AI Is Not Your Expert: Post-Purchase SEO, AI Pitfalls & Adaptive Strategy | Judith Lewis
Welcome back to The Search Session! I’m Gianluca Fiorelli, and in this episode, I have a great conversation with Judith Lewis—digital marketer, conference MC, and wine-and-chocolate expert.
We cover a lot of ground, especially how AI is impacting our work in SEO and digital strategy. Judith brings her sharp insights and humor to topics that are shaping the future of search.
In this episode, we talk about:
How AI is changing B2B and B2C search—and what that means for visibility.
Why post-purchase optimization is becoming a clever SEO play.
Using AI to speed up tasks and process data efficiently, but not treating it as a replacement for real expertise.
The risks of AI hallucinations—when it gives a convincing but incorrect near match.
How to design flexible SEO strategies that align with business goals but adapt quickly when needed.
Why old-school spam tricks might work for ChatGPT in the short term—but aren't a sustainable strategy.
All that and plenty more. It’s a fun and inspirational conversation filled with practical takeaways, real talk about AI, and a few great stories too.
Video Chapters
SEO Today: Stability, Strategy, and the Post-Purchase Opportunity
B2B vs B2C in the Age of AI Overviews and The Great Decoupling
How AI Can Speed Up Work and Where It Falls Short: Recognizing Hallucination Risks
Creating Adaptable Strategies That Support Business Goals — Without Getting Stuck in Rigid Plans
Transcript
Gianluca Fiorelli: Hi, and welcome to a new edition of The Search Session.
Our guest today is a 360-degree digital marketer. She doesn’t just do SEO—although that’s how many of us know her—she also works with social media, email marketing, and more. She’s fully engaged in all aspects of digital marketing, able to design strategies that are both nuanced and fresh, like a good Albariño, but her work can have a strong impact like a good Barolo. It’s tasty like a sweet chocolate, but also has that touch of irony to be bitter like a 99% dark chocolate.
Our guest today is Judith Lewis. Hi Judith, how are you doing?
Judith Lewis: Hello! I’m very good, thank you. That’s a lovely introduction. Everybody loves a good Barolo with dark chocolate, though—so, good pairing you’ve got there.
Gianluca Fiorelli: Yes, I don’t actually drink much wine, but when I do, I prefer red wines. You know, some white wines—depending on how they’re made—can give me a stronger headache, so I try to avoid white wine. My wife, though, is a big fan of Galician wine.
Judith Lewis: Ooh!
Gianluca Fiorelli: Barolo—sorry, not Barolo. I meant Albariño.
Judith Lewis: I was going to say—that’s an interesting area for wine!
Gianluca Fiorelli: Yes, especially the ones grown close to the coast. They have that slightly salty note in the flavor.
Judith Lewis: Yeah, no, they’re lovely. And she’s got great taste. I do love Albariño because it expresses itself so differently depending on where it comes from. It’s so nice and light and refreshing—it really cleanses your palate. So if you’ve had really rich, fatty foods and then a nice cool glass of Albariño, it’s beautiful.
Whether it’s from Spain or Portugal, that grape is just such a lovely wine. The grape really expresses itself well. So, good taste!
Gianluca Fiorelli: Thank you! I’ll let her know.
And just before we dive into digital marketing—so listeners can get a better sense of why you speak so knowledgeably about wines—would you like to share a bit more about what else you do and what you’re involved in as a digital marketer?
Judith Lewis: Yeah…I used to work at an agency where they said, “We’re a full-service digital agency.” And I said, “I don’t think you mean that in the way some men mean that...”
So yes, I’ve done email marketing, paid search, organic search, and social media. Email marketing is so much fun—it’s the unloved middle child of digital marketing.
But beyond that—because we all need a life outside our day jobs—I’m also a WSET Level 3 in wines. I wouldn’t say expert. Masters of Wine, they’re the real experts. But I am very well-educated in wine. I judge wine competitions: the Berliner Wine Trophy, Wine Lovers Wine Awards in Hungary, the Georgian Top 100 Wines in Georgia—the country, not the state.
Gianluca Fiorelli: Yes.
Judith Lewis: And several other wine competitions. So I’m very judgmental in search—but also in wine.
And I also sit on the board of directors for the Academy of Chocolate in the UK, which works to increase education about fine, good-quality chocolate. Not the low-cocoa-percentage stuff you sometimes find at the supermarket, but the higher-quality chocolate—which, in some cases, you can actually find at your local supermarket.
For instance, mine—well, I don’t know if I can see that. M&S Milk Chocolate, 46%. That’s a very, very good quality chocolate. And there’s actually a lot of good-quality chocolate you can find in your local supermarket; you just have to keep an eye out for it. And, of course, marketing plays a part in that, right?
Gianluca Fiorelli: Yeah, right. I mean, that’s also why bad chocolate tastes so good sometimes.
Judith Lewis: It’s the sugar and the fat content. There’s a lot of replacement fat—palm oil, for instance—that gets added to chocolate. Palm oil gives it that oily, fatty texture we love. And the sugar content is so high because it’s often around 18% cocoa solids, which means the rest is all milk, replacement oils, and sugar.
So out of 100%, less than 20% is actually chocolate—the rest is additives.
Gianluca Fiorelli: Yes. And addictive, because sugar is addictive.
Judith Lewis: Yep.
Gianluca Fiorelli: That’s why when people say they love chocolate, they end up wanting more and more—because it’s really about the sugar.
Judith Lewis: Exactly.
SEO Today: Stability, Strategy, and the Post-Purchase Opportunity
Gianluca Fiorelli: Returning to the world of digital—with a question I ask all my guests—how’s SEO treating you lately?
Judith Lewis: Luckily for me, I’m being treated very well by SEO and other forms of digital marketing. Your mileage may vary, of course, but SEO’s been good to me.
