Prabin Yonzon is the Head of Organic Search and CRO at Pure SEO.
In our latest Good Content episode, we sat down with Britney Muller, a distinguished authority on machine learning, AI, and SEO, to discuss the recent AI boom and its influence within the digital marketing realm. Rollan Schott, our Director of Content Marketing, and Britney delve into an expansive discussion around AI’s impact on marketing, AI’s influence on search and content marketing, and where business owners and marketers looking to succeed can start.
You can listen to the whole podcast or the key sections below! Scroll down to read the full transcript and find useful resources mentioned in the podcast.
You’re listening to Good Content, the official podcast of Pure SEO, New Zealand’s most-awarded search agency.
To make sense of the AI boom in 2023 is to be perpetually behind on the latest news. Advancements are occurring faster than businesses can adjust. Yesterday’s big, audacious goals are today’s yesterday’s news. Countless businesses globally are experiencing the disorienting, thrilling changes caused by the AI boom, but few industries have felt these disruptions more than the field of digital marketing.
How can businesses future-proof for an uncertain future while adjusting to a destabilised present? How are rapidly evolving AI technologies changing the world of search? What are LLMs? What do marketers need to be aware of? What should you do first if you hope to get started?
I’m Rollan Schott, Director of Content Marketing at Pure SEO. We’re joined today by Britney Muller. Britney is a recognised leader in the fields of machine learning, AI, and SEO. That’s precisely why Pure SEO reached out to her to help drive our AI initiative forward. It’s wonderful to have Britney here in New Zealand, speaking face-to-face.
Hi Britney. Thanks for joining us today. Do you want to start by telling us a little bit about yourself, how you found yourself one of the foremost authorities on AI and SEO, and how you came to find yourself here with us in Auckland, New Zealand?
Yeah, that sounds good. So, at a very, very high level, I essentially live at the intersection of SEO and AI, or machine learning. And I came about through a very unconventional path of sort of doing SEO for fun when I learned about it after college, and then doing it for a job, getting fairly good at it, starting an agency, moving in-house with Moz in Seattle, going back out on my own—and then I worked for an AI company last year, called Hugging Face.
And about, I would say, eight years before that, I was “all in” in machine learning. And so, I’m excited to see the huge explosion of AI in the last couple of months and be able to participate a bit more in the marketing conversation of it, whereas it wasn’t such a hot topic just even a couple years ago. So yeah, super excited to see those two worlds collide and look forward to going more down that path and applying it.
Mass adoption of AI—and then the evolution of these AI technologies—they’re kind of feeding each other. We’ve got this pretty wild snowball effect that’s been very disruptive to the industry. I’m curious if you saw this coming, and if so, how long ago?
Yeah, it’s so funny—I knew machine learning was the future—I didn’t think it would erupt quite like it has been the last couple of months. I thought it would be a bit slower.
And in terms of actual public adoption, I think the speed at which we’re going at isn’t super realistic or supportive of actual AI applications. We need to educate non-technical people and public stakeholders along the way to make sure that this is done in a safe and appropriate way and use cases.
So that part of it I’m very interested in. And it’s been really fascinating, again—just even the last couple weeks or months—around the governance, the AI ethics, the different applications, and the common misnomers of, “this is kind of the golden bullet that solves all your problems”, which certainly isn’t the case.
There’s a lot of things that most people just aren’t aware of in terms of the technology, common pros and cons, and navigating that changing landscape. It’s so important to have a core compass upon which you can really move through this stuff.
You’ve spoken a lot in your time with us about safety and the risks involved [with AI] and the shortcomings and biases. A lot of these issues we still have to confront in some way, and people are still working on them. It’s interesting to me to hear so much of that [the risks of AI] from someone who’s so excited about all the changes we’re experiencing.
How do you bridge that gap and manage the two sides of the blade here? Because there’s a lot to worry about, but there’s also so much to be excited about.
Oh, absolutely, there’s so much to unpack on either side of that. And I think where I like to operate is somewhere in the middle, right? Awareness of both polar sides of this technology—what it’s really, really good at, what it’s bad or harmful with.
