← Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
Voices of Search // A Search Engine Optimization (SEO) & Content Marketing Podcast
How brands can prepare for the Direct-to-Agent era of AI discovery
00:38:42
Loading summary
Transcript41 lines
- [00:00]
A
The Voices of Search Podcast is a proud member of the I Hear Everything Podcast network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit iheareverything.com welcome to the Voices of Search Podcast. A member of the I Hear Everything Podcast network, ready to expedite your company's organic growth efforts. Sit back, relax, and get ready for your daily dose of search engine optimization wisdom. Here's today's host of the Voices of Search podcast, Tyson Stockton.
- [00:44]
B
ChatGPT alone processes more than 2.5 billion prompts a day, including 330 million from the US alone. But those billions of prompts aren't just replacements to old searches. They're the start of a new discovery model where agents are making decisions for users. Optimizing for end consumers isn't enough anymore. Your brand needs to show up when AI agents are shopping. And that's just the beginning. My name's Tyson, and joining me today is Brian Stempak, co founder and CEO at Evertune. His team runs over 1.2 million AI prompts monthly for Fortune 500 clients. So he knows everything there is about how to stay visible in AI search. Today, Brian and I are going to be walking through how to influence AI agents and not just users, and why brands mastering generative engine optimization will dominate the next era of search. So with that, Brian, welcome to the podcast.
- [01:38]
A
Thanks for having me, Tyson. Glad to be here.
- [01:40]
B
I'm excited to hear a bit more about, you know, some of the findings you have because you have access to, I think, a wealth of knowledge as far as how brands are looking at prompts, what are the trends within those. So maybe just to kind of set the stage, tell us a little bit about, kind of like the data sets you're working with and how you're looking at performance in generative AI.
- [02:06]
A
Yeah, for sure. So at Evertune, we're trying to help large global brands and agencies figure out what's happening in AI search. So our starting point is understanding what are consumers doing. So I think we've all seen in our own behavior and others, people are moving quickly to use ChatGPT, they're using Gemini and Google. Sometimes they're using Meta's tools inside of Instagram or WhatsApp, they're using Perplexity. There's a whole range of AI models that they're not searching on. So step one is understanding, like, the patterns of how many people are now using those types of tools and then look at the types of things they're searching for. So we actually have a big data set of 25 million panelists. So we can see what are people searching for in real life and how is what they use AI for, how's that different from how they might use Google search? So easy example, in traditional Google search, average keyword length that somebody enters into Google is 3, 4 words. They're spending 10 seconds on Google and they're clicking somewhere else, whether it's an organic web link or a paid search ad. In AI it's different. So average prompt length is more like 20 to 25 words. So the average person is giving more context. Imagine you're shopping for a coffee maker. You're telling them what kind of coffee you like, whether you want an espresso machine, how big your kitchen is. You're giving additional detail because you know you're going to get a longer response. And so you get a longer response. Where the AI model summarizes, here's some good coffee makers or espresso makers for you. Here are a few options, here's the strengths and weaknesses of each one. And on average people are spending six or seven minutes in AI. So it's a very different experience from traditional search engines where it's less about quick navigation and getting to where you're going now it's more about give context, give detail. Because I'm going to get detail in return I can ask follow up questions. And so one general trend we're seeing is that the considerations cycle or purchase funnel is shortening. It's condensing where people now or they might have looked at from Google before they read a story on New York Times wirecutter, they go to cnet, they go to Consumer Reports, they browse around the Internet increasing, they're consolidating those searches to AI, they're doing their upper funnel research and they're comparing a couple brands and now lower funnel they can actually purchase something in the AI model. And that's new as of the past month here back in October 2025. And so we're seeing this big pattern of consumers moving quickly into AI tools and rapidly consolidating a lot of their information consumption. So it almost broaden it from just search to this is how they traditionally have used publishers and social networks and other places you might gather information. A lot of that is now moving into the AI model. And so we're seeing this happen and we're helping brands figure out how to respond by understanding what do the Models say, because interesting thing here again, different from Google search. The AI models give a different answer every single time. These are probabilistic models. They vary the answer and they'll also vary the answer based on your search history, your geography, things like that. And so if you're a major brand trying to figure out what's going on, you're Keurig trying to figure out coffee maker sales. You can't just ask it one question and see, okay, we're ranking third in terms of overall results here. We prompt on average a million times a month per brand to understand what's happening in the world of coffee makers, in the world of Keurigs, how often they being recommended versus Nespresso, giving them all the insights about what to do with that. And then how to use content pr, their own website, affiliate to go influence the AI model.