My clients aren’t necessarily seeing massive gains from the recent changes, but they’re not being hit either. They’re either stable or seeing a little growth—sometimes due to seasonality, sometimes thanks to the work we’re doing—but importantly, they’re not losing ground.
In fact, I’m working with some companies right now to increase their visibility across more of the chain of searching. For someone in SaaS or selling a product, everyone focuses on getting people to buy. But no one thinks about the post-purchase journey.
That’s what I’m focusing on with one of my clients: improving visibility for post-purchase queries. So, if people have questions about support or how to do something, we’re making sure they can find those answers in search results.
So people aren’t calling us up, they’re not bothering us with emails, they’re not costing us staff time. They’re getting everything they need directly from the search results.
So, not only are the latest updates and changes from Google, Bing, Naver, and Seznam treating my clients at least neutrally—or even positively—but everything is moving forward. People are starting to recognize the value of post-purchase optimization.
B2B vs B2C in the Age of AI Overviews and The Great Decoupling
Gianluca Fiorelli: Yeah, you’ve really summed up what a good digital marketing strategy should look like: a strategy that sees synergy between all the different channels.
In this case, you’ve highlighted the specific role SEO can play. Especially now, with all the changes we’re seeing—where classic click-through metrics are declining because of things like zero-click SERPs and AI Overviews.
We’ve seen how, particularly for informational queries, AI Overviews are really cutting into click-through rates.
How much would you emphasize this kind of holistic strategy for a company that wants to stay visible—knowing that maybe their traffic will be less, but wanting to make sure that the traffic they do get is of higher quality?
Judith Lewis: Yeah, I mean, we didn’t always convert on those kinds of queries. Most of my clients are B2B, but I do have a couple of B2C clients. And I think for B2C, it’s much easier to illustrate the impact because B2B tends to have lower search volume, and informational queries often still lead to a click—even with AI Overviews—because of the complexity of the query and the type of results needed.
They don’t just want a simple answer. They need a lot of information around it. And often, that kind of detail doesn’t come through in an AI Overview.
For example, it’s not just about knowing what a nonwoven textile is—they need to know things like the size of the extruded material that goes into making the nonwoven fabric. It’s more complex than what Google can summarize in an answer box.
However, for my B2C clients, where we’re seeing what many are calling “the great decoupling”—where we’re not getting as many click-throughs from informational queries—we weren’t converting super well on them.
One of my clients has been running a joke factory for many, many years. They do have some products placed in there, and they track whether those products actually sell.
They track clicks from that joke section into the main section to see if anyone’s actually converting. Conversion rates are extremely low.
Those are the kinds of lighter queries—not really informational in the B2B sense, but more casual questions about a product. And honestly, as long as the information is accurate in Google, I’d rather people get what they need there. It’s better than them clicking through, looking around, and leaving.
Because then we’re left wondering: were they really interested in that service or product and didn’t find what they needed? Or was it just an informational query—someone answering a question for an exam or a research paper?
In B2B especially, particularly in airline elements, where you're looking at PSS and things like that, there can be a huge volume of click-throughs that never lead to anything. They’re just not our customer.
And that really messes with my analytics because I can’t easily tell who I was actually targeting versus who just came through looking for an answer for a quiz or a term paper.
So, I’m not completely against AI Overviews. But where I do see them having an impact—especially on some of my former clients—I’d say it’s in industries like pharmaceuticals. I’m not working in that space at the moment, just by choice, but I’ve seen how AI Overviews can affect them.
For example, in the past, we’d see queries like “Am I pregnant?”—those show up in Quora, and they have significant search volume as well.
Now, obviously, Google can’t tell you if you’re pregnant. And this isn’t just a Google problem—it’s a societal one. If you don’t mind me recommending a book, I really love Everybody Lies by Seth Stephens-Davidowitz. It’s somewhere on the shelves behind me.
The book explains how people lie to others but tell the truth to Google. So, when someone types in “Am I pregnant?” in Google, at the moment, or the last time I checked, I wasn’t served an AI Overview for that kind of query. But that’s where the real danger lies: incorrect information being pulled into an AI-generated answer.
Something like, “If you’re sick every morning, then you must be pregnant.” Well, that’s a quick way to stress someone out unnecessarily. They might not be pregnant at all—it could be cancer, a food allergy, celiac disease, or something else entirely.
So AI Overviews worry me in the pharmaceutical industry. I’ve dealt with congestive heart failure—which is so depressing because, basically, you’re gonna die. I’ve dealt with pregnancy content, as I just mentioned, nappy cream, and so on—a whole range of topics, including maybe some things we shouldn’t even mention.
And I’ve worked across what I call the three Ps of marketing.
Gianluca Fiorelli: Yes, I understand you. The terrific thing, just to make a notice about AI Overviews and health, is that they are really present. On the Advanced Web Ranking website, there’s a free tool that monitors the evolution of AI Overviews' visibility in different industries.
One of the industries with the highest percentage of AI Overviews showing up in informational queries is Health. And that really surprised me because I remember Google saying, “No, we won’t show AI Overviews” for queries that involved your money and your life. Health is clearly one of those categories.
Evidently, they’ve decided not to show AI Overviews for things like health clinics, which should be eventually cited if someone is searching for information—like finding a clinic near you.
It’s not like a classic three-sentence encyclopedia entry. Otherwise, people would just start using Google as their doctor. Or even worse—using ChatGPT as their doctor.
Judith Lewis: Yeah.
Gianluca Fiorelli: That could lead, at best, to a massive increase in hypochondria—like a 2,000% jump.
But going back to what you were saying about your B2B client. That’s something I also noticed with one of my B2B clients. We’ve recreated informational content because one of the types of personas we target are people in that specific industry, but they’re not the engineers—they’re the ones buying the goods to supply to the engineers.
