorelli: Yes, I remember.
Martin MacDonald: And the only backlink database we had—and it was fantastic, I still miss it to this day—was Yahoo’s Open Site Search, or BOSS. That was a free database of links.
The day Yahoo shut that down was actually what gave birth to Moz—or SEOmoz, as it was known back then. And then, of course, came SEMrush, Ahrefs, and all the other tools we now see out there—delivering insights and products to tens, if not hundreds of thousands of clients. That’s actually one of the ways we can measure how big the industry has become.
For example, if you look at the SEC filings from SEMrush, you can see exactly how many active subscriptions they have. So just from that alone, we know there are hundreds of thousands of SEO practitioners through that reason alone. None of this existed 20 years ago.
And it doesn’t really yet exist for LLMs either—which is why observation as to how things are working and then coming up with ideas as to how to manipulate those results as best you can for your own or for your own client's goals has become incredibly valuable again. And that’s why I’m saying: I love this moment in SEO more than I have at any point in the last decade.
We’re back into the scenario where everyone has to play in the sandbox again. Everyone has to get back to where we were previously: there are no SEO blogs you can just read now that tell you how to optimize for this. You have to work it out for yourself.
And this is where a big change in the industry happened—probably between 2007 and 2012. In that five-year period, we went from an industry of people who were testing, refining, and then planning to an industry of people who—and there’s nothing wrong with this; it’s simply how the industry developed—once we’ve handled how Google worked, learned their trade by reading blogs, going to conferences, and studying what others had already tested. They never learned those skills themselves—because they didn’t need to.
Now we’re suddenly back in a world where that’s critically important again. And those same people who learned their trade in SEO by studying what others published are looking for content on LLM optimization or Google AI Mode optimization.
Frankly, as I was saying a moment ago, that content doesn’t really exist yet. Even with all the testing—and this is something I do daily across 10–15 large enterprise clients—what works changes frequently, which pages get traffic changes frequently, and the order in which companies are mentioned changes frequently.
So, as an industry—here in September 2025 (and if you’re watching this in the future)—we’re all working it out as we go along. That’s terrifying… but also incredibly exciting as well.
Gianluca Fiorelli: I totally agree with you. And we should also admit something: because we needed to experiment, our generation of SEOs—even those who are highly respected now, like you—tried everything. We wore the white hat, the black hat, and the grey hat because we had to test. And without going through that phase ourselves…
Spotting “AI Bros” vs. Real Experiments
Gianluca Fiorelli: For example, if there was a short checklist for recognizing when someone’s scamming you—as spammer-style AI bros—versus when someone is honestly presenting their own experimentation. How would you explain to the new generation, who are eager to read and watch what others are doing, how to recognize fluff from the truth?
Martin MacDonald: Yes, it’s a really good question. At this moment in time, anybody who says anything with absolute certainty is untrue. There is no certainty.
What people need to look for, in any content being published today, to gauge its accuracy and honesty, is the testing framework behind it. Look at how things were measured. Then look at the changes that were made—and what the outputs were.
In a multivariate world, you want to see: in this test group, this was done; in this control group, nothing was done. This was the impact here; there was no impact there.
Because we haven’t reached an industry consensus yet, you need to do your own research on other people’s research—specifically, how they arrived at their conclusions.
The number one red flag? People speaking in absolutes: “Here are the 10 ways to optimize for ChatGPT,” “These are the 10 must-dos,” etc.—with no context on how they reached those recommendations. Those are almost universally made-up, clickbait nonsense. So having a critical mind about how any of this works has become exponentially more important—because we don’t have industry consensus right now.
If you went back five or seven years, you’d hear the same advice from every reputable person: build content, align it better with the query, and provide a good user experience. Sure, things changed every 18 months or so, but they changed because that’s what mattered to Google at the time.
