Scaling SEO & Content with AI

Webinar Transcript

Mohamed Hamad: Good morning everybody. How’s everybody doing today? We’re very excited to be back here. We’ve got CT Moore of Social Inc. Today talking to us again about SEO, and scaling SEO with AI.

I’m your host Mohamed Hamad, founder of Third Wonder. And this is going to be an interesting topic because it’s top of mind, especially after Google IO yesterday and a lot of conversations about how to use SEO, or scale SEO with content, that is produced with AI without getting dinged, without getting penalized, without getting that AI slop that people don’t really want to engage with.

And you know, there’s a lot of fir, there’s a lot of conversation and discussions about that. And yeah ct, what are your thoughts on all of that?

CT Moore: Well, you know there is a lot of AI slop out there and obviously that type of content will get you penalized and none of your users will engage with it. I think what we have to understand about SEO is what a lot of Fortune 500 companies are starting to understand now, not about SEO, about using AI, especially for SEO, is that AI is great at helping people who know what they’re doing scale their work or ramp up their work. But it’s not quite at a place where it’s actually replacing people who know what they’re doing. So when it comes down to working on SEO or working on content, it’s not so much using AI to do it, but doing it in a way that’s AI assisted so your team can produce more or figure out problems quicker, but not fully replace that human layer or that human touch.

Mohamed: Yeah, I mean especially with a lot of the promises that we’ve had in terms of AI allowing you to scale things, write faster, research faster and all of that. But how do we go about, from, let’s say you’re doing research with AI and you’re taking all of this data, which is all of the data found in the Internet, and digesting it.

So you’re getting digested information from first party information and then you’re trying to use that with AI, which is regurgitating and manipulating all that content so that you can write a new piece of content. How do you inject the human experience or the expertise that you have, or the perspective?

Because that’s now becoming the more important thing when creating content, is the human experience. Seems like LLMs don’t really care for things that it already knows it wants new stuff.

CT: Right? You know, there’s an old saying, I like to speak in platitude, so you’ll have to forgive me. There’s an old saying that if you torture the data long enough, it will confess to anything. So when it comes to using AI for research and it’s pulling in data from multiple sources and also providing a layer of filtration or interpretation, you obviously, when starting out with your research, you can’t give it the full benefit of the doubt.

You as, as the content marketer, as the SEO or whatever, your, whatever your role is, you in this, in this torturing process, you have to be the grand inquisitor. So you know, you got to know what type of confession you are trying to get out of AI.

And I know that’s not a perfect example, but think of it this way. Think of AI as really how it was originally marketed in the early days. Like something closer to an assistant or a virtual assistant or in this case an intern. You got to bring in this intern. They’re going to do all this grunt work a lot faster than you can. They’re going to help you pull all this stuff together. But you got to train this intern, then you got to onboard this intern and finally you got to manage this intern. So you don’t want to just go in with a, you know, a random version of chat, GPT 5.5 or something like that and, and just trust it.

Before you start doing any research in any field, you want to train it for that field. Give you an example. Let’s say you’re an E-commerce company. You want to grab all your product information that you have, you know, any of the frequently asked questions that maybe customer support is getting from, from customers and you want to feed that to the AI as a training base.

And once you know it has some context and you have this custom agent, and it understands your product and it understands your industry, now it’s time to ask it to go out and do research because it’s going to be a little bit more focused. And then of course once it has that training, you really got to onboard it.

You know, you’ve given it some training, now get it used to working. The first few times you go through the motions with one of these agents, the work is going to be, you know, if it’s trained, it won’t be slop, it’ll be closer to slop than a polished product. And what you want to do is manage, this agent, have training interactions with it, giving it feedback and be patient for the first week you’re using it.

It’s going to still be quicker than you know, using an actual intern, but it’s not going to be perfect. But as you give it feedback, tell it what you liked and like, verify which sources are more trustworthy, kind of communicate your goals of what you’re trying to do.