- [05:41]
B
No, you hit on an interesting piece with like the longer prompts, more context. There's going to be, as you said, like you're going to get varying responses whether it's on, you know, context in the prompt history, whatever else, timing of it, maybe. How does that change then? How we're looking at knowing what's going on in the market where it's like before, it was a very traditional keyword research, identify it. People look at competitors. Great, I have my buckets that I now have to go after. Like, how do we deal with this? Like I say, like volatility, but like fluctuations and changes that happen so much more prominently in LLMs. And then understanding how are we succeeding within that.
- [06:26]
A
Yeah, So I think the starting point is establishing the baseline. So clients we work with, I'm thinking of like an automotive client we work with. Right now they're using us to measure what's happening in 40 different countries. So right there you've got a lot of complexity of saying, what is ChatGPT or Gemini saying differently in the United States versus Canada versus France versus Singapore. Right. And so they're starting by saying like, let's break this down by geography. And then they go and say, let's look at this by. By model. So they're looking at their suv, they're looking at their sedan, they're looking at different models. And then within each model they're looking at different Personas. So they're saying, okay, let's look at the family buyer who might have young kids. Let's look at the single more adventurous buyer. They have their kind of preset Personas they use for media planning and ad buying. And they're replicating that in our platform. And so what they're doing is they're creating hundreds of different line items or variations of what response does their brand come back with. So how often, just making this up, how often is Ford vs GM vs Toyota vs Honda being recommended in each of those categories? Because the models give very different answers based on geography, Persona, the matching of the user to the right make and model of the car. And so with brands that we're working with, they're starting to develop really more of like a heat map. Where do they have areas where they're quite weak and they need. They know that they're not being recommended for family buyers? Where do they have areas where they're quite Strong? Where for SUVs, for single, adventurous types living in the city, they're doing well. That understanding is kind of where the starting point of the models are. Not because they're probabilistic, because they bear the answers. The starting point is a baseline of like, well, how well are you doing across geographies, Personas and product lines? Which again, is a very different way from how you would think about search. Search is a bit more universal than this. This is much more customized. And I think that's really important for brands to consider, which is this is more like dealing with people the same way that you might run a focus group or market research. We've used a lot of those learnings to inform how we sample the models. That's how you very quickly get to hundreds of thousands or millions of prompts because you're exploring all these sub variations, what the model might say for very specific use cases.
- [08:44]
B
That's interesting. Yeah. So it's not like what would have been maybe one keyword that you're looking at, you're looking at on the Persona level, variances from it. Of course, you have like geos from that as well within some of the different models. Are you seeing significant levels of variance between. Like if we took that last example and we were looking us within ChatGPT AI mode, I mean, however many ones we want to throw in there. But like, how much variance are you seeing kind of between these different models?