What I have gotten so, so excited about the last several years even, is that people, non-technical people, hold the power right now. This technology is becoming more and more accessible. It will be non-technical people who think of high-value applications—and what they do every single day—that take this tech to the next level.
It’s not going to be the engineers that developed this—and I’ve had the pleasure of working with some of those people [engineers], and they’re brilliant, absolutely wonderful people. But it requires a different level of perspective and domain expertise to really connect this technology to specific use cases.
That’s what I want to empower people with. You know, I think it’s an intimidating field. There’s lots of jargon that is used to sort of confuse and—not that researchers mean to confuse non-technical people—but it’s a lot more accessible than people think. It doesn’t require tons and tons of knowledge to unpack the basics—to understand what’s going on under the hood of the common models being used today and to know enough to know when a problem is really well suited for AI and when it’s not. And I think that’s what I’m excited to just empower people with.
And again, it’s not going to be me; it’s not going to be anyone that’s been in the field forever or is very technical. It’s going to be—and not that I’m the most technical person, I’m not saying that either—it will be non-technical, domain-rich people that think of the ideas.
So, when I’m speaking at conferences where I’m talking to different companies, I am dying and so excited to hear your ideas. And that is what fuels me—seeing these brilliant applications and sparks that come out of those conversations. That is essentially a cascading effect of all the incredible things you can do.
What would you say you’re most excited about for the next six months to a year?
Oh, I think the thing I’ve been most excited about is how good large language models are at code assistance. For anyone dabbling in programming or building different things—there’s really no excuse—you have the best assistant at your fingertips. And my dream, and something I get excited about, is putting that in the hands of underprivileged individuals who haven’t had the resources or the influence to introduce them to this world of work.
I think tech has sort of become a bit of a bubble. And I would love to see new up-and-coming diverse communities entering it and providing loads of value in different ways.
How do you see AI technologies changing the world of search?
That’s a great question. I think quite a bit. I mean, this is still wild, wild west. Search companies have yet to figure out how to even wield this technology properly.
There’s a lot of issues that have bubbled to the surface, even recently, in the last several months with generative search. But things will get better. There will be more nuanced searches than ever before.
I think the potential to follow a search journey will be more powerful than we’ve ever seen. And there are ways to optimise for that—and so it no longer becomes just search engine optimisation; it’s really AI optimisation; it’s customer journey optimisation; it’s UX. It becomes much more human-focused, which I think is a great thing.
I don’t think SEO is going away—I know lots of people say that—I don’t think SEO is going away anytime soon. I think it will evolve as it sort of already has but at a quicker pace. And so, to really understand that AI is a tool [is essential], and we can wield this in really powerful ways to help automate some of the boring tasks that we do and focus on some of the higher-level strategy and higher-level thinking.
Yeah, I think that if there’s any theme running through a lot of the episodes we’ve done on this podcast talking about SEO, it’s often—you optimise for humans, and then Google will catch up. You just deliver good value, deliver a good experience and let Google close the gap.
And AI, in so many respects, is going to complicate that but also maybe accelerate it. There’s a lot to look forward to there. There’s so many new buzzwords that have just slammed into the lexicon.
Yeah. And I mean, even AI is the new marketing buzzword. It’s marketing jargon. You know, if you really want to get down to the specifics—we still have not reached AI—and, quite frankly, intelligence is such a vague term that researchers are really grappling with how to define it.
There’s historical references to the Turing test to identify whether or not we’ve reached artificial general intelligence. But it’s really important to understand that everything we work and deal with today is narrow intelligence or ‘weak AI’. Strong AI or AGI (Artificial General Intelligence) has a generalised knowledge of the world; we’re not quite there yet. A large group of people feel we’ll never get there, and another large group of people are working hard to make it a reality.
What’s interesting with AGI is—I hate comparing it to humans because I don’t want to anthropomorphise this technology—but if you think of these large language models—it’s essentially like someone being born in a dark cave. And the only things that they know have been read through text. They have zero concept of the real world. They have no ground truth. And once you start introducing multi-modalities, multimedia, or content to these large language models—like objects, gravity, physics, audio video — these models are able to develop a way more contextually rich understanding of the world, of how things interact with each other.