So, they usually have a good level of knowledge, but not deep technical expertise. And we have a kind of “canary in the coal mine,” so to speak. When we see traffic increasing from India—and this website targets the UK—that’s a sign. It means, okay, the informational content is working well, but the traffic that’s coming is mostly from people looking to scrape content or searching for quick answers.
You can even see it from the queries in Search Console. The variable queries usually look like they’ve been copy-pasted from somewhere, just like you were saying.
Judith Lewis: Yes.
Gianluca Fiorelli: And so, when we saw a lot of our content appearing in AI Overviews because of this, we weren’t too worried if overall traffic was diminishing. We noticed that the traffic to the more important pages was still holding steady.
Plus, because of that AI Overview visibility, even competing with really big multinational brands, we saw an increase in direct branded traffic. And from the homepage—where we monitor user behavior—we could see people navigating deeper into the site.
And after a very long search session—which can stretch over months in this case—we know they start micro converting. Maybe they complete a micro-conversion during the first visit, but from the CRM data, we know they finally convert after six months.
Judith Lewis: Yep. Yep.
Gianluca Fiorelli: This is industrial products, so yes—I totally agree with what you’ve said.
And for B2B—as maybe also for B2C—but especially on the transactional side of things, it’s not really a problem. The real problem is, as we know, for publishers. Especially for small, independent publishers. I’ve discussed this with many other guests as well, like Lily Ray and Duane Forrester.
How AI Can Speed Up Work and Where It Falls Short: Recognizing Hallucination Risks
Gianluca Fiorelli: And when you mentioned AI, we also, even if we didn’t say it directly, touched on hallucinations.
I really liked a talk you gave recently—people can search for it. It’s about… well, the title is AI-Powered SEO Report. That was a recent talk, and people can find it—just Google it. Of course, we’ll also include a link to the video in the description of our conversation.
In that video, you present what I think is one of the most fascinating uses of AI for us in SEO: using AI to improve and speed up our work. It’s about using AI to process data faster than we could manually, so we can focus on interpreting that data—creating real insights from it.
In fact, the core message of your talk is that the best approach is to combine AI with human expertise. And you also touched on hallucinations—and even gave some tips on how to recognize when AI is basically joking with us, when it’s hallucinating. Would you like to share a few insights about that here?
Judith Lewis: Sure. I mean, I loved doing that talk—it was so much fun. I really wish everyone could see it. If they get a chance to search for it, it’s definitely worth a look.
One of the tasks I had AI do was based on something I hear a lot: people saying, “Oh, AI is taking my job.” People are starting to use AI as an expert instead of actually asking an expert. And I think that’s a really interesting tack.
If you look at some of the analyses of how AI is being used, that’s exactly what’s happening. People are asking AI to be the expert instead of them. And when you don’t know what the answer should be, it’s very easy to get fooled into thinking AI is giving you great results.
One example I gave in the talk was when I asked ChatGPT what I needed to know about “Sequoia.” And it said, “Well, Sequoia could either be a tree or an investment firm.” And I was like, “No, no—it’s the macOS.”
So I asked again, “What do I need to know for my laptop about Sequoia?” This OS came out in 2024, so it’s not new.
And it still replied with, “That doesn’t seem to be relevant to a tree or an investment firm.”
Eventually, I even gave it a URL, and it still got hung up. It just didn’t understand. And that’s one of the biggest risks with using AI in this way. If I hadn’t known exactly what I was asking about, I might’ve just accepted whatever answer it gave me.
That’s actually what happened when I tested it with Hotel Chocolat. I said, “Right, do a technical audit for me.” I’m feeling lazy, I’m tired, I’ve got a thousand other things going on. Or maybe I’m in an agency and my boss is telling me I need to do this and this and this. So I think, fine, I’ll ask ChatGPT: “Give me an audit of the Hotel Chocolat website.”
And it says, “Yes, no problem, here you go.” But then I started asking it expert-level questions. I said, “Where did you get these orphan pages from?”
And it said, “Well, I took my best guess. This might be it, but I can’t guarantee it.”
Then I asked, “Can you give me some example metrics I’d need for a technical or a monthly report?” And it said, “Okay, sure—here’s your organic traffic, here’s referral traffic, here’s some direct traffic, and here’s how it’s up or down month-over-month.”
And I stopped it right there and said, “Wait—where did you get that data? I haven’t linked anything to you.”
And it replied, “Oh, the data I just gave you? That was just my best guess. It’s not real.”
But it hadn’t said that! That’s why I’m very cautious about using it in that way.
Now, taking large chunks of data in a closed environment—taking big amounts of data and trying to make sense of it—that’s a really good use for ChatGPT or similar systems.
However, my point was: we, as professionals, should be the ones interpreting the information that comes out of these tools. We already have everybody.
We’ve got AWR, we’ve got Ahrefs, we’ve got SEMrush, SE Ranking—the people who did the reporting and AI video. We’ve got all these tools already. We are the experts. We take the data and interpret it.
ChatGPT is going to lie to you. It’s going to say, “Hey, I found these orphan pages,” or, “Here’s the data for the number of visits each month.”
And when I saw that, I was like, “Excuse me—you what? Where did you get that from?” “Oh, I made it up.” Thanks, babes. That’s not a really useful report, then, is it?
So, the video really does illustrate this well. But the hallucination part of it—it’s something we’ve been dealing with since the dawn of search engines, and that is near matches.
Think about it: we’re always searching, looking through results to find the right match. We have the term pogo-sticking—when someone clicks into a search result, pops back out, clicks into the next one, maybe stays, maybe leaves again. That behavior teaches Google what is or isn’t relevant for a specific query.
That’s essentially a near-match failing. And what happens with AI-generated responses—these conversational, almost human-like answers—is the same kind of thing. It’s not quite an exact response—it’s more like something being dictated to us.