- [09:18]
A
A ton. So we drill into a few different areas. So one, a good example is looking at ChatGPT. So the underlying large language model versus ChatGPT when it adds search. Because so what happens is there's the underlying LLM that the scientists there have trained and so you can use the API of ChatGPT and ask that questions and that'll just have the kind of the purest answer of Asking the question, just the language model. Then you can ask question to the more normal user interface that you or I would experience if we use the ChatGPT app or the website, of saying, okay, what's the best SUV for family that goes skiing a lot. And when you ask that question, the core model will sometimes supplement with a search of going into Google or Bing and adding search data. And you can kind of see it doing this real time. On Gemini, for example, you'll see it running a search in the background, running 50 or 60 different queries, summarizing that information and adding it back. But even on ChatGPT versus ChatGPT plus search, there can be meaningful differences. There can be a 20 or 30 point swing of okay, the Honda CRV is being recommended 50% of the time versus being recommended 80% of the time. And so that starts to indicate is you might be weak in the core model but really good at SEO, or you might be strong in the core model, but we can. And so there's all these variations of what can happen of when you add search to the equation, does it make your standing worse or does it make it better? And so even looking at just chatgpt alone, you get different answers depending on are you asking the core model or you're asking the core model plus the search engine. If you zoom out from that a bit, the models have different opinions, right? They have a different point of view about the best sev from ChatGPT to Google to AI overviews inside of Google to Claude, from anthropic to Deep seq. And so we monitor all these different models because they swing drastically. And you might say, well, why is that? They're training on different data sets. So simple example, Google pays upwards of $5 million a month that we know of to license Reddit data. So they're licensed the entire Reddit data platform. Perplexity is not using that data. They don't have the license or the they're not paying for it. And so you have large content sets out there that are playing a role in some models and not playing in others. And then you of course have the data scientists have assigned different weightings to say here's how we value this content or this answer. And so you get very different points of view and even very different distribution. So for example, with Meta's AI model called Llama, we see the model recommend a lot longer tail of brands, so it'll recommend 20, 30, 40, 50 brands when another model like Gemini might only be recommending 5 to 10. And so that is like a more of something we've seen across product categories where it's more fundamental to how they designed that language model. And it has different preferences and different ways of saying, here are the best brands in this category.
- [12:14]
B
Interesting. I like, I like how you separated the kind of traditional versus with search because I feel like a lot of times people, they kind of struggle like, great, I know that I'm not doing well, but now is it SEO factors is the other. So I do like that separation of looking at those differences in whether it's like companies you've worked with or even talking to others. Have you seen people differentiate strategies based on performance in whatever models? So if they're like, I'm doing quite a bit worse in ChatGPT and then stringing together a more unique series of initiatives that would attack that versus Gemini or Ammo, whatever else.
- [12:58]
A
Exactly. Yeah, totally. So we see brands and I kind of think about it as when you're trying to influence the core model, it's harder, it takes more time, it takes a larger effort. The models don't retrain every single day on new data. It's only going to happen when there's either a refresh of their training set or they update the entirely new model. And so you're not going to have next day results from trying to influence the core underlying model. But when you do influence that model, it's longer lasting. Right. And so, and it has an impact of what it might go search for because it's supplementing what the core model knows. And so we kind of think of that that's like the more mid to long term approach, whereas more on the search side, that's more similar to traditional SEO where you can see faster changes in terms of, okay, the ChatGPT model initiated a search on your behalf when you're looking for the best coffee maker and it went out in search and then you can actually see the data, you can see the citations or the sources that come up in your search. When we run a million searches, we see that at scale. So we can start to say, oh wow, it's referring to this Reddit forum quite a bit, or it's referring to this coffee maker blog quite a bit, or it's going to this ratings website that's breaking down lots of prices for different coffee makers. And so we can pinpoint for the brand for some of these quick hits on citations and sources that are being searched for. These are places where your PR team should be engaging. Or if there's an affiliate option, you should do a sponsorship of that publisher or if you're doing a media buy, maybe you partner with them. And so being very data driven about the brand's content and media strategy is how we think about it. Because there's the short term benefit of getting better in search and then long term, how do you look for the sources of data that are having the biggest sway on the underlying language models? And so it's kind of the yin and the yang of those two things that will help a brand do better in overall AI search.
- [14:53]
B
Yeah. And then you're just layering into okay, hey, this is my initiative that's going to be that 6 12, 18 month kind of timeline. But then filtering in those quicker wins to keep the momentum within the organization, show some immediate impact and be able to kind of just progress towards those larger initiatives.