We don’t think about all the common things that we do every day that are very uncommon to a computer, right? And so teaching computers the easy bits tends to be the hardest part. It’s not something we’ve read all about, and we don’t have a ton of text on it. Navigating that is a whole fascinating field.
And then within the narrow AI—which we operate—there’s really two different primary types, which is generative AI. So that’s your Midjourney, that’s ChatGPT, that’s your large language models. These are probabilistic in nature. They’re basically prediction engines—and you can build anything that has time series data. You can create that generatively with these models. So, think music or brushstrokes, art, video, audio—all of that can be generative.
And then the other side of that coin is deterministic. This classifies or categorises data, and it’s a lot more specific to the data set. There’s no probability guesswork at play there, hopefully. It tends to be a bit more accurate in nature—or, you know, that’s the goal.
We’ve spoken a bit over the past week, trying to drum up all the different ways in our business that AI could streamline things and power things or replace menial tasks. And I think one of the more interesting things that came up while we were doing that was how frequently AI was not the answer—or how infrequently AI was the answer—and how often, in fact, probably the best solution was machine learning or some other automation process that we probably already have and have had for a while. I’d be interested to hear your thoughts on—especially with things so much in flux right now—the risk of over-application of AI and how that compares to the risk of under-application of it right now.
Yeah, that’s such a good point. And I don’t think people realise how computationally intensive machine learning models or AI is. In order to even get something up and running—it’s very expensive, it requires all this compute power. It has a large carbon footprint when in most cases, different automation, Python scripts, or heuristic models get you there.
We’re seeing so many people just throw generative AI at everything. And that’s not the right use case. It’s not a proper solution. Something that I think you all helped me work through as much as I hopefully supported you with this week, was really understanding that breakdown of problem framing for AI. It is quite engineer-minded in terms of—you have to break a problem down piece by piece by piece. If you were to, let’s say, engineer a program to do X—only when you do that can you understand if a heuristic model or a Python script could solve for this. Or, is it really qualified for a machine learning model to possibly categorise [for example] photos or content? Or do some natural language processing with text? Only then can you really start to work out which are proper use cases and which aren’t. Yeah, it’s wild! How do you feel about it? It’s a lot.
It’s disorienting. Especially as a Content Marketing Director with a team of copywriters. When you consider that the first mass adoption tool was effectively a content generator, there was this, I don’t know—I wouldn’t call it stages of grief, but there was this cycle of, “It’s exciting!”, “We’ll all be unemployed in six months!”, “We can probably master this”. The swings of calamity and utopia.
As you said, I think you wind up somewhere in the middle. And right now, the more that I learn about it, the less I’m convinced it can do what our team does. So, I think the thing that stuck out to me most from hearing you speak the last few weeks was making clear that this is not a ‘research tool’—ChatGPT specifically—is not a research tool.
Not ‘information retrieval’.
Yeah, that’s the word you used! Information retrieval.
If I could have gotten one thing [clear], that would have been it. It’s so important.
Asking these tools to do your research for you and then compile it into engaging content is very risky. And, you know, we work with clients across a lot of industries, and more than a few of them are in fields where false information has real liability.
We work with clients in the legal space and wellness [industry]. Spaces where you have to do your homework. And so, it’s really important that you are combining what this can do, or defining what it can do—and then confining what it can do to spaces where it’s safe to do so. And then leaving the real research and the job of understanding the business or understanding a topic to the talented copywriter.
One hundred per cent! And what’s acceptable risk, right? For the different applications? It’s interesting to consider that in those different spaces you mentioned because they would all have different levels at which would be considered acceptable.
And then, on top of that, you have what’s known as ‘edge cases’ [a problem or situation that only happens in extreme circumstances]. They are things you really can’t quite prepare for but are bound to happen with the deployment of some of these models. And that’s where I feel very grateful that we operate and live in a space that’s not life or death. We’re not doing autonomous vehicles in our world of work, but that’s a great example space of, you know, people’s lives are the training data.