What happens when we only get one or two answers, like with Sequoia? Well, it got it wrong and wrong again. We’re not given the opportunity to look down a list of options, like I showed later, to see what Sequoia actually matched against in the search results.
And as a result, that search result that I was looking at gave me what I needed—it showed all the near matches. That’s what a hallucination is: AI picks one thing, and it’s the wrong thing, but it’s a near match.
And I don’t think people always realize that. It’s basically what we’ve been dealing with for years in Google—we just didn’t really have a name for it before.
Gianluca Fiorelli: Oh yes. Yeah. It’s the classic problem of disambiguation.
Judith Lewis: Yes. Yeah.
Gianluca Fiorelli: And yes, I think you explained it really well about hallucinations. Hallucinations are like good lies—a good lie always has a little bit of truth in it.
Judith Lewis: Yes, the best lies always have a little bit of truth.
Gianluca Fiorelli: Exactly. And that’s why so many people fall for these hallucinations. They only notice the really obvious ones—like the classic hallucination about glue and cheese on pizza. But those kinds of things were everywhere about a year ago when SGE came out and became AI Overviews.
In fact, when it comes to using AI in technical SEO, I personally prefer not to rely on it too much. The only time I really use it is when I’m dealing with a huge amount of data and I don’t want to lose my eyes trying to dig through all of it manually. So what I do is format the data properly, and then I send it to Gemini. Lately, I have even just used Gemini directly inside Google Sheets.
It’s faster, of course. But if I upload it to ChatGPT, I always make sure to explain the context—what this data is, what each column means. I’m very precise. Then I ask very specific questions—the same kind of questions I would ask myself if I were analyzing things manually.
Especially when it comes to finding patterns—that’s really where these tools are super useful.
But when it comes to the classic question: “What would you do?”—that’s not something I ask AI. Instead, I might ask, “What hypothesis do you think could explain this data?”
That’s more like a brainstorming exercise. From there, I can think: okay, this could be true, this might not be, or maybe it’s something slightly different. But I never ask AI for solutions.
And this ties back to what we already know as SEOs. We all love tools like Screaming Frog or Sitebulb—they guide us with their insights and suggestions. But we also know that many times, what they immediately flag as a problem—yes, it’s a problem, but it might not be the real issue. There could be deeper issues behind what the crawler signals, ones that are more important and may actually be causing that surface-level problem in the first place.
So, it’s the same grain of salt we use when dealing with crawler tools—we should apply it even more when working with something that isn’t as specialized as a crawler.
Judith Lewis: I would say—oh, sorry.
Gianluca Fiorelli: No, no. Go ahead.
Judith Lewis: I was going to say—my father was a lawyer before he retired. And the best example I have of how hallucinations happen—because near matches are difficult and the algorithm is just doing its best to match what we’re asking—is actually from law.
What they did was create a restricted database made up solely of case law. That’s it—just remedies, tort, and other kinds of laws. A single, specific area. They uploaded all of that and then asked for cases related to a specific topic.
And even then—even in that tightly controlled environment—AI got it wrong.
It couldn’t find an exact match, so it gave a near match that was completely inaccurate. It took the judge from Case A, the defendant from Case B, and the plaintiff from Case C, and mashed them all together—because that’s what it thought matched the query best.
So even with a highly restricted dataset and an algorithm specifically designed to stay within that scope, it still hallucinated—and badly.
I think it was Dawn Anderson who talked about some research showing that hallucinations actually get worse the more specific we get.
So, hallucinations, well, I hate calling them hallucinations. The near-match algorithm mistakes happen more frequently as queries become more specific because the algorithm struggles to match like-for-like.
So, when we try to be very precise, the AI does its best to match what we're asking—but based on how it understands it. And that increases the chances of an incorrect best match. In other words, it gives a near match instead of an exact one. And that’s what we end up calling a hallucination.
Law is the best example of that. Even in an incredibly restricted environment, it still couldn’t get things right. That’s why I always say: be careful, be cautious—and remember, you are the expert.
I used Screaming Frog for a client, scraped the competition, and fed it all into ChatGPT—title tags, meta descriptions, and so on. Then, I asked ChatGPT to classify them into groupings, and then I had to go in and correct it.
That said, it did the bulk of the work much faster than I could have done manually. So yes, you can use it—but no, you absolutely cannot just send it off to the client without reviewing it first.
Gianluca Fiorelli: Totally agree. Because otherwise… well, it becomes pretty embarrassing to justify the mistake.
Judith Lewis: Yes.
Gianluca Fiorelli: “It was my assistant… named Chat, surname GPT.”
Judith Lewis: It was—I don’t know—Chat… Cha? It was my cat!
Gianluca Fiorelli: He’s a French collaborator.
Judith Lewis: Yeah, Cha. He got it wrong. He’s an intern now.
Gianluca Fiorelli: Yeah, yeah. He’s not working with us anymore.
Judith Lewis: Just blame it on Bob.
Creating Adaptable Strategies That Support Business Goals — Without Getting Stuck in Rigid Plans
Gianluca Fiorelli: I just want to go back for a moment to the topic of strategy—because I love strategy. I think it's the most rewarding part of our job. Of course, there's the implementation, working with the client, and all of that. Sometimes I feel—I don’t know if you know the movie Amadeus, about Mozart?
Judith Lewis: Yeah.
Gianluca Fiorelli: There’s a quote in it that I love. The people at the court are asking, “Where is the opera? Where is the concert?” And he says, “It’s here, it’s here,” pointing to his head. But they respond, “Yes, but where is it? We have to play it. Where is the paper?” And he says, “It’s just a few notes to put on paper.”
And I relate to that. I love having the strategy fully formed in my mind and then putting it down. But the project also has to live within the client’s organization—it has to be responsive. Otherwise, I’ve put in all this brain effort to design what I believe is the best approach, and it ends up taking forever to bring to life.