With this research, it will start getting better at pulling in research that’s more ready made and ready to be implemented as part of a content strategy. And honestly it’s the same time if you’re trying to get it to create content, it’s a question of providing it with some training materials, onboarding it through the process and then working with it to actually work out content that isn’t slop.

And that’s gonna resonate, your brand’s voice and resonate with your target market.

Mohamed: Absolutely. And a bit of a shameless plug here. We at Third Wonder have a program called the Jumpstart where we basically help organizations do that. How to create those training materials, how to develop the guardrails and the policies around what, what your agents will do.

And that all depends on the different types of LLMs. So for instance, you’ve got ChatGPT and they have projects and custom GPTs. You have Claude with projects and skills and you have Gemini with custom gems and whatnot and let’s say Notebook LM and all of that.

But the one thing that you mentioned was giving it feedback and giving it feedback. Nowadays a lot of these LLMs have memory. So as you keep developing your voice and tone and give it that feedback, it’ll start to remember things and it will hone down that voice because it’ll start to pick up on the nuances of what you, you actually like and want to sound like.

But now what are the shortcomings, what are the things to watch out for when you’re working with these? I mean it’s all great to talk about training it. You know, a lot of people, even though they’ve trained the LLMs, sometimes hit a roadblock. And I’ve done that a lot. Like there are things that just, it just won’t budge.

CT: Yeah, well I think a lot of it comes down to a feedback loop. Like to keep running with the intern analogy, you never fully trust an intern to be fully autonomous. You’ll constantly remind them of certain SOPs or best practices throughout the days or weeks. I’ve found that is very necessary, with any type of custom, agent.

It’s one thing, you know, you give it all your product information, you give it your, your brand guidelines. Every time you kind of start down on a new project, you’ve got to remind it to refer back to those things. It’ll, it seems like it’s still in its memory but becomes a more distant memory.

So you kind of have to jog that memory. I find there’s a lot of redundancy in interactions with AI to get good quality content. You know, every few prompts referring back to whatever training documents you gave it. So there’s no ambiguity there.

Just like you would, you know, like a summer intern that doesn’t have a lot of work experience but is very bright. They’re not going to know the industry inside out the way you do or the way your client does. So you have to remind them, you know, for this exercise, don’t forget to refer back to this document or this knowledge base or these brand guidelines and things like that.

So you almost have to treat it, like a brilliant or ingenious toddler that’s able to do everything but still has the attention span of a child.

Mohamed: Yeah. On our side we like to use review agents. We create an agent whose sole purpose is to just review things based on specific guidelines and structures. So even if you do create content with a creation agent, for instance, you have a separate one that’ll come in and will review the content and double check everything based on guidelines so you’re not mixing the two hats between them.

And that kind of goes into what you would do in real life. You’d have a copywriter or a content creator. But you’d also have a director and an editor to kind of come in and oversee and say, hey, nope, slash this, this is not in our voice. This doesn’t make any sense. And double check things.

CT: Yeah, and you know, when we’re training an agent, we run into two problems sometimes. Like, you know, a client doesn’t have brand guidelines, they don’t have a lot of product information offhand, you know, when you don’t have any of that, it’s also a good reminder of like, oh, we got to do this foundational work. How are we going to train this agent if we don’t already have it?

At the review stage at least, you know, for the Types of content we’re doing or using AI to assist us and create. We wouldn’t create a separate agent for that as much as we will feed it back all our feedback and all the clients feedback, write down paragraph by paragraph. So you know, we don’t just copy and paste in a revised draft and say here’s everything will change, we’ll go through the sections where the changes happen, we’ll feed it in, we’ll ask it to you know, highlight what was actually changed there. And that kind of adds to the memory you were mentioning earlier that allows it to perform better on an ongoing basis.

So I guess in addition to the training, onboarding and, and management aspect of it, there’s also a maintenance aspect where you want to keep reminding the agent or the skill set of what the goal here is or what the background context is.