And it’s very scary to think about those edge cases. Like an Uber model was trained [a self-driving car] on people riding bikes and pedestrians walking—it had no idea what a woman was, who was walking her bike, and she was struck and killed.
That’s one of many examples of these super dangerous edge cases. And sure, they might be in the greater minority of examples or training applications for this technology. But these are really happening. And when you think of jobs, or HR departments using this to sort through resumes, and being biased—they [AI tools] are going to magnify biases that they were trained on. It becomes very problematic. It starts to automate inequality. And it’s going to do so in really sneaky, sneaky ways—just by the pure nature of common data sets and the way that they misrepresent minorities and different groups of people.
So, that’s something—as this technology continues to explode, and people just throw it at different things—that really does start to scare me. You know, that’s a scary thing to think about. That part keeps me up a bit at night.
Yeah. Just stay sceptical of this technology, stay sceptical of what you hear people doing with this technology, stay sceptical of different advancements that you hear about—the news is just full of hype right now, which is problematic for a host of reasons.
And then also to be asking questions. You know, when you hear about a specific type of AI being deployed in a space, consider what data went into it. Even medical diagnostics—what data went into that? What data went into, you know, a tax or law Chatbot? There’s lots of ways that we can keep the scepticism in the back of our heads to better navigate some things as they come forward.
Yeah, we’re racing into the application stage of mass adoption, and so there’s going to be so many different applications that are asking for permissions, that are handling data differently, and handling data with varying degrees of transparency. And there’s definitely going to be a level of digital health and wellness—digital safety—that people will need to become fluent in fairly quickly.
Yeah, beautifully said. I cannot agree more—especially that [about] reinforcement learning and human feedback. Many marketers, many SEOs don’t understand that literally how you interact with these models trains future iterations of them. You want to be really careful about what you feed them in terms of private information, medical stuff, financial information, and just kind of clean that out [don’t include it]. Because that will inevitably lead forward into future versions of these models.
Putting ourselves in the shoes of marketers and small businesses that are doing a lot of this stuff themselves, what else do they need to be aware of right now, with so much in flux?
Yeah, that’s a great question. I think just familiarising themselves with the tools currently available. Only through interacting with this technology do you start to get a really good grasp of, “Oh, it was really good at this” and “It had some like funny outputs over here”. Getting a ground layer of how this stuff works is essential.
But on top of that, it’s really the ‘doing’, the ‘experimentation’, the ‘throwing things against the wall and see if it sticks’. Playing with Midjourney [for example], why does it have such a hard time with hands? You know, getting hands-on experience.
I think for a lot of green, and even somewhat experienced marketers—I don’t think people quite realise the confidence that you can gain through doing the stuff, through experimenting, through starting to understand how difficult it is to get Midjourney—unprompted—to generate a black person, for example. It’s having that hands-on experience; it’s interacting with the tone and the voice of ChatGPT versus Bard. It’s trying to get them to do different things. It’s integrating them into Google Sheets and trying to do something at scale. It’s learning about the APIs.
I really do believe that marketers will be so, so valuable, who know how to use some of this technology and use it in the right ways. And I think, as an industry, we’ve been quite robbed when it comes to data science and statistics—just fundamental statistics, even. I find it very hard when even very experienced marketers don’t quite understand or can’t explain what a distribution is.
Data is the future. These models revolve around data sets. And we, as marketers, need to have a much stronger understanding of what that data is that’s being fed into the models. What outliers are there? Are there any inconsistencies? How can we better clean the data? How do we—you know, within data science, describe, predict, prescribe? Doing that from a marketing perspective, I think, will bring all marketers and SEOs really to the next level when it comes to what it is they do each and every day. I mean, think about all the data that you even work with at a content level. Being able to apply some transformations and a richer understanding of how different content is performing and at what topic level. I mean, it really does surface insights, unlike any other area.
So, I’m excited. I’m excited for that. And I think there’s a huge opportunity for people to learn a little bit about it. And it’s this huge value add and tool you can have in your tool set.
It was put to me some time ago—I think when I was just getting started at Pure [Pure SEO]—why there will probably always be a space for agency-side experts like us or for even client-side experts in the field of SEO. Specifically, that there’s no silver bullet for this stuff. There’s no automatic way to the top. Google can’t just say, “Here’s 10 things you do”, or “These 10 things you rank number one”, because then everyone does them, and they can’t all rank number one.