I’ll give you an example: Core Web Vitals. If we’d had this conversation seven years ago, I would’ve said, “Right—web performance. That’s the thing.” And I’m speaking from an enterprise perspective here: “You want your websites as efficient as possible for Google to crawl, parse, and index. And that’s the most important thing. ”
Now, that was said with the understanding that you might be working on a site with millions of pages—and those pages were individually relevant to individual queries people were searching for. But that happened because Google’s number one expense was crawling the web and maintaining the index. So it became very important for them to keep that cost base as low as possible—which is why they pushed web performance.
Go back five years before that, and it was all about mobile—mobile optimization, mobile viewports, mobile everything. That’s because Google was moving from desktop crawling to mobile crawling, having seen the inflection point where more people were using mobile than desktop.
How LLMs Really Build Answers
Martin MacDonald: So, the major structural and systemic changes in SEO over the last 15 years have largely been about making Google’s job—indexing the web and returning results—as easy as possible.
Now we’re in a situation where we need to think about LLMs in exactly the same way. Because… how does ChatGPT crawl the web?
Two months ago, the answer was: they’d licensed Bing data and were using the Bing index. So, all of a sudden, optimizing for Bing—or at the very least, Bing indexation—became really valuable. And I know some people at Bing will hate me for saying this, but it became really valuable for the first time ever.
But we’re now in September, not July 2025, and for the last month or six weeks, ChatGPT has very clearly been crawling Google a lot to get there. I mean, look at what we have right now—this week, the big scandal du jour in SEO is the &num=100 tracking in Google.
Gianluca Fiorelli: Yes, everybody’s talking about it as something against scraping, in the sense of SEO tools. It’s not really about the SEO tool; it’s about OpenAI using the SERP API for scraping Google.
Martin MacDonald: Precisely—100%. And from that, when we look at—and I don’t love all these terms floating around right now, but let’s use one that’s common—query fan-out. If you install this amazing extension by Ayima called ChatGPT Path—I was just checking in my browser to see what it was called.
Gianluca Fiorelli: Yes, I have it too.
Martin MacDonald: Right—so if you install ChatGPT Path, the Chrome extension, and then go to ChatGPT and run any search, it shows you the sequential searches ChatGPT makes to answer your query. You start to understand which queries are being made.
Previously, for each of those queries, they were looking at the top 100 results, then going and checking information from individual pages they deemed relevant based on that. That’s where a lot of the ChatGPT traffic—and the actual hits on your web servers—comes from. I’m not talking about human traffic here. That’s where a lot of that traffic was coming from.
Now, Google was doing a terrible job of filtering this out—which I know because I’m tracking 10–12 million keywords daily for their impression counts from Google. There were definite patterns that a number of people in the industry pointed out, which proved to be very valuable.
So ChatGPT, et al., were going to Google, running searches, pulling the top 100 results, then independently fetching content from those pages, and finally building your response using the LLM part of it.
This is the part that a lot of people aren’t disambiguating correctly in their minds. Right now, you could argue that ChatGPT—now that it relies more on Google results as opposed to the Bing index—is a wrapper for Google Search. It uses Google’s results, then parses that content to formulate its answer to give it to you.
Rewinding three or four months, it was very important to rank well in Bing for terms to rank in ChatGPT. Now, arguably, it’s just as important—maybe more important, who knows, because we don’t know the distribution between these two—to rank competitively in Google for these things.
And now that Google has removed &num=100 from their search results, it suddenly becomes exponentially more important to rank in the top 10 results for the terms ChatGPT is searching for in its “query fan-out” model. Again—I’m putting that in air quotes because it’s a made-up SEO term at this point.
Gianluca Fiorelli: Yes, maybe the more accurate term would be something like query refinement or query reformulation.
Martin MacDonald: Yes. Absolutely.
Gianluca Fiorelli: Yes, I think so too. It was especially evident with AI Overviews. People said, “Look, AI Overviews cite sources that don’t even rank for the main query.”
But that was a brief, early read. As we later discovered—though not immediately—with the introduction of the AI Overviews, all this query fan-out buzz came out quite late, more around February–March. Because query fan-out is so specific, a completely different type of search result can appear.