That leads me to something I’ve become more aware of lately: related to how quickly everything is changing now. So, how do we design strategies that aren’t rigid?
How can we build something structured enough to guide the work, but flexible enough to adjust quickly when needed? Because sometimes, strategy gets so well-architected that it becomes too rigid.
I have a feeling you’ve had the same experience—and probably asked yourself the same question. So, how do you answer that question when it comes up for you?
Judith Lewis: I think that’s a really difficult question—because so often, it’s the clients themselves who ask us for the strategy. And they want a year-long plan. They want to know how to resource people.
And that’s where the rigidity comes in—they need to know what level of resourcing they’ll need to achieve what’s been laid out in the strategy.
So, when I’m working with clients—like I am right now on a new strategy—I start by looking at the overarching goals of the company. Everything I do is underpinned by data. As all of us do… I hope. I’m looking at you, kids. Everything we do should be based on data—or at least, most of it—not on wishes, hopes, and dreams.
So it all ties back to the company’s goals. If the company’s aim is to be seen as a leader in airline technology or a leader in microphone technology, then the question becomes: how do we do that?
What are we doing, as SEOs, to support that goal and help boost the perception—at least through search results—that this company, Samsung for example, is the leader in microphone technology?
That’s the wiggle room in there. But the real challenge is resourcing, right? And if you have to pivot at the last minute. Well, I might have people scheduled. Real human beings who rely on this work, whether they’re part of the company or freelancers, for their income, for their rent or mortgage.
So what happens when those eight articles that were scheduled for this month suddenly become irrelevant? Because someone at Google presses a button—as Danny Sullivan has done more than once—and everything shifts.
Now, those articles aren’t going to help us at all. In fact, they may be a waste of time.
That’s where the rigidity of strategy becomes a problem. We need to be able to brief in eight new articles quickly. Maybe not all at once—but maybe two right away, two more a few days later, and the remaining four before the end of the month.
At the same time, everything still has to tie back to the overarching strategic goals of the company.
Are they aiming for £25 million in sales? Is that the target? Are they trying to be seen as the world’s leader in soap?
You know, it could be something mundane—or it could be something sexy. Maybe they want to revolutionize how we approach DNA research. Or maybe they just want to improve people’s experience going through an airport. And I said “just,” but I shouldn’t—because a smooth airport experience is actually really important.
But it doesn't matter what it is, as long as we're relating it back to the overarching target and strategy of the company itself—each step down.
So the overarching SEO target might be: I want to publish 200 articles, 150 of which underpin the company's positioning, and 50 of which are direct response. So, we aim to make money right away.
And then, because you've been broad there, the specific ones underneath that become: these are the topics or the articles that we're going to write. So, each step down gets more specific.
It does require us, as SEOs, to step back out of our comfort zones—whether that's technical SEO, or content SEO with keywords, topicality, and entities, or even link builders focused on getting links from relevant websites that actually send traffic and generate sales.
We need to look at the bigger picture. And I know that's not a comfort zone for a lot of us. But if we can step back and see that bigger picture, and then align our strategies, our tactics—to that overarching strategy, each layer down we go, if we’ve given ourselves enough wiggle room, it means that when we hit that month where Google’s like, “Well, screw you”—which honestly, I feel like they’re doing to me personally—we can pivot really quickly.
Because we’ve said, “Well, we’re targeting eight articles. Those eight articles are no longer going to make us any money. Let’s change that”.
And when you say it like that—to the CEO and everyone below—“These articles are no longer going to make us the money we targeted them to make,” everyone will be on board with you.
But if you say, “Well, the keyword optimization for the entity relationships we were originally targeting is no longer valid, and we need to develop a new entity framework to address that,”—nobody’s going to be listening.
If you say, “They’re not making us the money we expected,” that’s when they’ll pay attention. So always relate everything back to money.
But remember the stages: start with the big picture for the whole company, then the big picture for your department—whether that’s SEO, email marketing, or whatever.
And then, you know, what are the tactical elements that will help achieve that big picture? And underneath that, how do you break down each of those tactical elements on a per-month basis?
And here’s the most dangerous part—never do this: don’t assign exact monetary values to each tactical element.
For example, if you say, “I’m going to spend £200K on articles, and I expect—based on the search traffic they’ll bring in and an assumed 2% conversion rate— an average order value of £6,745.” Or euro or whatever. Obviously, I made that number up—it’s too low for B2B and probably too high for some B2C, unless we’re talking luxury goods.
But if we do think in that way, then we can say, “Right, this isn’t meeting the goal anymore—those numbers are no longer matching. So, let’s move things.” But never share those money numbers with the CEO. Ever. Ever, ever, ever.
Internally within your group or team? Yes. Make everything relate back to money. And then at the end of the year—report on that. Report on that money.
You will look so good, so hot, so amazing, so sexy to the C-suite and to everyone below. Whether you're in-house or at an agency, you'll not only keep your job—you’ll get bonus, you'll get retained, you'll get the love—because you made the money count.
So always think about it that way. Is this making me money? Is it making the company money? Is my boss going to look good because of this? That’s how I think about it.
And that's what we’re doing right now with one of my clients. We started by asking: What is the company aiming for right now? What is SEO aiming to support? And that’s why we’re now looking at post-conversion. So, post-purchase, we're creating a large number of articles that don’t make money initially—but they do make money by reducing the churn rate.
Right now, we have a percentage drop-off, and our goal is to lower that percentage. And we’re doing it without directly communicating it to the CEO. Because we don’t want to get in trouble for not targeting what they thought they wanted. Instead, we’re focusing on what’s actually going to make us money. And retaining a customer is always cheaper than acquiring a new one.
Gianluca Fiorelli: I agree. To sum it up, there are two key concepts here.