- [15:14]
A
That's right, yeah. Generally speaking, I think when we see brands really lean into this and they put forth an effort of adding, we kind of direct them like okay, add content about this topic area on your website, go to this and affiliate, go to this. In PR we typically see results. It's not overnight, it's more like two to three months and it'll be a material result where doesn't mean they have to do 100 PR pieces or a thousand new articles on the site. A relatively modest effort. We'll see like five to ten content pieces can have a meaningful impact if you're talking about the right things. If you're really focused about what subject areas you're trying to move the model on. If you're trying to do everything all at once, it's a lot harder. But when you pinpoint, here's the factors that matter. So like let's stay with Keurig as an example. Cost per serving might be one thing to look at or reliability of the coffee maker or the variety of drinks it makes. If you try to own one of those topic areas and really make an effort, that's where we see the models have the most movement.
- [16:11]
B
Interesting. Okay, so let's build from that. And I mean there's been a lot of conversations, a lot of talk going on recently on aging. I mean one we've been hearing and seeing stuff both in the e Commerce Area. OpenAI has had several partnerships announcements coming out, but I mean we're also seeing components and atlas just recently came out as well. How do you see in this Keurig example the interaction then with agents?
- [16:40]
A
Yeah, this is one of the big trends that's happening that is going to be is really substantial for most marketers so we'll stay with Keurig. So a couple things changed in the past month. One is the ability to actually buy something in AI. So ChatGPT launched this where they're getting more lower funnel, where you can actually go and say, hey, I don't just want to research coffee makers. Show me a couple of different products and show me where to buy them. And they're directing straight to the landing page. So you can go buy that Keurig coffee maker at Bed, Bath and Beyond or from Keurig.com or from Amazon. And ChatGPT is taking a very small percentage of the overall price tag. So they're taking kind of that affiliate commission as part of that. And that's the first step that we've seen in ChatGPT moving to commercialize in a way outside of subscriptions. So to date, all their revenue has been people paying 20 bucks a month for the consumer subscription or people paying for the professional enterprise license. And so this is the first step that they've taken where you're seeing them start to play more in the commerce side of things. But I think it's smart because it doesn't jeopardize the trust of the consumer in any way. There's no feeling that the answer is being biased or changed. It's simply presenting, hey, you wanted to buy this thing, this coffee maker. We're going to surface different prices and the different options from retailers there. And I think that's a really interesting starting point, because as commerce starts to move into ChatGPT, the other big trend that I think will start to happen over the next coming couple of years is having an agent that you trust to make decisions on your behalf. So am I going to trust an agent to go buy a car for me anytime soon? Probably not, right? I will stay involved in that decision. And even like a coffee maker, do I want to necessarily outsource that decision? Maybe, maybe not. I'm willing to take AI's advice on that. But then go down a click further, like the K cup or the pods that it's buying. Do I really care if it's like one brand of coffee or another? Do I care whether I buy it from Amazon or from Target? Not particularly. So, like, imagine a world where I'm saying, okay, AI, grocery shopping agent, you know that we're drinking Keurig at home, we go through 15 or 20 cups a week. Find me the best price for the Starbucks, the Dunkin Donuts, K cups, and go order me 200 of them. And I don't Want to think about it, that's where this gets really interesting, where the agent, whether it's from ChatGPT or another model like Anthropic and Claude Power, a lot of the agents being built on the, on the business side of things. So this could come from anywhere. But as these agents arise and it could be that's embedded in, if you shop on kroger.com or Instacart or somewhere else. But when you start outsourcing a decision to that agent, then this is a very interesting question for a marketer because you've gone from a world of saying, I'm trying to influence this consumer with all the TV ad spend I do and the digital ad spend, my paid search campaign and SEO and everything I've been doing. And now the consumer has said I don't actually care. I'm going to let the AI bot decide. And so that's a really powerful shift. Again, I don't think it's going to happen necessarily for a car or even a coffee maker overnight, but I do think for the more everyday kind of mundane decision, when I'm grocery shopping every week I shop for groceries online. Some of those decisions, I don't really care. I'm just trying to get it done. And also, you know my past purchase behavior, you kind of know what I like. You know as well as I do of the things I've bought in the past. That creates this really interesting future of what happens when people start trusting the agent to make the decision for them and how do you influence the agent? So it goes beyond like the SEO side of things of how am I showing up and is my brand being recommended and goes a click further to say, well, if this agent is built on top of Claude, I gotta make sure Claude really loves my brand and recommends my brand and knows how to find my brand and knows the truth about the ingredients or the price or the SKUs that my brand offers. And so it's hard for me to predict is that in the next six months or is that a couple years from now. But I think we're headed towards this path where, I mean, we've all been on that path anyways in E commerce, where here's recommended products for you and that whole, we're all used to that type of thing. I think it's a fairly short jump to say, hey, this is what the model is recommending for you. Just click to add it to the cart or it can just add it to the cart for you. That starts to get really interesting from the influence Standpoint of the very simple example of an AI agent.