So yes, I agree with you on how to identify the “AI bros” as I call them, from honest people who are doing experiments, or simply explain what they are doing and what the results are—correct or not.
Structured Data’s Second-Order Effects in AI
Gianluca Fiorelli: That’s why another polemic has been: how much do LLMs rely on structured data to better understand the meaning of the content? Honestly, I think it’s good that SEOs are somehow obnoxiously curious about even the little details.
But I think that even if—let’s say—JSON is tokenized, the structured data means creating, substantially, patterns, even if they are tokenized in the raw text as individual patterns like: this "Person" with this "name" has this "jobTitle" and is "sameAs" this link.
So there are patterns—and even if, let’s say, these patterns are not fully understood, this is just speculation, let’s be honest, if people—for instance, in a biography (I’m using the example of a profile page)—were using a list of things that in schema.org/Person are indicated as meaningful for describing a person, we would have really better profile pages.
So, even just with that, I would still say: follow a structure, so you have, as a byproduct, clear content. So—let’s go with your comment.
Martin MacDonald: Absolutely. I totally agree. But I want to address something—I’d seen a study. I’ve not even read the study, and I know this is terrible of me, but I’m only going on the headline I saw and whatever was shared on social media.
The point that was raised—and I honestly couldn’t tell you who did this—was that structured data has no impact on LLM ranking. I call BS on that, for one reason and one reason alone.
In addition to what you were just saying: if we’re in a world where AI/LLMs are extracting their answers from search, and structured data helps you rank in search, then by proxy, having well-formulated structured data markup (Schema.org) on all your web pages—if that enables you to be returned by Google or Bing as one of the answers for those queries in the fan-out—then indirectly you’re positively impacting your likelihood of being returned by LLMs.
So it’s not the same thing at all. It’s a myopic view to say, “When ChatGPT hits my page, it’s not paying attention to the structured data.” Screw that, it makes no difference. If you’re not in the corpus of web pages being returned in the first place, then you’re not appearing in LLMs. And it’s not something you can test in real time—it takes weeks to test.
Gianluca Fiorelli: Totally. But if we want to go technical—in SEO, since the very beginning of Schema (2010–2011), we’ve known they’re not a ranking factor. This is one thing that’s always been clear. The problem is that many SEOs don’t know how search works, and they always forget this sequence: crawling → parsing → indexing.
It’s in the parsing where structured data matters. The better the parsing, the wider the query set for which your specific content will be indexed.
Martin MacDonald: So this ties exactly back to what I was saying earlier: the things SEO became over the years were largely to help Google. Structured data is one of those. I mentioned page speed and mobile optimization before—but structured data is another of those exact items where it was fundamentally thrust upon us so Google could reduce their CPU overhead: the processing cycles required to understand the context of a given webpage.
So structured data has been critically important since Google started using it—for that reason and that reason alone. Anyone saying today that it’s not important for AI is—without being pejorative—showing a lack of understanding of the second-order effects.
And one more point on people’s understanding of crawling, parsing, and indexing: at SMX Advanced Seattle in 2017—maybe 2018—Paul Haahr from Google gave a fantastic masterclass on this. I’ve cited it so many times to product teams in large enterprises.
His surname is H-A-A-H-R. If you Google—or ask ChatGPT—for “SMX Advanced Paul Haahr indexing,” you’ll find it on YouTube. Gianluca, you should link it in the description so people can click through after watching. It’s a terrific 30–35 minute masterclass on exactly how Google works.
It should be required watching for anybody who works in SEO—and for anyone now trying to optimize for LLMs, assuming LLMs are using search indices, which all evidence and demonstrations suggest is how results are being built. So yes, that should be considered absolutely critical viewing—and structured data is a strong part of that.
Experiment Framework: Mining GSC for LLM Query Patterns
Gianluca Fiorelli: Staying on the topic of LLMs—and more specifically Google’s side of LLMs—you said before that you’re a man of many experiments. Is there one specific experiment you can, and want to, share with the public? Maybe something you started doing for things like AI Overviews or AI Mode—if you want; it’s not an obligation.