First, avoid what I call sales interest. When I say that, I compare it to the Middle Ages—when people believed everything revolved around the Earth. No, nothing revolves around SEO. Usually, we’re not the center. The sun is the client and their business needs, and we’re one of the planets in that system.
And second, don’t focus only on the single tree. Try to see the whole forest. Because otherwise, you’ll miss the fact that there’s more forest beyond—or that the forest ends just behind the tree you’re staring at.
The Myths of Generative Engine Optimization
Gianluca Fiorelli: And now, let me ask you another question. When I was reading your bio on LinkedIn to prepare for this interview—of course, I know you, but I wanted to be a good student and make sure I didn’t miss anything—I noticed something: next year marks your 30th anniversary in digital marketing? If I’m not wrong, you said you started around 1996?
Judith Lewis: Yep. Yeah.
Gianluca Fiorelli: So here’s the question—because you have such a long and broad experience, not just in terms of time, but also in the variety of roles and changes you’ve seen in the industry.
We’re now entering this new phase of search—let’s call it AI Search. I’m not going to get into all the acronym nonsense, because honestly, it is nonsense. But they call it AI Search. And every era in search has its myths.
Back in our day—when we were all just getting started and breaking websites—we had things like the keyword density myth.
I’m curious—are you already seeing any new myths emerging around AI that make you just roll your eyes with that “oh please” kind of look?
Judith Lewis: I love that—yes.
My younger brother is a mathematician, so he and I often talk about algorithms. And I’ve worked in machine learning too—specifically building algorithms for agriculture. It’s not the sexiest industry, but it’s essential—because it’s how we get fed. In that case, it’s how we get wine.
But a lot of people think they know what’s going on, and I don’t understand how. I mean, even I don’t know everything that’s going on. So the biggest myth right now is that someone can be an expert in generative engine optimization or AI.
First of all, the debate about whether this is even AI or not still hasn’t been settled, in my opinion. Is it artificial intelligence or is it just deep learning or machine learning?
I don’t think we’re using the terminology correctly. And because of that, people are calling themselves “experts.” And that is the myth.
There are no true experts—at least not outside of the people working directly in the math area. They’re the experts.
But a marketer? A marketer is never going to be an expert in an algorithm so complex that even Google’s own engineers can’t currently explain why search results behave the way they do—because the system is constantly learning.
So you can never be an expert in that. There will almost never be an expert in generative engine optimization.
Now, I’m saying that today—in five to ten years, maybe I’ll be proven wrong. Maybe there will be guardrails that will ensure that we actually can be generative engine optimization experts.
But right now, as things stand, it’s incredibly difficult to call yourself an expert and be taken seriously. Why?
First, it’s a brand-new, rapidly evolving area of technology. And if you consider the logarithmic rate of change in functionality, even engineers are struggling to keep up—let alone marketers.
And second? You didn’t do the math. So, I really don’t believe you.
That’s my biggest bugbear: people calling themselves experts.
The other myth I see is the belief that you can optimize for AI Overviews. I would say that there are methodologies you can incorporate that may improve your chances or potential to appear in AI Overviews.
But much like in the later stages of Google, you can do a lot of things—and sometimes they just don’t result in anything because you missed something along the way.
So I think the myth that you can magically or automatically guarantee inclusion in AI results—like we used to do with Google—is, honestly, massively premature. Possibly even a myth, and therefore incorrect.
Experiments, Spam Tactics, and the Limits of Persuasion
Judith Lewis: We can improve our chances, but there are no guarantees in life. There were never any guarantees in Google, and there are no guarantees in life, and there is no sense in trying to guarantee your way into ChatGPT’s results.
And if you remember, we did an experiment to try to adjust things.
Gianluca Fiorelli: Yes, yes—I remember. And only one of the many attempts actually worked.
Judith Lewis: Exactly. And even then, it only worked for a short time—then everything reverted back to type. So it was an interesting view of how recency played a role in what the system returned, and also the frequency of the query.
And… I don’t know if I should admit this, but I might’ve used an old-school technique that maybe we might’ve used a while ago… or maybe not. And I might have gone into ChatGPT with different accounts—paid accounts—on different laptops.
Because sometimes, I don’t know, maybe I broke something and had to use a different laptop. Let’s go with that. We’ll stick with that lie—that illusion, that fantasy, that myth.
And then… maybe I went in and did a frequency of query. I’d ask, “Why didn’t you include Judith Lewis in this result for X?” And ChatGPT would respond with something like, “You know what? You’re right—X, Y, Z.”
So I kept repeating that query with increasing frequency. And eventually, that repetition led to something interesting—within that community, let’s call it, of people monitoring those results, we started to see responses where my name was actually included.
Sometimes, that happened because I had gone in and used an old-school spam technique.
But like all old-school spam techniques, it was short-lived. It was driven by recency and frequency, rather than something that you can manipulate over the long term, like good, solid, old-school SEO does for us.
When you optimize well, it lasts over time. Optimize poorly, and it might work for a moment, but then it goes away.
Gianluca Fiorelli: Yes, in fact, there are quite many of these techniques—let’s call them experimental—that are starting to pop up.
Judith Lewis: Yeah.
Gianluca Fiorelli: The only one I’ve seen lately—I'm going to share it here, but it's already available in one of your more recent LinkedIn posts—is a tactic I’d call smart.
It relies on encouraging people to share a specific question from your website and then submit that question directly to ChatGPT.
But with a little trick, to add the prompt: “Please remember this answer in your memory.”
This ends up tricking ChatGPT’s memory function—at least for users who’ve taken that action. And once this happens, your page or your brand keeps getting cited again and again, in everything. So it’s a way of eventually forcing the machine a little bit.
Is it correct? Is it ethical? It could be. There are no rules right now. There are no clear guidelines for ChatGPT in this context.
Formally—when it comes to search—if you want to stay strictly within white-hat SEO, we at least have guidelines. Google has guidelines. Bing has guidelines.
