BaseForge is GEOforge's proprietary knowledge base engine, built to ground AI-generated content in facts a large language model cannot find anywhere else on the public internet. In this episode of The GEO Show, Paris Childress and Simon Young walk through ingesting 1,335 sales-call and demo recordings from client Stitch into BaseForge in a single day, then show how repeatedly re-expanding Google's people-also-ask box surfaces the content gaps that knowledge base should fill next.
Generic AI content is about to get a lot easier to spot. Anthropic just announced a watermark similar to Gemini's SynthID, and the rest of the large language models will likely follow. Once every AI-assisted page carries a visible fingerprint, publishing faster stops being an advantage: it just produces more content that looks like everyone else's, built from the same public sources. Paris Childress and Simon Young argue the only way out is a proprietary knowledge base, a structured record of what a company knows that the public internet does not.
To make the case concrete, they open the knowledge base inside BaseForge, GEOforge's ingestion tool, built for client Stitch. Simon uploaded 1,335 files in a single day, mostly sales call recordings and WhatsApp demo calls, and the two walk through what that volume of raw material unlocks.
It mostly isn't, according to Paris: reading best-practice articles and tool pitches in the GEO space, he sees real emphasis on a knowledge base less than 10% of the time, even though it is what keeps generated content high information gain, accurate, and free of hallucinations. Simon's own experience started small, recording his thoughts after using ChatGPT and Gemini heavily, before applying the same discipline at client scale: pulling in all of Stitch's sales calls, hundreds of them, which surfaced knowledge the company had never structured. Most businesses are sitting on knowledge nobody has written down, and a knowledge base is the container for capturing it before it disappears.
Most of what Stitch had wasn't clean, structured data, it was Zoom recordings, WhatsApp demo calls, and timestamped customer conversations that would take a person weeks to review manually. Simon's standing recommendation to clients, independent of GEO, is call-tracking software like WhatConverts, so every call gets recorded and can later be mined for sentiment and objections. With BaseForge, that manual review step disappears: the AI runs through the raw calls directly and reports what's working and what isn't.
Information gain is Paris's term for content that offers something genuinely new: an expert observation, original data, a tested methodology, a contrarian conclusion. He frames it as an evolution of the SEO acronym E-A-T, because it depends on firsthand experience rather than research. His argument for why prompting alone can never manufacture it: an LLM can only index what a person is thinking once that thinking has come out of their mouth, been recorded, and been published somewhere it can be crawled.
Simon illustrates the point with a client meeting about an HR software company's Google Ads messaging, where the group worried about sounding as generic as every other HR vendor talking about "time management" and "employee management." One offhand line, the suggested headline "Sarah's late for work again," was the kind of specific, human idea that gets lost the moment a meeting ends unless someone captures it. Paris also points to wearable AI recorders, one priced around $179, as an early sign that always-on capture of stray thoughts is heading toward becoming routine.
The same logic that captures internal knowledge also works on public data about competitors. In a previous episode, Paris and Simon scraped roughly 500 G2 reviews from a Stitch competitor called Wati and found the reviews were a strong proxy for the voice of the customer. Stitch's 1,335 files do the same thing at far greater scale: with an estimated several thousand unique questions across those calls, Paris calls it potentially "the largest Q&A index that could possibly exist" for the business.
"ChatGPT can become your best salesperson in that respect, when people are prompting with knowledge of your brand, looking for questions about you or to compare you to somebody else."
Publishing that Q&A as FAQ content does two jobs. It helps a generative engine recommend the brand to someone who has never heard of it, and it means that when someone asks how Stitch's Pipedrive integration compares to Wati's, the model has an accurate, brand-supplied answer instead of a guess.
Simon's process starts with a single question typed into Google, for example, "is it possible to integrate WhatsApp with my CRM system." Below the AI Overview sits the "people also ask" box. Clicking the top question open and closed repeatedly, five or ten times, keeps expanding the list underneath with more related questions pulled from Google's own index. Hand that list to someone who knows the product, identify which questions they can answer better than what's ranking, group the strongest into a themed set of around five, and build one piece of content, ideally video, that answers all of them directly.
Paris connects this to video's outsized role in AI Overviews. In one live example, roughly 25% of the visible space around a search result's AI Overview was video citations, and the pair pointed to a channel run by Lakshit Ukani, with 17,000 subscribers, where a single video published three months earlier had already reached 164,000 views, most in the first 20 to 30 days.
The shift from SEO to GEO does not remove the need for high-quality content, it raises the bar for what "high quality" means. Content generation is still central, but only holds up if grounded in something proprietary the business actually owns. Paris and Simon plan to turn Stitch's knowledge base into an actual content strategy next.
Okay, we are live. Hey Simon, how are you doing today? Good, looking forward to this one, should be an interesting topic for us. How's your week going so far? What have you been working on? Spoken to a few new potential clients about GEO, and there's a lot of interest. Obviously there's loads of people exploring it. I'm seeing it because you and I are deep in the game, right? We're talking about this stuff all the time on LinkedIn and other platforms, and I get every single advert you've ever seen, all the different software. I must get pitched about 20 times a day. I don't know whether you see the same.