There’s a million different factors to consider. And a lot of it is objective, and it’s weighted. And I think some of the initial concern for LLMs, like ChatGPT, from a content perspective is, well, now everyone can do this stuff. But again, they can’t all rank.
So, to me, I’m seeing these AI tools from a content perspective—and probably elsewhere [other business areas]—raising the floor of what’s possible for a lot of people who maybe didn’t see a lot of success. But it’s raising the ceiling, as well. And there’s still space for you to have stuff done at an exceptional level.
So, if you were hoping to get started with this stuff, if you were hoping to engage it to empower your business in your marketing and do better than you’ve been doing before—where should you start?
Yeah, so I think sometimes, SEOs and marketers forget all of the data at their fingertips and the type of strategic insights they’re able to serve to clients. Client, SEO, or marketer relationship—your job as an SEO marketer is to deeply understand that key KPI.
What is that [KPI]? And then doing the research to understand what drives that. Write what might support that in ways that the client often doesn’t know. When you’re coming at this from a data perspective, and you’re doing your digging, you’re doing keyword research and also possibly real-world research—you’re able to surface these insights. [You can explain to the client] “These are the inconsistencies we’re seeing from this product to this product. Do you see an increased demand over here?” and they’ll possibly see that or maybe provide you with feedback. And then you’re also able to ask all of these deeper questions around—what people are asking, what does this mean, what kind of question is this.
Getting to the root of that information to really support your strategy is what drives results. And if you don’t do some of that ‘reverse engineering’, from a value perspective, what commonly happens is that clients don’t always know what they want—but they know when they’re not getting it. You could be ranking for all these incredible pages until the cows come home, but if that’s not supporting the client’s true KPI metric—whether that be revenue or signups or whatnot—you likely won’t have a future with that client.
And I think that’s kind of the getting back to the basics, getting back to human-connection marketing. That’s going to be more important than ever in the future. Because—you’re exactly right—sure, companies can take the easy way out and use some of this technology, but it won’t support them over time, and it’s certainly not going to get those results that they’re looking for. And, yeah, the whole goal is to have long-term, evergreen, ‘snowball effect’ results for clients.
So, again, it’s so interesting how it all boils back to data. But it also comes down to that human element of creativity—detective work. What are those questions? Go sit in the car repair shop of your client and start figuring out what brings people in in the first place. What kind of questions are those? How could you possibly get this business visibility before they even make that phone call? Or provide people with the support and value?
I think we’ve forgotten the facets of creativity and visibility that we can really get for clients in fun ways. And I think that’s not only a huge value add to clients, but it makes our job so much more fun.
Again—do—start by doing. Google Code Labs is a great repository of models and tutorials to build something on your local computer. Google Cloud also has a large variety of courses to teach you how to build and deploy different things on a cloud server. Kaggle is the number one data science competition website in the whole world.
And it’s fascinating to see what companies are creating competitions to solve for. It’s very telling. TSA has had some very interesting ones in the past that I like tracking. Other resources… Andrew Ng came out with a really brilliant prompt engineering course, completely free. There are lots of great resources available to start really learning and understanding this information. I’m excited for people to explore that.
I’ve personally been working on a large language model guide for the last embarrassing amount of time—several months now—but really look forward to also providing that for free to marketers and SEO specifically, in hopes that it really helps guide them to do good work, and also to support clients.
I think that’s another unspoken territory, the “How do we educate clients about this technology?”, “What sort of information or guidance can we provide?” in terms of what this is and why it won’t solve all their problems, and how to think about it.
Great. Britney Muller, thank you for joining us today.
Thank you for having me.
It’s been a pleasure. [To the listeners] Thank you for listening to Good Content, the official podcast of Pure SEO, New Zealand’s most awarded search agency. Visit pureseo.com to view our suite of digital marketing services, and get in touch with a member of our team to learn more.
Subscribe to our blog and get awesome digital marketing content sent straight to your inbox.