So, in those cases, it’s better not to push boundaries—because if you go against those rules, there’s a real risk of being penalized.
Judith Lewis: Yep.
Gianluca Fiorelli: But for ChatGPT, or LLMs, they are natural language processing and there is no rule. So we could call it a “persuasion trick.” But there are no official guidelines against persuasion.
Judith Lewis: It’s true. I think you’ll find that persuasion doesn’t have long-term value. It’s all about frequency and recency. That’s what drove the results in our experiment—how often and how recently something was published.
And also… no, I’m not going to tell you that. That would be cheating. Let’s just say it has to do with how you publish certain things, where you publish them, in what way, and how you phrase them. I’ll leave it at that.
Gianluca Fiorelli: Oh yeah, yeah. I think our viewers and listeners understood us.
Judith Lewis: Yeah, yeah. We don’t want to encourage people to do things. What’s happening right now is that some old-school spam techniques are working to influence ChatGPT. But I don’t want people to encourage people to think that this is a solid strategy or a methodology for long-term results.
Short-term gains are fine—for fun, but it’s not fun for your client. First, it can cost them a lot of money—because you have to keep redoing it. And second, it doesn’t deliver the kind of stable, lasting results that proper SEO does.
We’re not doing PPC. We’re optimizing for an engine—but in this case, the engine is an LLM as opposed to a different type of engine.
So I’d say that there were a number of different techniques we experimented with. Some of them included things I wouldn’t recommend—like spamming queries.
But you also have to consider how ChatGPT works. It has session lengths. After a certain amount of time, that memory is flushed—so you’re not going to see a lot of what we were doing held in memory.
And even if we say, “Remember this,” there’s a particular period of time after which ChatGPT won’t remember it. And we could probably demonstrate that using that joke—that test we ran—which clearly showed how recency and frequency encouraged results. But look at how quickly it dropped off.
For example, I once got a link off of The Telegraph, and it boosted my personal blog massively—hugely—while the article was still on the homepage.
But as soon as that article moved off the homepage, the impact started to drop in terms of how it was affecting my site.
It was ranking my site for all kinds of terms—many I wasn’t even interested in—just because those terms were covered in the article and mentioned on my site in places, in an off-handed sort of way.
It was like a near match—almost like a hallucination from ChatGPT or AI Overviews. It did help—briefly. But then, it quickly tailed off. And that drop-off happened fast, simply because the article moved away from the homepage.
And so the amount of value that was being passed by that link dropped—and hopefully I’m not saying anything people don’t already know about how links work. But that’s a really good example of frequency and recency in ChatGPT’s results, how we were able to spam it, and why it didn’t last.
As soon as we stopped the frequency with which everyone was posting—and those posts began to age—the effect started to tail off. Then it went back to form, showing what it had originally shown.
So it’s a short-term, spammy technique. I wouldn’t recommend it to a client. I wouldn’t recommend it as a methodology for anyone else either.
Sure, it’s fun—like when we used to call it “graffiti” back in the day, changing Google’s search recommendations. You know, when you’d type “chocolate chip cookie” or “chocolate chip” and it would finish with “cookie recipe,” at least for me. And then you could put something else in.
We called that graffiti. But don’t do that for clients—not for long-term goals because it doesn’t last and ot’s too much work. You could be spending your money better elsewhere.
Gianluca Fiorelli: Yes, totally. And I think that—just like in the old days—we were doing experiments. I mean, the wiser ones among us weren’t doing it for clients; we were running tests on our own experimental websites.
Judith Lewis: Yep.
Gianluca Fiorelli: That was how it worked back then, because Google was still relatively young. We didn’t have the whole corpus of documentation on Google for Developers like we do now. So we had to really try things out to understand what the alchemical recipe was to get visibility.
So we started doing all kinds of dirty stuff. But it wasn’t because we wanted to be spammy. It was about pushing the limits of the system to see how much we could do and how much we could integrate the mechanism.
And yes, you can try this kind of experimentation on your own experimental website—but never with a client's site. Otherwise, you're going to lose the client. Not only will you lose visibility in ChatGPT or because of everything we just talked about, but you'll also lose the money the client invested.
Judith Lewis: Exactly. Yep.
Gianluca Fiorelli: I was just looking at the clock—we’re already done! That went by fast.
Judith Lewis: Wow! It’s because I just go on and on.
Gianluca Fiorelli: No, it’s because you’re interesting! Otherwise, it would’ve felt like one of those never-ending hours. But you say such wonderful, interesting things.
The Fireside Questionnaire
Gianluca Fiorelli: Before we wrap up, just one quick question—so people can get to know you a little more. Well, people already know about your expertise in wine and chocolate, but tell me: is there a wine that surprised you the most? A really good one?
Judith Lewis: Wow. The wine that surprised me the most… oh, I’d have to say Xinomavro. I was at a conference a few years ago in Plovdiv, in Bulgaria. You fly into Sofia, then take a bus to Plovdiv. There was a press tour to the Northern Greek area before the actual conference, and that’s where I had the Xinomavro—served in a magnum bottle. I never thought I’d become this much of a wine snob, but it was absolutely beautiful.
It was rich, full of deep red fruits. Imagine a bowl of ripe, dark red cherries—all pitted—so you could just scoop them up with a spoon. Add to that some raspberries, maybe chunks of really ripe, rich plum, and just a little hint of prune. Not too much, just a touch. Every mouthful was luscious, juicy, and smooth. Not too many tannins, not too acidic. Everything was really what they call “in balance” because it’s not too high anywhere. You could taste all those fruits coming through, and it was just gorgeous.
But there were still tannins, and there was still acidity—which meant you could have stored it for a while under the right conditions, and it would still have been this beautiful melody of fruits coming through. It was so delicious.
