Podcast: Play in new window | Download
Search is evolving, and if you’re not optimizing for large language models (LLMs), your brand could be invisible in AI-powered search tools like ChatGPT, Gemini, and Perplexity. SEO expert Crystal Carter returns to the show to break down exactly how businesses can increase their visibility in AI-driven search, why entities matter more than ever, and the key strategies to stay ahead as search continues to change.
How to Make Your Brand Visible in AI Search: A Small Business Guide
Is your business invisible to ChatGPT? You’re not alone. As AI-powered search continues to reshape how people find information online, many businesses are discovering they don’t show up in tools like ChatGPT, Gemini, or Perplexity. But here’s the good news: you can change that, even if you’re a small business.
The AI Search Revolution Is Here
Look, I get it. With everything else on your plate as a business owner or marketer, optimizing for AI might seem like something you can put off. But here’s why you shouldn’t: AI search tools like ChatGPT are already getting nearly 3 billion views per month. And according to recent predictions, we could see traditional search engine use drop by 25% by 2026.
More importantly, people use AI search differently than traditional search engines. Instead of typing “automotive repair Portland Maine,” they’re asking questions like “What’s the most reliable auto repair shop near me that specializes in European cars?” If you’re not optimized for these conversational queries, you’re missing out on potential customers.
Understanding Entities: The Key to AI Visibility
Here’s something that blew my mind during my conversation with SEO expert Crystal Carter: AI tools don’t just look at your website’s keywords – they try to understand what your business is as an entity. Think of it like this: when you Google “Beyoncé,” you get a knowledge panel that tells you she’s a singer, entrepreneur, and cultural icon. That’s because Google understands Beyoncé as a distinct entity.
Your business needs the same kind of clear identity. Instead of using vague descriptions like “award-winning solutions provider,” be specific about what you do: “We’re an automotive repair shop specializing in European vehicles, serving Portland since 1985.”
Practical Steps for Small Business AI Optimization
- Start with Your Google Business Profile
This is the foundation of your digital entity. Make sure it’s complete, accurate, and detailed. Why? Because this information helps form your knowledge panel – how AI tools understand and represent your business. - Use Schema Markup Strategically
Think of schema as a way to explain your business to machines. Add schema markup to your website that matches your Google Business Profile information. This helps AI tools understand exactly what your business is and does. - Be Distinct and Clear
Look at your website content. Could someone (or something) immediately understand what makes your business unique? If not, it’s time for a rewrite. Remember: if AI tools can’t tell you apart from your competitors, neither can your customers. - You might not have a Wikipedia page (yet), but you probably have connections to entities that do. Partner with:
- Your local library
- Chamber of commerce
- Community events
- Local newspapers
- Alumni associations
When you’re featured on these established platforms, you’re building connections that help AI tools understand and trust your business.
Measuring Success in AI Search
Traditional SEO metrics won’t tell the whole story when it comes to AI visibility. Here’s what to track:
- Use GA4 to monitor traffic from AI sources
- Add “ChatGPT” or “AI Search” to your “How did you hear about us?” forms
- Regularly check how your brand appears in AI responses to relevant queries
- Monitor the accuracy of information about your business in AI tools
The Future Is Already Here
The shift to AI-powered search isn’t coming – it’s happening now. While the big players are still figuring things out, small businesses have a unique opportunity to get ahead of the curve. Start with your Google Business Profile, be clear about what makes you unique, and build meaningful local connections.
Remember, you don’t need to be a tech giant to succeed in AI search. You just need to be intentional about how you present your business online. Start with these steps today, and you’ll be better positioned for the future of search tomorrow.
How to Optimize Your Brand for AI Search – Part 1 Episode Transcript
Rich: My next guest is an SEO and digital marketing professional with over 15 years of experience. Her global business clients have included Disney, McDonald’s, and Tomy. An avid SEO communicator, she hosts SEO webinars and podcasts, and her work has been featured at Google Search Central, Brighton SEO, Moz, Lumar, Deepcrawl, SEMrush, and more.