- [20:59]
C
Time for a one minute break to hear from our sponsor, Pre Visible. So you're looking for SEO help and you got a couple of options. You could start replying to spam from agencies that claim they can get you to rank number one on Google. You can pay an hourly rate for a consultant who will inevitably nickel and dime you with hourly charges. Or you can work with a cookie cutter agency to quickly launch a strategy less project with low success rates. None of those sound very good now do they? Well, that's where Pre Visible's integrated consulting model comes in. Pre Visible draws From a collective 40 years of SEO and digital marketing experience to unlock your organic growth opportunities. They build custom solutions that combine strategy, technical expertise, content and reporting to effectively operationalize SEO for your business. Pre Visible's four stage approach ensures that your SEO programs thrive by starting off with a strategy first approach. Then they support you in your efforts to create quality content, help you identify technical issues, and most importantly, they'll work with your cross functional teams to integrate your SEO strategies to make sure that your SEO budget actually drives results, not just your agency's bottom line. So join brands like Yelp, eBay, Canva, Atlassian Square, all who rely on the SEO consultants at Pre Visible. For more information go to Pre Visible IO that's Pre Visible P R E V I S I b l e.IO.
- [22:27]
B
Which and I agree like that's a difficult timeline to predict because it's like you're dealing with like public opinion and trust factors and things like that. I do agree that that like consumer goods as, you know, paper towels, like stuff that's like you're not overly brand sensitive to begin with on that, you know, low risk, whatever, it's a commodity. Where is something that is, you know, more nuanced is going to be something probably with that heavier involvement from your last piece though of like okay, if this, I don't know, anthropic was producing like the agent then that was interacting with it From a optimization standpoint, do you think still just going towards whatever the LLM model is or are there other nuances that people should be considering with like how their content would interact with an agent rather than an end consumer?
- [23:24]
A
I think the big thing we're often advising brands to do is with consumers when you market to them, it's a different thing because you have a limited amount of their attention, you have a consumer's attention very short period of time and you need to break through the clutter and get them to remember something about your brand or have an emotional association or know something about your product. AI is kind of the opposite. An AI model has all the time in the world to learn about your brand, and it's insatiable for as much information as possible. And so one of the things we counsel brands we work with is in the past, if you're a brand New York Keurig, you're trying to focus on, like, what is the one tagline or the two lines of copy, or the three product features that we put on the website or in our advertising or in our copy that will really stand out and differentiate. And that's the right thing to do for a consumer. With AI, you have a lot more time, you have a lot more word count, you can get into a lot more detail. And so step one is making sure the AI model knows everything about your product set, that it knows the difference between the Keurig model that's meant for the workplace versus the Keurig model that's meant for home, that it knows, like, what temperature does the coffee maker get up to? If somebody wants really, really hot coffee, can you make it strong or weak? Like, all these details that might be overkill for a consumer other than somebody who's like, really crazy about Keurig. Right. But it's going to inform how an AI model answers the question. And the more data it has to pull from, the better answer it can give. So if we move from Keurig as an example, the other place we often see something like this is increasingly in B2B. So in software, where the business buyer, this is if you're buying, let's say, CRM software, and you're a small business, you're thinking, okay, do I do HubSpot or do I do Salesforce or do I do something else? Increasingly, what we're seeing from B2B buyers is they're kind of running that RFP that they might have done in the past. They're doing that with AI. They're having a research agent go out and doing a really deep search saying, okay, go evaluate CRM software platforms by cost per seed and what's going to cost me in two years and integration with email and like this long list of features that they might look at an RFP or have somebody fill out in a spreadsheet. We're seeing that happen in AI. And so if you are a software company and the AI agent coming to read up on your website, read up about your case studies and essentially fill out this rfp, it's going to have the information and so long way of saying whether you're Keurig or this or HubSpot or a B2B software company. Step one is make sure the AI models have as much information as possible about your product and your brand so they can recommend it, so they can get the right price and the right sku. And that's not always how marketers think. Right. Marketers often are thinking where do I get the most bang for the buck with the consumer or are there things that I want to hold as more proprietary and not share? And that's a pretty big change in terms of how you think about content marketing, which is for this audience it's the more data the better.