Mohamed: Yeah. Now talking about like what, you know, you’ve gone off, you’ve created it, you’ve edited, you’ve pulled through all this content. How does that affect SEO in general? Like there’s a lot of conversation about AEO geo and we just found out recently that GEO really doesn’t matter.

Google Search console released this guide and pretty much put the kibosh on this concept of geo.

CT: Yeah, you know last time we did one of these webinars, I think it was, in October I said something that people have been saying for the last 20 years in SEO. Every time there’s a change in the ecosystem or in the algorithm, SEO is dead. Long live SEO.

So when all these new buzzwords like AIEO and GEO started coming out, I laughed. I, I, you know, excuse my language, I was calling bullshit on that. They were great acronyms for people in the industry to virtuue signal that they’re, they’re up to speed with the latest changes. But at the end of the day the best practices are all the same. Producing good quality content that speaks to your users needs, not spamming the system, you know, having proper technical optimization on your site whether it’s structured data or page speeds.

And if you do all of that right, you know, not just the search algorithm but the LLMs will continue to, to reward you. And you know it’s, it’s a little different optimizing for the AI overview and search than it is optimizing to be recommended by LLMs.

But the best corp, the core best practices are still exactly the same. Maybe what we need is a new acronym DEO discovery engine optimization, in case anyone’s looking to coin that one.

Sorry, go ahead.

Mohamed: For those who are not familiar with a lot of SEO best practices, maybe we can kind of run through a couple of those. Yeah, right. It’s very obvious. Writing for humans versus machines, making sure your content has a unique perspective or human perspective.

You’re not regurgitating stuff that is already very openly accessible on the web that everybody said. But what else is there?

CT: Well, if you think about what search engines are first set out to do, they want to give users the most relevant information possible. Right. And what it deems is relevant is the popularity of the information, but also the expertise, the authority and the trustworthiness behind that information.

So essentially what you’re going to do is you’re going to think, you know, really put on Your, your, your Mr. Dress Up Hat and say, okay, if I’m one of my customers or a member of my Personas, what problems am I looking to solve that my products and services could solve and then start generating content around that.

For some people it’ll be FAQs. For some people it will be how to guides. For some people it’ll be product comparisons. For most brands it’ll be a mix of several types of content. But first, understanding what type of content do my users actually need, are actually looking for, actually going to benefit from, benefit from.

Then once you kind of have that figured out, it’s time to flush out that content. So yeah, there’s going to be a lot of things. You don’t want to be repetitive and you know, repeat everything that’s been in a top seven listicle all over the place. That being said, some of that stuff is still, you know, foundational. So you’re going to have to touch upon it, but touch upon it in a unique brand voice kind of way.

You know, you don’t just scrape it off of a competitor’s site and just change a couple words so it’s not duplicate. But really think, you know, if my brand was a person and it was really speaking from these brand guidelines point of view, how would it phrase that question? How would it answer that question? What tone would it have and who would it be talking to?

So that’s what happens on the content side for, you know, maintaining all the authority and trust or establishing some authority and trust. You need to, you know, rank well in the search engines and rank well in LLM recommendations. But then the other side of it is technical and this is old school technical SEO best practices.

You’re going on pages that are structured properly. They’re using HTML in a clean way, they’ve got header tags. So content isn’t just all jumbled together, it’s well structured. You’re using schema markups, structured data markups. So when you know an LLM or when Google’s bot crawls, the page knows.

Okay, this section is a product description, this section is an FAQ section. And this is a way of making your content more modular using technical best practices. A big thing LLM and you know, AIEO looks for is the modularity of their content.

So it’s not about throwing up a big guide with some product recommendations and some FAQs that that piece of content I just described has three components. It has an article component, it has an FAQ component as a product description component. You want to get that technical, structured data markup in there.