She’s returning today to the Agents of Change podcast to talk to us about how we can optimize our businesses for LLMs. I am always excited to talk with Crystal Carter. Crystal, welcome back to the podcast.
Crystal: It’s an absolute pleasure to be here, Rich. Always a pleasure to have a chat with you and to discuss how cold it is in all of the various locales.
Rich: Yes. Crystal shared with me how cold it was today in her neck of the woods, and it was a balmy 42 degrees. As I would say, as I told her that with windchill, it was two degrees in Portland today. She doesn’t even understand what that means. She’s going to have to ask an LLM to explain it to her.
Before we jump in, Crystal, I caught a glimpse of your T-shirt. For those people who can’t see it, tell me what’s on your T-shirt and tell me the story behind it.
Crystal: So I am wearing a T-shirt for the Women in Tech SEO Festival, which is an event that’s put on by the Women in Tech SEO group, which is a group that is free to join and is for women who work in SEO and tech SEO. And people say, “Oh, I’m not a technical SEO.” It’s like, if you’re a content marketer, that’s fine, too. You’re welcome to join. “Oh, well, I’ve only just started SEO”. That’s great, join. They have a mentorship program, and you can ask questions. It’s literally the friendliest, most active benchmark for best practice Slack channel I’ve ever been in.
And the community events that they put on are really, really great coming together of their actual community. So the people that are there are people from the community and they’re really happy to see each other and to hear from folks who are speaking there. And yeah, it’s a great event and a great organization.
Rich: I love it. And I’ve got a couple of women on my team who are doing SEO. If I wanted to send them on their own journey, is there a URL specific or a website that we should be pointing people towards?
Crystal: Yeah. So the Women in Tech SEO group, I think I’m going to have to look it up now. So Women in Tech SEO, just going to Google that, give them a little bit of organic traffic there. So it’s womenintechseo.com is where they are. And yeah, shout out to the folks there. And yeah, they do a great job.
Rich: Fantastic. And we’ll have that link in the show notes.
All right. For today, Crystal, we are talking about LLM optimization. What the heck are we talking about?
Crystal: Yeah, so this is something that I started thinking about a while ago. And it’s something that, you know, basically, as soon as ChatGPT dropped, I did the same thing that everybody else did and I quote unquote “Googled myself”. Basically, I said, you know, who’s Crystal Carter? And ChatGPT was like, who? And I was like, you know, Crystal Carter, that famous lady. And they were like, I don’t know. I don’t know her. Like that Mariah Carey gif, “I don’t know who she is.” And this hurt my feelings and made me sad.
And then months and months went by. And then I said, ahem…Mr. ChatGPT excuse me, who is Crystal Carter? And they were like, oh yeah, she works for WIC studio. And she does this and that. And I was like, a-ha. This is interesting. This is very, very interesting. I’m interested in this.
And this got me thinking about some of the conversations that I’ve been having with people where they’d say, you know, we’ve checked for a brand and we can’t find our brand on ChatGPT, they won’t say who we are. And basically, when we think about being visible in LLMs, it’s essentially so that when people query LLMs, like ChatGPT, Perplexity, Claude or Copilot or Gemini, that you show up. Because search is changing significantly.
I very often use a chat assistant, an AI assistant to get information, and sometimes a website is what’s useful and sometimes a ChatGPT type tool is useful as well. But your customers and more and more users are going to be accessing these, and in order for your business to have to have a presence in these spaces, you need to optimize for those.
And some people think, oh, there’s nothing you can do. But that’s not true. It’s just not true that there’s nothing you can do. There are definitely things that you can do to move that needle in your favor. And as an SEO and a busy body, I took it upon myself to have a go at this, and I’ve seen some interesting results. And I’ve been sharing that over the last few months and people have been really engaged with it. And I think that there are wins that can be made for people of businesses of various different sizes. And I think that it just takes a little bit of demystifying of what these LLMs do and how they work, and engaging and using the tools that we have and know moving forward.