- [26:23]
B
Interesting. So there's more of that coverage play to ensure that they have that information that then can be pulled from where it's like to me I feel like a lot of the conversations that you've been hearing around AI mode in the space is more of like remind me of traditional consumer behavior, like more advertising where it's you're wanting to build that brand reputation and brand awareness for those late stage kind of clicks. But with how you're describing this too on kind of like a rise of more of like an agent driven interaction, you have that more lower funnel, high volume like coverage play to then drive into that in conversion.
- [27:05]
A
That's right. And so I think the step one is kind of give as much raw data as possible to the models to train on step two is to start pointing them in more of the right direction. One common trend we've seen recently has been in both our consumer panel data and from brands asking for this, it's thinking a lot about use cases. So again when the consumer is not saying best coffee maker but they're saying I've got a family of six and we all drink coffee and this got to be a higher volume machine like which is the best one for me being able to meet that use case. And that means creating content that maybe you wouldn't have otherwise. And so that's where we're seeing brands again and they're starting to use AI generated content as a starting point of saying let's build content for this use case, let's build content for this use case so that when the model comes across that type of user they can do more of the matching. And so I think that's again a shift in how people think about both search and how they think about content overall of where is this product to fit. And I think it's also Worth noting that like the AI models in general are discerning. They do. You've probably seen this in your own behavior. They will go through and say Keurig is great at this and not great at this. Nespresso is great at this and not great at this. They do give pros and cons. And so it's kind of like a more traditional marketing exercise of what is your strategy, what are you, which Personas, which use cases are you really trying to own and building your content around that versus trying to own everything. I think that's harder for because that's where your brand is going to put out as much content as you want. But there's also independent reviews, there's user reviews, there's people on Reddit, there's a lot of other factors that are going to shape that. And so kind of where to place those bets is a big. Is a big question as well.
- [28:49]
B
Interesting. And it, I feel like I'm racing through a few different like ideas from that. But like with that concept of like coverage from it, like one thing that we worked with in SEO for a while is like yeah, you can have over indexation. We're very concerned especially with like the really big websites around crawl efficiency, crawl budget. With this I feel like it's almost introducing like a new like do we need to challenge some of those previous thoughts on our coverage and crawling? Have you found pet companies that are maybe going a little extreme on that coverage play that they've seen kind of less significance or less weight to what they're putting out there? Is there a adverse consideration for hey yeah, but we still can't just flood it with all these different variations of potential use cases that may or may not be extremely relevant.