Martin MacDonald: So, the way I’d answer that question is this: the framework within which you run experiments on your own site and content is the important takeaway. Nothing I can tell you right now—based on what I’m working on across large enterprise sites—might be valid in a month or two when you’re watching this. It changes week by week.
So the important thing isn’t the individual results of any given experiment—it’s the understanding and the framework by which you should be doing these experiments. And by that, what I mean is that the only rich source we have in 2025 for keyword, search, impression, and query data is the Google Search Console API. You can run about 1,200 queries per minute through that API. It takes a while to get through everything, but you can build tools to do this perfectly fine—using Claude, ChatGPT, or just PHP and MySQL. You can build your own tool that does this.
What everybody should be doing is understanding their data as much as they possibly can from their own Google Search Console—through the API—to see what queries are new, what queries you’re being surfaced for, and what queries appear unnatural. This is a sensitive point right now.
People should be looking at how to analyze their search query data for unnatural-looking queries happening in bulk and happening regularly. When I say unnatural, specifically, I’m referring to a similarity between search types that doesn’t appear to be human. So look at very long searches.
What I want everyone to do is download—adding a dimension to the operator—so you’re filtering by device or country, and do it by day. Then download Google Search Console’s page, up to 25,000 queries. Download every single query you can over, say, a 12–16 month period (that’s about as much as you’ll get from GSC). Then look for commonality among the longest searches you can find.
So, simply take all this keyword data—and it could be 10 million different queries, it could be a hundred million. To start, just add a length field to every single one, okay? Then sort descending so you’re looking at the longest queries over the last 16 months of data.
Once you start looking at that, each person will see things no human would type into Google. At that point, you’ll start seeing patterns—patterns in how search has been triggered that led to impressions on your site.
Critically important: you’re not looking for queries with clicks. You’re looking for queries that have got impressions and zero clicks, and then you’re looking for patterns within them—and the highest repetitions of those patterns.
Once you’ve found that on any individual website, you’ll have a better idea of the keywords—the search queries—being used to populate LLMs with information about your site. Because that’s how LLMs are actually working at this point in time.
Now, I’m a super nerd for all this, and I love the experiments the best and brightest in our industry are doing—I find it fascinating. But I’m struggling to see, right now, how my fascination for those things actually improves outcomes for clients or for my own site. This piece of information I’m giving you directly leads to optimizing for your clients—or for your own site. So take all of your keyword data, find the longest queries, and look at them.
Gianluca Fiorelli: Yes, and I see what could be the second part of this experiment. With your individual patterns, you can cluster these patterns. From very long, verbose queries, you can shrink to topics—the topics implied by the query—and then go see if you’re actually visible for those topics. In this case, for instance—and I’m citing this tool not just because we’re hosted by Advanced Web Ranking— you can use Advanced Web Ranking for tracking visibility for topics. But you can also use—citing a tool from our common friend Dixon Jones, like Waikay for topics visibility, not single-prompt visibility.
And so you can start to understand where your gaps are. I see this as the second part of everything you’ve shared. So yes—thank you.
Martin MacDonald: Well, if you break it down even more simply—if you look at the commonality between these machine-generated queries that are happening—they’re actually very easy to optimize for. People just haven’t known what to optimize for. So if you’re seeing an 18–24 word query in your GSC that appears 20 times a day and has never had a click, start optimizing for those queries and see what happens.
Gianluca Fiorelli: Yes, but you also have to add context. I did something similar for a client, and at first I thought, “Ah, these are surely generated by an AI.” But a simple exact-match search on Google showed me they were copied and pasted questions from test students on the internet.
Martin MacDonald: Yes, but this is what I’m saying. Rather than giving a specific tactic you should do right now, the important thing is understanding how to figure out what you should do now. I’m being a bit meta because I’m not giving a specific hint.
Gianluca Fiorelli: No, it’s totally fine.
Martin MacDonald: I’m giving a “this is how you should be thinking about it”.