But ideally, the name of the game is to make sure that your name shows up for relevant queries. So, you know, if someone was looking for a marketing team and you want them to find Rich Brooks’s team, that would be the people that you need them to find. If somebody was looking for somebody who was an incredible SEO and had an incredible likeness for Beyonce Knowles, they’d be looking for me, obviously. And so I want people to find that stuff.
And also, when people search for your brand, that they will be able to get accurate information about your brand as well. And that’s essentially what we need to need to do when we’re thinking about optimizing for LLMs.
Rich: That’s a perfect overview, because really what I want to accomplish today is kind of understand what it means to optimize for LLMs, and then specific things that we can do as owners, as marketers, as SEOs, to make our brands more visible.
So I guess one of the places that I want to start is, as we’re talking about LLM optimization, is this the same thing as optimizing for AI overviews, which appear at the top of the Google searches, or is that something different?
Crystal: Yeah, so AI overviews and optimizing for LLMs is different. AI overviews will use some of the tools that LLMs use in order to create the content that they create, but they’re different.
So when you’re thinking about an AI overview, the journey is different. AI overviews are essentially a search feature, which are very different from somebody who’s going to ChatGPT, and they go to ChatGPT and they make a query, they make a request, they ask for something. AI overviews just pop up. Nobody asks for them. They just show up and that’s a completely different thing.
Whereas an LLM is a search tool. And also, users will choose to use an LLM, whereas AI overviews just kind of happen upon you. So I think that the user journey is very, very different. So if you’re thinking about optimizing for AI overviews, that’s a different conversation from optimizing for LLMs.
Rich: All right, so we’ll keep the focus on LLMs today. I think one question that I would have is how big a deal is this today? You know, I had Wil Reynolds on the show recently and he was talking about how small a percentage of the search is that’s going to LLMs. And I haven’t, I admit, done the research into our own GA4 reports to see how much LLM bot traffic we’re getting.
But is this substantial? Should we be, if we have limited resources, limited time, is this something that’s really worth the effort?
Crystal: Yeah. So I think that what you need to do is you need to have a look at what’s going on with your own traffic. And you can go into GA4. There’s a lady called Jes Scholz, and has shared a really good regex.
Also there was an article on Search Engine Land that also shared a really good regex for this. And basically you can filter out, with regex, the traffic that’s coming from tools like Gemini, tools like Bard, Complexity, et cetera, et cetera. You could filter out that traffic and you can see how much traffic is coming to your website using one of these regexes.
I have an article on the Wix SEO Learning Hub that talks about this and includes a link to it as well. So I think that that’s where it includes link to it in my deck. So it’s worth having a look at that.
But then I think also when you look at it more widely, Kevin Indig recently shared an article that showed that if ChatGPT continues at the pace that it’s going at now, it will surpass Google within the next sort of four to five years, because it’s continuing to grow and grow in usership.
However currently, ChatGPT is getting just around just under about 3 billion views per month. The last time I checked it, it was around 2.6 billion on similar web, but it’s getting more and more. And again, that is going to be mostly their desktop.
There’s also going to be people that have the apps directly and are using it there. So it’s getting a lot. Compared to Google, it’s still fairly small. However, the time that people are spending on ChatGPT is significant. So people are spending like six minutes per session on ChatGPT, for instance, just alone.
And they’re also not the only game in town. So I mentioned a few as well. So there’s ChatGPT. I very often use, Bing has a co-pilot app that’s pretty low touch. It’s pretty light as an app goes. And I use that really regularly just for general questions. If I’m just like, “Can I substitute oat milk for regular milk?” And they’ll be like, “Yeah, you can do that with this, this and this”. And I’m like, great, thank you. That’ll save me looking it up with my fingers. And so I think that there’s that sort thing. So there are lots and lots of other tools, and there’s lots of tools that are both expanding and developing.