And finally I think the big thing, and you know, this can’t be under, this can’t be overstated enough. And it’s something that’s been really, really important ever since Google introduced mobile, first indexing years and years ago is page speeds. How fast is your website?

You don’t need the fastest website, in your market or in your industry, but you do need to be on par or a little bit faster than your competitors. Obviously there’s challenges there. You got tracking scripts and other things going on, slowing down the site. But you have to look everywhere where you can increase page speeds.

Because again, when search engines and LLMs want to recommend the most relevant and trustworthy source of information, they want that information to be accessible. So if you have page speed load problems, you’re going to lose out there. So that’s another big thing you got to be looking for.

Mohamed: Yeah. And to say a little bit about the technical side for those who are not familiar with what structured data means, it’s basically machine code on a website that defines what a piece of content is. So for instance, if it’s an faq, it’ll mark it up with a tag that says that this is a question and this is its related answer.

Or if it’s a team member’s profile, that is a person with a title and a name and a bio and so forth.

CT: It’s literally a way to categorize blocks of content.

Mohamed: Exactly. And for page speed, it’s an interesting one because the web is becoming more and more multimodal. It’s a Lot of images, a lot of video, is being added which can slow down a website.

Because if you’re loading a website on a mobile device while you’re on the go and your, your network isn’t as fast as at home on let’s say a fiber connection on WI fi, that needs to be fast and responsive.

So someone who’s on the go is not waiting for a minute or 30 seconds for a page to load. And that’s the important part that Google optimizes or prioritizes. Right, right.

CT: So you know, you’re a lot more familiar with the technical implementation, implementation side of things than I am. But there are ways to prioritize what loads on the page first. So really think about it, there’s what’s going to load first that’s going to benefit users from an LLM or search algorithm point of view, but also what’s going to load next that’s going to benefit your business from a conversion point of view.

You know, so getting a little bit of content to load right before all your product results load. But you know, don’t wait for all the content to load before your product’s results load. Another thing to do with speed and this is something you and I were talking about earlier this morning, but I want to bring up for whoever’s sitting in is the problem with the page, page speed loads and bot traffic which has come up.

So you know a lot of brands are using tools like cloud flare to filter traffic, spam traffic from different sources. There’s a lot of bot traffic out there designed to, you know, slow down websites, scrape content, that type of thing. And then there’s also features that products like Cloudflare have to I guess throttle LLM bots so that they don’t scrape your content.

So a couple things I just want to say there are caveats is be careful how you configure those tools. If you’re throttling all your bot traffic too extensively Googlebot will have trouble getting through. We’ll start getting 404s, you’ll see your site start getting de-indexed.

Similarly blocking all LLM crawl bots poses a problem because if they can’t crawl your content, they can’t recommend your content.

Mohamed: Yes.

CT: So you know that’s, that’s a kind of a sit down moment you have to have with your leadership, your marketing team and your dev team and say okay what tools are we using to reduce these, these bot attacks or this bot traffic that might be inflating our hosting costs?

And what are our goals we’re trying to achieve with our content and try to find that happy medium? Because I’ve had two clients just kind of throttle bot traffic to the point where they completely disappear from the Google rankings within a week and undo, you know, two years, three years of solid SEO work that’s not always easily restored.

Mohamed: Yeah. Just want to say, if anybody’s got any questions, please throw it into the chat. We will be going into a Q and A session in a few minutes. And we’d love to answer, what you guys, what’s top of mind for you?

So we talked about content, scaling it, training, training the AIs and the LLMs, how to review it, how to give feedback, how to structure your website to be fast, and to make sure that the content is machine readable.

I think there’s a storytelling aspect that we don’t really talk about as much when it comes to SEO and content creation and using these LLMs to really be a little bit more interesting with the type of content that we create.

And a lot of the times we focus on rankings and the AIs and you know, telling a very, very, specific narrative that is, is optimized but it doesn’t come off as human, it doesn’t come off as connective.