Gianluca Fiorelli: The important part is that you gave an example where the framework is the most important thing. And speaking of Search Console, I know you’re a big fan. You even built a tool once around how to use Search Console in your daily practice—a tool that, I think, has since been substantially—maybe unknowingly—redone by others.
Can You Trust GSC?
Gianluca Fiorelli: But here’s the problem: how can you “really trust” the data Google gives us? When—and I’m going back to the &num=100 parameter—it’s so biased by bot traffic hitting Google’s servers, without Google giving us a clear filter to distinguish AI-driven traffic from classic traffic.
Martin MacDonald: Yes, very good question. On the first point: I’ve spent the last eight years—actually, more than that, I’ve spent the last 25 years building SEO tools—but I’ve spent the last eight building my own internal suite, which is what I run my business on and use for all my enterprise work.
I’m going to show you something after this call is finished—and everyone else can, I don’t know, follow me on Twitter @searchmartin, because I’ll be putting that out publicly in the not-too-distant future, maybe even by the time this is out on the tubes.
But to answer the question of what you can depend on and what you can trust, there’s something very critical and interesting here: you need a solid understanding of how that data looks and behaves in the real world.
So, the first thing I’d say is that the queries I mentioned earlier are those with a large impression volume and zero clicks. If it looks unnatural, it probably is unnatural—and it’s going to be bot-driven. You have to understand as to what the point in that bot traffic is, and then you can optimize accordingly. But for the rest of the traffic—where you’re actually getting clicks and sessions—it’s spectacularly valuable.
But—and this is a huge but—the sampling Google provides through the GSC API is getting more and more aggressive every year. Five or ten years ago, you basically got all your search queries. In the pre–“not provided” days (pre-2012), we had all this data in Google Search Console. Every single year, the number of impressions being hidden that we’re getting through the GSC API increases.
This is particularly relevant across different individual sections: searches relevant to a person are highly filtered, searches relevant to an address are highly filtered, and it kind of goes down from there.
The only way you can truly measure it is by looking at actual traffic on pages via server logs. So, how many times a page has been accessed without an HTTP referrer—what appears as “direct traffic” in Google Analytics or any other analytics tool. And you need to look at the amount of traffic you’re missing on a given page type to understand how much you’re missing from Google Search Console.
There is one other way of fixing that in GSC for about 90% of queries. But—you know—this is the allure of SEO: it’s the kind of thing you need to test your way into. That’s why I said people should follow me on Twitter—yes, I appreciate it’s a hellscape—at @searchmartin, because at some point in the next month or two, I’ll probably put more info out there, as soon as I’ve finished rinsing it for myself. I’m going to be cryptic on that.
Gianluca Fiorelli: Yes, well, it’s a good cliffhanger.
Enterprise SEO at Scale: Breadth > Single Keywords
Gianluca Fiorelli: You’ve mentioned it a few times during the conversation, and I also said it in my introduction: you specialize in enterprise SEO—you’re actually working with a lot of enterprise companies.
Just for context for our listeners—in the past, especially in the travel industry, you worked with Expedia?
Martin MacDonald: I’ve worked with everybody.
Gianluca Fiorelli: Great. So my question has two parts. First: How mature are enterprise companies in everything related to search? Second: What impact has AI search had on their prioritization of work?
Martin MacDonald: So, an answer for the first part—how much more sophisticated are they? If we want to put it that way, not at all. Enterprise websites are perhaps less sophisticated than medium-sized companies, and that’s largely due to structural constraints.
I was Global Head of SEO for Expedia and all its consumer brands. Before that—this ties back to your conference naming, Inounder—I was Head of Inbound Marketing for Expedia’s B2B division, working with most of the large travel companies worldwide and helping with their marketing.
Since then—staying in travel—I’ve consulted for three major airlines worldwide, I spoke at TUI’s annual general meeting about seven years ago, and I’ve worked across most of the big travel companies except Priceline/Booking, because they will always remain the enemy, because I’m an Expedia alumni.