I saw an article the other day that was saying that people think Bing is Google’s competitor. It’s actually Perplexity. And Perplexity is a great AI tool, it’s got a really good mix between search and LLM. So not only are we seeing growing use of these channels, but also we’re seeing a breadth of these channels. So, lots of these channels are popping up as well.
So, I will use NotebookLM and ChatGPT and Perplexity to accomplish the same task. Like, across a similar task I will use all of those things to help me get where I’m going. So I think that it’s important to remember that though Google is still strong right now, there’s a lot of traffic coming through LLMs. And Gartner is proposing that with all of this LLM traffic, we could see a drop in and search engine use by as much as 25% by 2026. And potentially organic traffic could drop by 50% by 2028. And it’s just a prediction, but sure, but it’s very interesting that people are feeling confident enough to make predictions like that.
Rich: So basically, although it may be a small piece of the pie right now, if we want to stay ahead of the curve, a step ahead of our competition, the recommendation is you need to invest some resources and time into LLM optimization.
I want to ask a question. I see the word “regex” all the time. I honestly could not define it. Could you tell me, Crystal, what exactly is “regex”, and how do we use “regex” in this world?
Crystal: So regex is essentially a set of code, but it’s basically like code parameters, basically.
So for regex for this process, you would essentially go to GA4, you can set up different workflows within GA4, and you could exclude certain things. And essentially, if you say an example of regex is if I said I wanted to – and in Google Search Console, you can also use Regex – you wanted to look for queries that had the word ‘shoe’ with an asterisk at the end. That means that you’re looking for things that are shoe and also things that are shoes. Right?
So that asterisk means, like, anything else. So it’s essentially, you’re adding parameters. Essentially you’re using a little bit of code to help you add parameters to your queries. So this regex will allow you to filter from multiple different channels like Perplexity, et cetera, et cetera, et cetera, in one string of text. So you would add that into GA4, and there’s… I can share you links to the one that, I think it’s Dan Taylor from SALT Agency, shared on Search Engine Land, and he explains as well. Jes Scholz also shared one as well.
And essentially, you take this one little piece of code, you can copy and paste it in into your GA4, and then it will allow you to filter out, and then you can see everything as an individual channel.
Rich: Awesome. So just what I’m hearing is that Regex is just a kind of complex or code-based way of asking a very specific question. It filters out the wrong answers and only gives you the results that you’re looking for when done correctly. Perfect. All right. Thank you.
Now in a lot of your conversations, you talk about this idea of entities and how they form a core component of LLM training. I feel like everything I’m asking is, can you define this for me? But, can you define what exactly is an entity and how does it fit into this LLM training?
Crystal: So years ago, Google published something or, you know, they started talking about the idea of things in strings. Things. And so, so essentially, like, when you think of a, when you think of an entity, essentially, it’s a thing.
It’s like a person, place, or thing. Like, when you’re, when you think of, like, 20 questions or something, it’s a person, place, or thing. Online, when you’re seeing entities, they essentially will tend to pop up with a knowledge panel. So for instance, if you were to Google ‘Beyonce Knowles’, she would have a knowledge panel. She is an entity, right? So is her business that she owns. So are her albums are also entities, and this podcast is an entity. It’s a thing that Google knows to be specific as to be a specific individual thing. Entities are defined.
It’s the way I also like to think of it is it’s a little bit like the scientific name for a tree or a plant or an animal or something. So, I think the scientific name for an oak tree is like Quercus. And so I might call it an oak tree. Somebody else might call it something else in another language. Somebody else might call it something else in another language. But for the science community, and so that everybody knows that we all have the same thing, Quercus is the one that we all know to be the sort of the core entity of that element. And so it essentially follows the same path. So the entity will be assigned there.
And entities are defined using the knowledge graph which is plugged into Wikipedia. And Wikipedia is plugged into Wikidata, and those things are really, really useful. And there’s lots of other data sets that help form the knowledge graph, and all of those things plug into LLMs understanding entities. So they use entities to understand the language that they’re transforming, that they’re reading.