And you know, what is it that we would be looking at to really reshift the way we think about content creation? To be more human and engaging and just, you know, cut through the jargon, and get.

CT: So if you’re trying to create content that’s gonna, that’s gonna feel human and you’re trying to do it in an AI assisted way, keep in mind that you’re doing it AI assisted. You’re not doing it with AI. It’s easy to take a brief or an outline of an article that’s fully researched and has all its talking points and its sources, hand it to an agent and say, write this blog post.

It’ll be readable, but it’ll be AI slop. Really what you’re doing is you’re looking to speed up your content team, so that they can deliver content more quickly, maybe create more, more, more content in a week or reduce the amount of time it takes to create the content they’re trying to replace.

But you know, use AI to help research your things, develop a Brief, but when it comes to writing, let your content people interact conversationally with, with AI. You know, maybe you could treat the AI agent as a junior copywriter.

And your content person sits in a role as a senior editor, you know, where they’re maintaining brand guidelines and human, a human element to it. So I’ll give you an example. Never try to, you know, prompt it to write a whole article in one go. It’s not going to work out like it’ll be fine.

Information people will be able to understand it but, but it won’t have that human element, that especially Google’s now looking in its official guidelines to make you stand out from the crowd, but rather go through, whatever piece of content you’re doing section by section, you know, asking the agent how would you write this section?

And now something that’s not quite perfect or not quite on target is handed to someone who actually knows how to write and they’re able to look at this and say, oh yeah, this is what I don’t want. Now I know what I do want to write. And as you do that, give it back your human created version so it can start learning from you.

And as you do that, subsequent sections will come out a little bit more closer to what you want. But you still need that constant human interaction and back and forth, you know, a human taking an output, a human who knows what they’re doing, giving it that really human touch or rewriting it completely, feeding that back into the machine and seeing what it can do with the following section.

But like I said, you’re using it, you’re doing AI assisted content, not AI driven content. And I think that’s the big difference. At the end of the day, it’s similar to, when a university professor, assigns 200 students an essay topic.

After reading those 200 essay topics, that university professor is going to be in a position to write the book and it won’t sound like an undergrad wrote it. I don’t know if that’s the best metaphor, but I think you guys probably know what I’m getting at there.

Mohamed: All right, we’ve got a question here, from Elif. If you are a one person marketing team for a small nonprofit, what’s the first tool you personally would reach for?

CT: Oh, wow. I guess for me that would depend on what is my priority, Is my priority technical SEO, is my priority market research or is my priority content creation? You know, on the content creation side, we found that a Properly trained agent in chat GPT or clothed tends to outperform the other models we’ve tested.

And same thing for the research side. We have a little bit more, we’re a little bit partial to chat GPT for, for market research because of the way it does deep dive research. But that’s what I do. I’ll let you take the other half of the question, Mo, because I know you guys use AI for a lot of other things.

Mohamed: I would say, if you’re trying to keep track of a lot of information, for instance, nonprofits will have a lot of reports, a lot of white papers. Something like NotebookLM is great to be able to bring in a lot of data and be able to digest it and tease out the information out of things.

It’s a tool from Google, it’s free. You bring in all of the sources that you have from YouTube videos to websites, PDFs and Google Docs or Word Docs and it’ll help you dive into that. And then from there you can tease out the information to create interesting content from it.

Whether you’re using ChatGPT or Claude. Claude is great for writing. It does have more of a human approach and it feels more connective. But ChatGPT does a pretty good job there as well.

Right? We have another question here from Sam.

Can AI pull information about a company brand in all languages or will it be strictly in the language it was asked in English, Arabic, French, etc. Now this is an interesting question.

CT: That is an interesting question.

Mohamed: I don’t really have the answer to that, but I’m going to make an assumption that it would probably be primarily in the language that you asked it in. So if you ask it in English, you’ll find English sources.

If you asked it in French, it’ll find the French sources. Unless you specify that you want all of the different languages in the research, or you give it multi language resources. So this one’s a tricky one.