- [29:49]
A
We haven't seen it yet. It's certainly possible, right? I mean like I think the AI world is going to go through its generations of algorithm changes and things like that that happen. But if you. I kind of take a step back and I look at the motives and what is the motive of every AI model right now? Get as much data as possible that it can train on. They're of an insatiable need for more data. Especially like I think if I had to bet, I think in the near future we'll see the models start to differentiate on which data sources they trust and which they don't. And so I think they've been to some degree kind of using Google as their, as their crutch for that. And all the work that Google's done in the past, I think they're likely start to develop their own points of view there because they're going to have to. Because the, the downside to this right, is that we'll keep playing with Keurig. They're not a client. I should note Keurig just published a thousand articles saying here's the reasons that Keurig is the best coffee maker, the top 10 coffee makers, all those kind of listical approaches that might have some impact. But over time what any AI model, just like Google searches realize this is a lot of pretty fluffy low quality content. And it's like over time our expectation is the model is biased towards quality. We already see that today in terms of the domains they cite, the sources they cite, they're looking for the right answer to the question. When they can't find the answer, they turn to more brand owned content or they'll turn to other options. But in general we are seeing that third party publishers, content creators, social media, those play a pretty big impact in shaping the models, not just the brand dot com.
- [31:31]
B
Now what would like from and I know this will evolve and change maybe even from the recording of this, but what are some of those hotspot sources that obviously, I mean we talked about Reddit, obviously there's paid partnerships at play there. What are some of those key kind of sources that people should be really looking close at?
- [31:55]
A
Yeah, a few that come to mind. So one is one of the things we track is has an AI model done a licensing deal with a publisher. So Reddit is a good example. Travel Leisure, a publication is another where the model, some models are, they all have data licensing teams, they're paying for that information. Well, when they have deemed it enough quality to pay for it, they cite it more often, they use it more often. So that's an obvious flag of like okay, they've knighted this publication so let's go invest in more partnership and stuff there. Another common theme, we see YouTube quite a bit, a lot of the YouTube transcripts. And so creating content on YouTube is one path there. Reddit has kind of ebbed and flowed. You know, Reddit was a big deal a few months ago. It's actually declined in the number of citations that we're seeing overall, at least on kind of the search side of things. But then, you know, in other cases, like in consumer categories, we've seen a lot of review websites. So like a Wirecutter, ratings.com these different sites that will rate electronics or appliances and have their own scoring system. Most of them make money off of you clicking the affiliate link so they're independent content with the affiliate link. But those are some of the things that we see most often in a few different categories.
- [33:10]
B
Interesting. And it has been kind of interesting to see this like, resurgence of like UGC content, whether it's Reddit, I mean even in YouTube in that sense, where I feel like we went through a period or an era that it was not necessarily penalized, but it was not viewed as very highly high quality content. And now that pendulum's kind of swinging back into it being a significant lever to play.
- [33:39]
A
I think that's right because I kind of look at Reddit as a really interesting example of this where it's some of the most human content on the Internet, where, and it's, it's vetted by people. You've got the content moderators, you got people upvoting it. And so it is fairly difficult to game. And so even if you went as a brand and tried to promote your brand pretty obviously on Reddit, you're going to get called out for it.
- [34:02]
B
Right?
- [34:02]
A
You're going to get, you're going to get downvoted. And I think that makes that data valuable to the models because it actually is objective and independent and more of how people think. And that, that's, I think that's why you see some of the models paying so much money to license data from Reddit.
- [34:15]
B
Yeah, it's like you have an inherent baked in voting system and like quality piece that's just built into the infrastructure of it.
- [34:24]
A
That's right. I mean, another instinct, try not to call out. And we look at this as an opportunity for brands where when you get into more, less covered areas. So there's lots of information online about coffee makers or buying a car. If you start getting into like, okay, what's the best soft CRM software for small manufacturing companies? Right. Like very more niche use cases. The models don't always have a great answer for that because in some cases, like in B2B, the good information might sit behind a paywall at Gartner or at Forrester or some kind of white paper type report. And so in some cases what we'll see is the models are citing sources that are pretty out there. So an example would be we've seen models in B2B software where they're citing a source from a job posting. And so when we see that, it's kind of a flag of wow, the models don't have a lot of good data in this category. They had to dig deep enough to go to like indeed.com and look at a job post and pull information. So if that's their best set of data, it's a pretty thin answer. And if you're a brand in that category, that's an opportunity. So that, that says that's a flag that hey, this is a place where the models don't know very much. It's an opportunity to, to inform them.