Are they more sophisticated? No. Not at all. They just aren’t. But in enterprise SEO, the sophistication of campaigns isn’t the point. It’s about the constraints of massive scale—and knowing how to optimize within those constraints, rather than optimizing for individual rankings for individual keywords.
I once had a conversation on Twitter where some—frankly—idiot in the industry said, “I’ll outrank you for any given keyword. You give me the keyword and let’s see who’s boss.” To which my response was: “I’m not interested in ranking for an individual keyword. I’m interested in ranking for five million keywords as much as possible.”
And that’s the difference between enterprise SEO—enterprise means scale of keywords, pages, and countries—and non-enterprise SEO. It is the breadth of what you’re trying to rank for, rather than the very pointy end of it. Now, clearly, ranking at the pointy end is important too, but there’s no difference there—whether you’re a small website or not; it’s about having the best content, having the most page rank, having the page indexed. That’s fine—but you know what? When you’ve got 300 million pages on your website, getting as many of those indexed as efficiently as possible is the difference. It’s about the breadth.
And that applies to any industry—not just travel. It just so happens that travel has the longest of long tails, I believe, of any industry on the planet. Ultimately, in travel, you’re optimizing for every hotel name, every town name, every city name in the world. Then add origin–destination searches—travel from A to B—and you’re looking at every combination of those worldwide. So you start looking at billions of pages. That is a different mindset.
Gianluca Fiorelli: Not to mention all the inspirational search space.
Martin MacDonald: Yes, totally. Like “best beaches in the Caribbean” is one thing—but if you could be number one for every single hotel name in the Caribbean, every single beach name, and how to get from A to B, where B is any of those places, that’s exponentially more valuable.
People look at enterprise SEO and think it’s some magical, mystical world they don’t understand. It isn’t. It’s the exact same thing—just at an exponentially larger scale. That’s the difference.
Gianluca Fiorelli: Yes, it’s similar to the concept of international SEO—you’re still doing national SEO, but at scale.
Martin MacDonald: Yes.
Gianluca Fiorelli: And with very specific localization nuances. But when it comes to the technical, it’s the same thing.
Now, for the second part of my question: how are enterprises reacting to the eruption of AI? Considering enterprises are usually bureaucratic and slow to decide, how reactive are they to AI search?
Martin MacDonald: Surprisingly reactive, in my experience so far. I have standing weekly or biweekly conversations with the CEOs of several Fortune 50-level companies where we’re literally exclusively talking about how AI has changed over the previous two weeks.
I’ve never had that before—CEOs of client companies regularly talking about SEO. I’ve been very lucky to work for two CEOs in my career who were SEO nerds themselves, and I’m grateful for that because it gave me the platform that I had internally at my previous companies.
But not once have I ever had a client come to me and say, “The CEO wants to chat every two weeks about SEO.” That doesn’t happen. It is happening with AI search.
Now, perhaps that’s a bit self-selective—coming from the corporate world and dealing with very large companies—but I’ve never seen this level of interest before. I think it’s because this is such a fundamental shift in how people will access information moving forward. It’s exciting.
Again—and this is the advice I give all the time—we can’t take knee-jerk reactions. Particularly in the enterprise world, you could spend three months engineering a solution, and in two months—maybe even one—that solution is no longer the right one.
So we have to be quick, we have to be agile, we have to be testing. We need to form assumptions from what the testing shows us, implement fast, and keep moving—because it’s changing so quickly. And that’s why I say it’s exciting again. I haven’t seen this level of engagement from large corporations about SEO in a decade and a half of working with enterprise. I’m seeing it with AI search, and I’m seeing it all the time—so it’s exciting for that reason as well.
Gianluca Fiorelli: Yes, it’s very exciting. Sometimes it’s a bit scary, but exciting. As I’ve said in other conversations, we’re not just optimizing title tags anymore, even if it was a basic thing—explaining that a million times had become pretty boring.
Quickfire Personal
Gianluca Fiorelli: So, well, one hour of talk. Let’s make a promise to have a second talk in six months, because things are changing so fast that an update would be useful.
Martin MacDonald: I would love to.