So one of the sort of a-ha moments that I had about this was I was using a tool called TextRazor. And TextRazor is something that uses NLP, natural language processing. And they will take natural language and they will pull it apart and they’ll pull out all of the little entities from all of the different things that they can discern from the elements. TextRazor is a super fun tool. It’s an oldie, but goodie. And it’s been around for a while, but they’ve got a demo that’s really great and it’s free. So you can take a piece of text, you can pop it in, and then you can see which entities it pulls out.
What I did with this one is I put the Barbie movie description into this tool. And the entities that it pulled out were very different from what was actually on the page. So on the page, the description for it said something like Barbie and can go off to save Barbie Land and, you know, hilarity ensues, something to that effect. They never mentioned a doll. They never mentioned manufacturing. They never mentioned a toy. They never mentioned Mattel. However, all of those entities were in the NLP pullout. So when the output came out, it said that the entities were Barbie, Mattel, Fashion Doll.
So then I went to ChatGPT, and I said, “Name a fashion doll”. And ChatGPT said, “Barbie”. I went to Claude, and I said, “Name a fashion doll”. And they said, “Barbie”. I went to Copilot, and they said, “The most iconic fashion doll is Barbie” I went to Perplexity, and they said, “Barbie is the one of the most iconic and well-known fashion dolls”. And they had that. I went to Gemini, and I said, “Name a fashion doll”. And they said, “Oh, well, there’s some names for a fashion doll, like Barbie would be a name for a fashion doll, it’s Ken and Skipper and things. What would you like to name your fashion doll?” And I was like, guys, you missed it entirely, but okay.
So with that, I say all of that to say, how did Barbie become so optimized for LLMs? Like, they didn’t exist and before when Barbie came out. But if you go to the Barbie Wikipedia page, which is like the HQ for the Barbie entity online. If you go to the Barbie Wikipedia page, the first thing that it says is Barbie is a fashion doll. That’s the first thing that it says.
And when we think about this in relationship to LLMs, all LLMs are trained on Wikipedia. Which means they’re all trained on entities. All LLMs are trained on Wikipedia because it’s 66 million pages of royalty-free, human verified, cross-referenced, maintained content. So they’re all trained on there. And so this means that when you’re thinking about your entity, it’s very important to make sure that if you want to do well on an LLM, that you also need to make sure that your entity is distinct, is correct, is accurate and is optimized for the kinds of queries that you would expect to see.
I’ve seen some times where, and there’s a video of it where it was the SEO office hours, and there was a team that were from an adult toy team, and that was something that he sold, their entity on their Wikipedia page was not safe for work. It said like, adult toys, da, da, da, da, da. Their competitors said that they were manufacturers, right? Same category, really, but their competitors said, we are manufacturers, we are retailers, we are this and that. So I think that you have to think about your entity, and particularly on your Wikipedia page, to make sure that it is searchable.
So for instance, on the safe search, for instance on that particular one, but it’s also relevant for the kinds of things that you do. So for instance, on that one, I would say that your entity, if you’re going to have it, if you’re a delicate vertical for instance, that maybe you have that as part of the entity in the body of the entity, rather than as the core element of the entity.
But then I think also, I think also it’s worth thinking about the kinds of queries that you’re going to have from your brand, and the kinds of queries that people are likely to make in order to get to your brand.
This is the other reason why I published a tool that helps you to monitor your brand and monitor the kinds of questions that people are likely to ask and where you fit with those things. Also on the Wix SEO hub, which is a downloadable thing. And basically you put in your name for your brand, and it will output lots of different variations of questions about your brand. And you can use those to ask an LLM to see whether or not you even show up.
I’ve tested this on a few different things like, “what is the best SEO conference”, for instance, and to see if some of the ones that I knew were showing up there. And there were times where some of them did show up, some of them were accurate with the responses, some of them weren’t accurate with the responses. But you should make sure that those things align with what your entity is, and that those things align with being accurate about your brand.