I had never tried it before, and it really surprised me. I was so used to international varieties—wines from Geneva, wines from Switzerland in general, Spain, France. These are the usual kinds of experiences most people have. I’d had some German wines too, but this was my first encounter with a truly rich, beautiful, and amazingly expressive wine made from an indigenous grape variety—something not considered “international.”
You know, like Cabernet Sauvignon—that’s grown everywhere, and we call that an international grape. But this Xinomavro was just deliciously amazing. It completely surprised me.
Since then, I’ve really sought out wines from Eastern Europe, Greece, Turkey, and other regions with lesser-known indigenous varieties. That’s actually how I ended up in Georgia—the country—trying Saperavi and Qvevri wines. Again, they were surprising, though I think they don’t have quite as broad an appeal as something like Xinomavro does as a red wine. They’re a bit more challenging to the palate.
They take you on a journey. They start you in one place, lead you somewhere else, and then finish in a completely different place from where you began. Wines like that require your attention and concentration. But most people just want to sit back and sip something that's simply delicious. That Xinomavro—that was the one that surprised me the most.
Gianluca Fiorelli: And speaking of chocolate—what’s your first memory of falling in love with it? Maybe back when you were a kid in Canada, where you were born?
Judith Lewis: It’s really hard to pinpoint the moment I fell in love with chocolate, because as a child, it’s just such a beautiful, sweet treat. It’s really delicious.
I have to admit—and I know some people might come at me for this—but After Eights. When I was younger, I swear they were more delicious! It felt like the dark chocolate was of higher quality—or maybe that’s just how it seemed to me as a kid. But that beautiful, almost liquid peppermint center, wrapped in a very, very thin chocolate coating. And you'd bite into it, and that oozing center would just pull away… you’d get that immediate melt of dark chocolate and peppermint. It was so delicious. Honestly, it’s probably why I have so many fillings now! I really loved After Eights. For me, at Christmas, that was the special treat.
But when I’ve gone back to try them again—they’re just not as good. Same with Turtles, which is a Canadian thing. They're basically blobs of caramel and pecans coming out at four corners, so they kind of look like turtles, all smothered in milk chocolate. I swear they don’t taste as good as they did when I was a kid.
Maybe it’s just childhood nostalgia, or maybe the quality really was better back then. But yeah, After Eights and Turtles—some of the cheapest chocolate out there—that’s where my love of chocolate began.
Gianluca Fiorelli: Yeah, same for me. It’s the classic—like Nutella, back when Nutella was really good, with not so much palm oil.
Judith Lewis: Yeah.
Gianluca Fiorelli: I mean, 30 or 40 years ago, when I was a kid, the best snack was just a piece of bread with Nutella on it, and I was so happy.
Judith Lewis: Yep. Totally.
Gianluca Fiorelli: So, to wrap things up—you MC many conferences and you also speak at many of them. So, let’s say after the summer ends—where are you going to be? If people want to see you in real life, where should they go?
Judith Lewis: Oh wow, okay! So, I’ll be at the SEOFOMO Meetup on September 4th in Barcelona—that’s going to be fun. Then, after that, I think it’s Digital Olympus on September 11th, in Amsterdam.
And… where am I after that? I’m literally going through my phone to check because I don’t even know where I am anymore. Oh! Right, DudaCon. I’ll be speaking online at the Duda Conference. Then in October, I’ll be at Search & Stuff in Antalya, Turkey.
And honestly, even just as an attendee, I’d go—it’s at an all-inclusive resort. Plus, if you’re single, it’s great. My husband doesn’t come with me anymore. He’s just not as into these things as I am. But seriously—for three nights, under 800 euros for a single person at an all-inclusive resort on the beach in Antalya? I think it’s even five stars! So yeah, that’s amazing. I’m really looking forward to Search & Stuff.
And then immediately, hot on the heels of that—and the reason I can’t stay late at Search & Stuff to get more of a tan—is the SERP Conf. Vienna.
Gianluca Fiorelli: Yes.
Judith Lewis: That’s always such an incredible conference. SERP Conf. Vienna—I really, really enjoy that one.
After that, I think it’s the SEO Mastermind group, which is happening in Dublin. But that one is invite-only. So if anyone’s interested, you’ll need to go to the SEO Mastermind website and look specifically for the Dublin event—there are lots of Search Masterminds out there. Once you find it, you can apply to be considered for the invite list.
Then it’s Brighton. I’m not speaking there, but I’ll probably be around.
And finally, at the end of October, the event I’m super excited about: Spotlight. I’ll be at the SEMrush Spotlight Conference! So that’s going to be exciting!
And in between all of that—I'm not sure if people will be interested—but I’m also doing a session in October on how to judge chocolate. That’s going to be virtual and probably only about a Fiverr to take part in, but unless we decide to do a chocolate package for participants, in which case, we might have to charge more.
So yeah, it’s a very packed September and October for me.
Gianluca Fiorelli: Yeah, I see that!
Judith Lewis: But it’s going to be fun! And I hope I haven’t left anyone out. If I have, I’m desperately, desperately sorry—I’m literally looking at my phone now to double-check. I think I’ve included everything. Oh! And I might be teaching a course somewhere, too. But yeah.
Gianluca Fiorelli: You’re really busy!
Judith Lewis: Yeah.
Gianluca Fiorelli: Indeed—thank you, thank you so much, Judith. It’s been a splendid hour with you, and I’m sure our viewers and listeners really enjoyed spending this time with you.
Judith Lewis: Thank you so much!
Gianluca Fiorelli: You're welcome. And to all of you watching—you know what to do! If you enjoyed this wonderful episode with Judith Lewis, give it a like, ring the bell to be notified about new episodes, and if you want this channel and this series to grow, please subscribe.
That way, you’ll always be updated whenever we publish something new. Thanks again—have a great day, and see you in the next episode of TheSearch Session!
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