Gianluca Fiorelli: Before closing, just a couple of questions. The first one—since I live in Spain, I have to ask: you spent roughly 25 years of your life in Spain. What from those Spanish years still pops up, unconsciously, in your daily life?
Martin MacDonald: Oh, Lord, that’s a good question. Okay, I think I kind of live my life with a very Andalusian perspective. I was two years old when we moved there, and 27 when I left, so most of my grounding in life and my perspectives are very southern Spanish. That hits every part of my life.
And frankly, that also trained me to live in the Caribbean—on the Caribbean, for my British and European chums. Because all of that’s bad about living in the Caribbean—getting stuff done, bureaucracy, and so on—is exactly the same today in 2025 as they were in 1985 in Spain. So I think of myself more as Spanish than British because of my upbringing there.
And what do I miss? God, I miss the food. I don’t miss the lifestyle anymore because I’ve found it here, which is why I feel so at home in Barbados and in the wider Caribbean—but my God, I miss the food. A proper tortilla española—a potato omelet—oh…you can’t get a good tortilla in this country. Jamón serrano every day—oh my God. You can get it here, but it’s so expensive. I grew up with a leg of ham in the kitchen all the time. You’d walk past it and slice a bit off. You couldn’t do that here because the humidity doesn’t permit you to have any food outside; it would be green in two days.
So there’s a ton I miss from growing up there. But the lifestyle I remember from Spain, I’ve found here, which is why I’m so happy in the Caribbean.
Gianluca Fiorelli: Yes, and this is the last question. Why Barbados? I mean, the Caribbean has so many islands. Why did you finally choose Barbados?
Martin MacDonald: Barbados was the first country to introduce a nomad visa. I was between Zoom meetings, went on Twitter, and saw Barbados trending number one. I thought, “What terrible thing could have happened?” I’d never been to the Caribbean before, I’ve never been in the continent.
I clicked and saw they were announcing the Welcome Stamp, a digital nomad visa: you apply, pay a fixed yearly sum, and that’s your visa. This was just as COVID was starting to spread worldwide. I’d been living in San Francisco, in the Bay Area, for the previous 10–11 years, and we were going into lockdown there.
So I escaped. I thought, “Right, I’ll go to Barbados for a year.” And here we are—five years and three months later—and I still haven’t left. So, yes. That’s the only reason I chose Barbados, literally because it was trending on Twitter.
Gianluca Fiorelli: So let’s say, in a certain way, it was a marriage of convenience that evolved into a true marriage.
Martin MacDonald: Absolutely. I couldn’t have put it any better myself.
Gianluca Fiorelli: Okay. Thank you, Martin. It was a real pleasure—and great to see you again after such a long time. It’s true that with the internet we can talk every day, but seeing each other face-to-face—and hopefully soon in real life—would be fantastic.
Martin MacDonald: Thanks for the invitation. I appreciate it, and I look forward to coming back in six months and seeing all the things that were wrong with what we thought today about LLM optimization. That’ll be good fun. I’m very much looking forward to it—and thank you for the opportunity and the conversation today.
Gianluca Fiorelli: Sure. And thank you all for being here—patiently listening to me and Martin talking and conversing about SEO, AI evolution, enterprise, and all these things. Remember to subscribe and ring the bell to be notified when a new episode comes out, so the channel can grow. Thank you!
Podcast Host
Gianluca Fiorelli
With almost 20 years of experience in web marketing, Gianluca Fiorelli is a Strategic and International SEO Consultant who helps businesses improve their visibility and performance on organic search. Gianluca collaborated with clients from various industries and regions, such as Glassdoor, Idealista, Rastreator.com, Outsystems, Chess.com, SIXT Ride, Vegetables by Bayer, Visit California, Gamepix, James Edition and many others.
A very active member of the SEO community, Gianluca daily shares his insights and best practices on SEO, content, Search marketing strategy and the evolution of Search on social media channels such as X, Bluesky and LinkedIn and through the blog on his website: IloveSEO.net.
stay in the loop