Rich: That makes sense. So in this case, the entities would be about ourselves or brands, depending on if we are out there in public. So it might be about flyte new media or Rich Brooks, those might be entities. But also, a part of our job is to make sure that how these entities are being seen by these LLMs is also critically important. Is that correct?
Crystal: Absolutely. And I think it’s also important that you think about making sure that they’re very distinct and that they’re unique. Because that will make it easier. So I think that there’s sometimes with entities and with the knowledge graph, Google sometimes has a difficulty understanding the difference between the two entities.
So you have like, lions as king of the jungle lions. And then you have like the Detroit Lions, for instance. those are two different two different things. You have Chicago the band, you have Chicago the city, and there needs to be differentiation between the two of them.
So if I went to an LLM and said, “What is Chicago?”, they would probably give me multiple answers, right? Because there’s Chicago the band, there’s Chicago the city, there’s Chicago the musical, there’s lots of different things. So if I am one of those entities, I need to make sure that I’m distinct.
I’ve had some people where they’ve said, “Oh, my brand never shows up in an LLM”. And then I go to their website, this was an agency, and their agency said, “We are a multinational, award-winning agency supporting lots of people in lots of different countries.” I went to another one of their competitors, and they basically had the same phrasing verbatim. And I went to another one, and they basically had the same phrasing as well. So if I can’t tell from looking at your website that you are different businesses, then how is an LLM going to be able to distinctly recommend you to anyone? Because there’s no difference between any of you. So it’s important that your entity is distinct.
Another good example is if I said, “Steve McQueen is a man known for his films”. That describes two Steve McQueens. There’s Steve McQueen who used to drive fast cars back in the day, and there’s Steve McQueen who’s currently working as a director and is a Turner Prize winner. Now if I wanted to be distinct, I could say, “Steve McQueen is a Turner Prize winning Academy Award winning director known for his films.” Now that does a couple of things. That’s saying that he’s the name, giving his name, which is his entity, and then tying him to two other entities which I know to be in the knowledge graph. And that’s something that’s important. So when you’re thinking about where you’re going to be in an LLM, not only do you need to think about your own entity but think about the entities that are related to you.
So you mentioned Wil Reynolds, you mentioned other people. If you are on stage with other folks who are also in the Knowledge Graph, who are also already in LLMs, that’s going to be more likely that you’re also going to come up in conversations related to those entities.
So for instance, I went to Kenyon College – shout out to the team, the folks at Kenyon College and on the Cacosing River in the heartland of Ohio. So if somebody looked up, “who are alumni of Kenyon College?” If I have on my blog, within my entity, that I went to Kenyon College, then they will say, “Crystal Carter is an alumnus, this person is an alumnus, that person is an alumnus, that person is an alumnus.” If I have not included that in any of my bio, then I’m not going to show up. So it’s the equivalent of getting more links by essentially getting more associations with more named, distinct, already visible in LLM entities.
Rich: Okay. That is a lot to digest, but one of the big takeaways that I heard is that we really need to have a unique brand. And maybe this is a time that a lot of companies should be reinvesting in their brand to make sure that they’re just not another award-winning global supplier of X, Y, and Z. And really to make sure that they stand out from their competitors and then they do a great job of explaining that in a way that the LLMs will pick it up and understand what their entity is all about.
Crystal: Entirely. Entirely. And also invest in your Wikipedia page.
Rich: Yes, we’re going to get to that. Because I want to talk about real world SMB stuff. And yeah, we’re going to get to that.
So you identify two different types of LLMs. One is the pre-trained, and then the other is search augmented. Can you quickly define the difference or what each one is? And then why we might need to tailor our approaches for optimizing differently for those two types?