- [35:38]
B
Interesting. I like that angle too where it's just you're looking at the strength of where those sources are because that's going to be the low hanging fruit for them. Assuming it matches kind of like your brand strategy or product.
- [35:52]
A
Yeah. Another example would be like we again to this theme of like openness or sharing more data. Many B2B companies have great white papers and case studies sitting behind a paywall. Not paywall, but like a login. You got, give me the, give me the email sign up, we'll give you some of the information. In some cases we're encouraging. Like you need to make that more open. Right. If your buyer, and in this case B2B, we see this a lot. If your buyer is going to claude or they're going to Copilot and they're asking questions to that, wouldn't you rather that that white paper inform that? And so again it's a shift in thinking of like how open are we with our information?
- [36:26]
B
Yeah, I mean and hopefully it is a changing chapter from the paywalls from a lead gen source which one's live.
- [36:38]
A
And well, I mean it's still going pretty. And I get it right. You want to generate all the CRM data and the leads coming from people filling that out. But at the same time, if a lot of your buyers are now turning to AI, then it's not a bad strategy to give AI more of that data.
- [36:51]
B
Yeah, I mean as a consumer I'm for that too. Just break them down altogether.
- [36:56]
A
Yeah, exactly.
- [36:57]
B
No one of the areas. So we hit on a few different industries international. Like obviously there's not even rollouts in languages, things like that like perspective or kind of insights from different international markets.
- [37:12]
A
Yeah. So we're global. We're operating with clients all throughout Europe, Asia, South America. And so we're operating in lots of different languages and different markets. And that's kind of the beauty of this is like these language models have all grown so quickly that they're available in most markets. One interesting trend we sometimes see is like even for a user in Australia asking questions about their local homeowners insurance companies, it'll sometimes bias towards American answers still. And so the models have adapted to give regional answers in some cases, but it's imperfect. And that's a place where we've seen, for the most part, they've regionalized. But it's kind of like Google was in the early days. It's not perfect quite yet. And so we're operating in all those markets in other languages. But I'd say that's evolving pretty quickly.
- [38:01]
B
Now with that piece though. And I mean, it would make sense because obviously these models have probably been trained, you know, on the American sources for a longer period of time. But are you seeing then in that sense the larger American global brands having like a disproportionate representation in other markets that may have just like thinner training sets?
- [38:28]
A
We are. So I think, I mean, in general, I think you see a pattern where the model's bias towards all the data they trained on. And so if you've got a company that's a large multinational brand that's been around for a long time, even if it's not the strongest in this category, the same way it might have the most brand awareness with consumers, it typically has a pretty strong presence in the AI models as well, at least to start. And as they start adding to that data or supplementing with search, that's where you can see things change a bit. But that historical legacy matters quite a bit. Another interesting anecdotal thing we've seen, again, differentiating the core model and adding search. The core models sometimes get confused by rebranding. They get confused by one company acquiring another. And so at times they'll refer to an older, out of date brand name. And so it'll often correct it with search. But when you have a brand with a complicated product strategy of like, okay, here's our premium product, our semi premium product or regular product or low end product, and they all share a similar kind of brand name, the models can get confused by that, they get confused by brand hierarchies. And so again, it's kind of oversharing of like, here's what each of these products is and defining it for the large language model.
- [39:44]
B
That makes sense. All right, so that's going to wrap up this episode of the Voice of Search podcast. Thanks again to Brian Stempak, co founder and CEO at Evertune, for joining us. If you'd like to get in contact with Brian, be sure to check out a link to his LinkedIn profile in our show notes as well as checking out his company's website at Evertune AI. If you haven't subscribed yet and you want a daily stream of SEO and content marketing knowledge in your podcast feed, hit that subscribe button in your podcast app or on YouTube and we'll show up in your feed in the following day. That's all for today. Thanks for stopping by and we'll see you in the next episode.