I’m not too sure, but I’d love to dive in and investigate that one.

CT: The closest I’ve, I’ve gotten to experience that. You know, because we live in Quebec where we operate in both languages, we were doing some research and I got my hands on government documentation to train an agent before diving into the research. We had DOC official documentation in both languages.

So I trained on that. And during that interaction I was able to get it to answer in either language or translate from one to the other without any errors. But again we trained it on identical sources from multiple languages so that made a big difference I think.

Mohamed: Yeah, it really depends. If you ask it primarily to focus on different languages, it will do that, but you have to be specific.

CT: Right.

Mohamed: Another question here from Mike. Many CMS products are adding gen AI functionality. Should I advise my users to use these or should they use standalone LLMs outside of the CMS?

CT: That is tricky. I don’t have any experience with AI functionality native to a cms. You know, a lot of our client work spans clients that are using all kinds of CMSs. Some of them are custom built.

We’ve basically relied on LLMs that are outside the CMS. I had played with some content tools that are inside, larger tools like take semrush for example and it has some AI assisted content tools. I found that they were, they were pretty decent but they were limited for generating that really human content.

But they were good for, you know, getting you a very solid skeleton. Maybe it has all the information but only 50% of the voice and it still requires some editing. But I think the only way to answer that question is maybe ask around for people who have more experience with using a CMS native version of AI.

Mohamed: My experience with LLMs that are inside specific products are opinionated to the product builder. So if you were using let’s say Squarespace, and there’s an AI built to create content within Squarespace.

It’ll be trained on a wide variety of websites and it’ll be very generic and it’ll be very hard to kind of fine tune it to your voice and your tone and what you’re trying to get at. Same thing with let’s say in CRMs, like let’s say HubSpot has a lot of what you call it, AI tools or let’s say you said SEMrush.

SEMrush for instance will take a lot of the data that it has, but it’ll digest the data and it’ll give you what it thinks the users want, but it won’t have that storytelling or that human aspect. And this is from my experience, it’s probably going to get better and better as people get feedback and train things.

But at the moment I haven’t really had a great time with them.

CT: And actually come to think of it, Mike, because we really are strong proponents of training a custom agent for you know, your task. I’m not sure if one embedded inside a CMS would allow you to do that. If it does, great. But if it doesn’t be cognizant that that’s going to be a limitation because like I said, AI as an assistant or an intern is still something you got to train and really onboard.

So if you don’t have that ability within the cms, you’ll probably get better results you know, with, with something like cloud or chatgpt. If it does offer that ability, I guess play with it and you know, compare to, compare it to what the other tools do and see what works.

But yeah, generally I’m aligned with MO on this one.

Mohamed: Okay, anybody else have any questions? We are at time. This was a very informative session. Thanks everybody for joining us today. There’s a lot going on in this space.

Obviously AI is going to be a continuing topic that everybody’s super interested in. If you are interested, we do a workshop called the Jumpstart, a third wonderful, where we help organizations train their LLMs, create those training materials and those feedback loops and a roadmap to implementing AI in the organization.

That’s a great resource that we provide. But also Chris, you work with organizations to help them on the SEO side when it comes to the AI training and on content as well.

CT: Yeah, yeah. So we, right now we’re, we’re heavily deep into helping our clients, in house content teams really make this part of the process and do it in a way that you know, makes less work rather than more work. So that’s something we’re, we’re very good at and if anyone’s interested in finding out more than more about that, I think you’re going to send an email out after this.

There should be a link in there for our free SEO clinic and we can kind of help diagnose your needs if you want to fill out that form and drop us a line.

Mohamed: Absolutely. We’re going to send a recap and a lot of this information in your follow up email. And thanks for everybody for joining today. Thank you Chris for all your insights and looking forward to the next time.

CT: Thanks for having me.

Mohamed: All right, thanks everybody and see you next time. Cheers.