Crystal: Yes. Okay, so, and these are my designations. So you have your pre-chained model LLMs, and basically these are essentially closed. They don’t have links out to other stuff. These are getting few and far between. Since the time that I started talking about this, these have reduced. So ChatGPT basic used to not have a search feature, and they’ve just added the search feature to everything so now we have SearchGPT available for everything.
Copilot, the basic one, like the little basic assistant that I have, it’s like a little app. They really don’t do links very often. And Gemini has historically not done very many, but they’re starting to do more and more and more. But essentially when you have those, they’re essentially like an encyclopedia or like the phone book, for those of us who are old enough to remember the phone book.
So what would happen when the phone book came out, you’d get this giant Yellow Pages, and it would come out in, I don’t know, let’s say January. And you would be able to see Aardvark Taxis and AAA Home Decorator, et cetera. And they’d be there. If something happened between like in February, if your business opened in February, you wouldn’t be in the phone book.
Similarly, what happens is LLMs are trained on certain data sets. And if your activity, your business, whatever you’re doing has happened after the data cutoff, the training cutoff date, then you are very unlikely, if it is not connected to the web, if it’s a static training set, you’re very unlikely to show in that data set.
Sometimes they with all of these things are constantly testing and back and forth. So sometimes you might pop it on one search, but not on the other, that sort of thing. But if it’s a static training set, then if there’s a knowledge cutoff date, and I think in my article I point out what the dates are, but if there’s a knowledge cutoff date that say 2001 or sorry, 2021, and the thing that you’re talking about, let’s say you launched your product in 2022, well, guess what? That’s not going to be in that data set because it didn’t happen. So they don’t know what happened after that. And essentially for those ones, you have to wait until they update their data set.
So when I asked ChatGPT initially, “Who’s Crystal Carter?” And they were like, “I don’t know who you are”, between that time and the time when I asked again, they updated their data set. And after they updated their data set, then I was in the data set. So I think that that’s worth thinking about. And I think that for those ones, it’s really important to, if somebody says, “Oh, we weren’t ranking. I queried us and we didn’t show up”, ask them which tool they’re using, which version, which exact tool they were using when they did that, and double check it to see what you’re working with. So that’s really important to think about.
Don’t expect to see brand new info if you publish it last week, it’s probably not going to show up even on the newer ones unless it’s hooked up to the web. So if it’s not hooked up to the web, don’t expect to see it. And also for those ones, you want to monitor your training set updates.
So ChatGPT publishes all of the training cutoff dates and all of the information for all of their models publicly. You can find them their health documentation. Gemini also has information about theirs as well, so you can see when the training model cutoff date is. So that’s useful.
And the other one that’s my favorite is, it’s really important to give feedback. So if you go to Gemini they say, please give us feedback, we want feedback. That is exactly what we want. Give us feedback, give us feedback.
Prime example. I was working on a project called site of sites and it launched in it launched in 2024 in the spring. And at the beginning of this, of the summer, I asked Claude, “What is site of sites?” And they said, “Oh, it’s something to do with archeology.” I was like, that’s not right. I asked ChatGPT, I said, “What is site of sites?” And they said, “Oh, well, it might be an art thing. Might be metaphorical”, et cetera, et cetera. They weren’t short. I asked Gemini in August 2024, “What a site of site?”. And they said, “There’s literally a website called ‘site of sites’ that curates website designs.” And I was like, okay, all right, cool. I like this. Good, good, good. And then I asked them in September, and I said, “What is site of sites?” And they said, “It’s a website that has all these sites and they’ve got an extensive collection and they’ve got categorization and it’s inspirational.” And I was very pleased about this and I was very, very excited.
And my next question, because I’m an SEO, was “Can I get a link?” So I said to it, “What’s the URL for site of sites?” And they said and they said, “Oh, I apologize for the oversight. Site of sites is not actually its own site, it’s part of Behance.” Which made me livid, because Behance is actually a competitor to this project. So I was very, very upset. And to this, I went to the feedback form. Because at the end of it, there’s always the upvote, downvote for the little arrows that pop on there, and I said, no, bad robot, Site of Sites, the URL for Site of Sites is here, and it is not part of Behance. Submit.
And then two months later, I went and I asked the same set of questions. I said, “What is site of sites?” They said, it’s blah, blah, blah, blah, blah. I said, what is the URL? And they said, the URL for site of sites is exactly what I gave them. So don’t underestimate the value of giving feedback, particularly when it’s verifiable. If it’s your opinion, that might be trickier. But if it’s the case that they’re saying that you’re part of a website that you are not part of, if they’re saying actually Larry Bird is the coach of the Detroit Lions, that’s just wrong. That’s not correct.
So you can give that feedback and they can check it across their data set, and they will update their data set. They want to get better. So if you have verifiable and verifiably correct information that you can give them, because they’ve given you something inaccurate, then absolutely give feedback because it absolutely is valuable.
Rich: Versus Larry Bird is the greatest player of all time. And your opinion may be, no, it’s Michael Jordan, or whatever the case may be. And you may have very good reasons why you believe that, but the bottom line is that’s opinion and probably not going to have much of an impact.
Crystal: Right. Exactly, exactly. I mean, it might be fun to argue with the LLM in that regard. However, that won’t be the thing.
Now when we’re thinking about search augmented ones, which is like Bing, Office 365 has Bing Co-pilot, and that is fully connected to the web. Perplexity is fully connected to the web. Gemini is connected to Google, and ChatGPT now has SearchGPT. When we’re thinking about search augmented LLMs, when we think about those, we also have to think about rankings.
So one of the things that I did in order to optimize Site of Sites, was we were having a little bit of an indexing issue. Between the time that I started querying LLMs and the times that the time that we got through that, you know, we started seeing more progress, we were having some indexing issues. And so I worked on the indexing issues, and we got those pages indexed. And so between that time, we were doing search engine optimization. Because all of the search augmented LLMs have a search engine at the back of them. So you need to optimize for the search engine as well. And the rankings that you have on the search engine will affect how you show in an LLM.
If you go to Bing Co-pilot, for instance, the links that will show in the Co-pilot responses will be very similar to the links that show as the pre links on the SERP for the website. So for instance, I was looking at Site of Sites. Again, there’s a completely new site, and they had like three or four links that were showing there. And then when I said, “What is site of sites?”, the links that it pulled out were all of the ones that were showing on the first page query of the search engine. But it’s also important to remember that just optimizing for Google doesn’t necessarily just optimize you for.
All of these other tools. So as an exercise, I had a look at 400 keywords that were all ranking top across Google and compared them to the same set of keywords to how they were performing on Bing. And actually, you see that Bing pulls through, on the one set I had 92 top ranking Google queries. But for the same set of keywords, only 66 of them were top 10 ranking on Bing. So that means that you might see different responses and different performance on the search enabled AI tool based on those queries.
So if it is the case that you perform less good on Bing, you might also see your brand have less visibility on Co-pilot, for instance. Perplexity is backed by Brave. So Brave is the search engine that powers Perplexity, so that’s worth thinking about. And Google Gemini is of course powered by Google. But, it’s also worth thinking about that. So if you’re not ranking on Google, if you’re not indexed on Google, it’s unlikely that you’re going to be showing within Gemini, for instance.
End of Part I
Show Notes:
Provide podcast feedback at https://www.theagentsofchange.com/survey by the end of February 2025 to be entered in our $100 giveaway at your favorite restaurant or store. It will only take a couple of minutes of your time, and it will help me create a better show for you. Thanks!
Crystal Carter is an SEO and digital marketing expert, specializing in helping businesses navigate the ever-changing digital landscape, including the rise of AI-powered search and LLM optimization. Be sure to check out her article on SEO for Brand Visibility in LLMs.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine, and founder of the Agents of Change. He’s passionate about helping small businesses grow online and has put his 25+ years of experience into the book, The Lead Machine: The Small Business Guide to Digital Marketing.