Yeah, I'm overwhelmed with GEO and AEO content, all of my social feeds. I guess every algorithm out there knows that I'm obsessed with this topic, so it just keeps reinforcing that. It's only going to get worse. Which actually makes me think very hard about how, in these types of settings with these LinkedIn Lives, we can stand out from all the noise and say something different. Look, first off, we're live, and not an awful lot of people are doing that. We're running a project together in the wild, and we're going to show people exactly what's happening, and then show them, more importantly, the results.
One of the things I said when I met you first off was: we need to be able to prove what we're saying, because the whole industry is awash with tons and tons of people saying what they can or might do. I referenced a post in the last live about somebody putting up a set of results that just looked completely made up, and it turned out they were. People have got to be very careful, and it is the wild west again. It's back to the good old days of there being loads of people peddling software. I saw a piece of software peddled today that was $49 and purports to do everything you'll ever need to do in GEO and AEO. It supposedly gets you listed everywhere. I'm like, for $49? Okay, good luck.
That's pretty suspicious. Yeah, but that's the game it's going to be for a while, until we get to the point where, well, we'll hold our hands up and say that not everybody understands absolutely everything within the space just yet. There is a level of experimentation, but there's a level of knowledge that people need to have before they get to even investing in it. One thing, there's a big, very popular product called Otterly.AI. It's one of the big names in the space, and I think their starter plan is about $50 or $60 a month. The thing that jumps out whenever I see a price point like that, below $100, for robust AI visibility ranking, is that the first thing I'm sure of is they're only doing one run per prompt every time they check. So it's a very randomized, highly variable experiment that they're running.
I don't think that tells you anything really, because if you run the same prompt, I think we said this last time, if you run the same prompt ten times in a row, you will likely get ten different answers. And somebody else, same thing. Sorry to interrupt, I was speaking to a client this week about, he's got a recruitment business and he wants to be listed in several different towns, and that's the exact conversation we had. He's been experimenting himself, searching in ChatGPT, looking for his business, finding it in one search, then not finding it, getting somebody else from one of his other offices to try and see if they're listed. That's not the way to go about it. You need a tool that's going to be robust and reference where you actually are, because I saw another post today, somebody say, "Oh, I've taken, I've got three times the share of voice I had two days ago." Or you came up another time, or you flipped a coin and it came up, I don't know. Just run it again and maybe things will change.
I'm just going to pull up any comments. For the audience joining us live, please drop any comments, because we are watching them and we'll try to reply in real time if we can, to prove we're live. That's right, to prove that this is pre-recorded. It's really weird, Simon, because I've got two screens. I've got the stream right in front of me here with you, and on my other screen I'm watching the event, which appears to be on about a ten-second lag, so I can see the comments. It's kind of weird, because I glance over at the other screen and I see myself talking from ten seconds ago, which is a little weird. Weirdly, when I do these Lives, and maybe it's just me, I tend to talk to myself, which is maybe odd, but it would be nice to think that as this series becomes more regular, because we're planning on doing a couple a week, as we get more into the more in-depth, deeper topics, we get a bigger audience as we go along, hopefully.
Yeah, I hope so. All right, let's dive into the topic of the day, because there's a lot to talk about here. We're going to be talking about a knowledge base, and this is one concept that I often do not see associated with the AEO and GEO conversation. I'd like to hear your thoughts on this, Simon: nine times out of ten when I read best practices articles, and when I see content in the GEO space proposing tools and solutions, probably less than 10% of the time do I see a real emphasis on the concept of a knowledge base, meaning a knowledge base that is going to ground all the content and keep it high information gain, keep it accurate, eliminate hallucinations. What's your opinion on that? Have you seen this as part of the broader conversation much?
No, not as much as you would hope. My own personal experience of building a knowledge base was, if I go back to when I was using ChatGPT a lot, and Gemini particularly, I just started recording my thoughts. I encourage my clients to do that, and everybody really should be doing that. One of the exercises we've done for Stitch, the client we're going to run this project on, we've actually pulled in all of their sales calls, hundreds of them, and that's pulled out tons of knowledge. But if I take it back to where I started, I just started building things that are in my personality, my own brand guidelines, the way I talk about things. Then you can overlay it with just things you've got, a little nugget of information you pick up when you're out and about, and you should record all of that. I don't think we necessarily do that, like just having a notepad, so to speak.
That's the fascinating piece, because you've spoken to me a lot about the new knowledge, the knowledge that's not out there in AI or on the internet, and that's what matters. When you speak to clients, you have to say to them, what do you know that somebody else doesn't, because that's where the gap is. Yeah, that's the concept of information gain. I think that's an evolution of the SEO acronym E-A-T, essentially, because it's your firsthand experience. You could even take it a step further and say the LLM can never actually index what I'm thinking unless what I'm thinking comes out of my mouth and gets recorded, and then gets published on the internet where it can grab it. If I have real original thoughts and ideas that come out in meetings or discussions, and I have the opportunity to record that, then that's really something new to the conversation, hopefully.
There's something I'm going to try to find and pull up really quick. It's a new device, a wristband, or you can attach it to your Apple Watch, and it's a recording device. You just tap it and it starts recording. If you're in a live meeting, or even in an interesting discussion, you can just turn it on, and of course there are privacy considerations, and as soon as you stop recording it sends it to your LLM of choice, and it just gives it that context. I think that's where we're headed here potentially, because probably the majority of meetings in business are still not recorded, and I think that's going to change over the next couple of years. I think it's going to start to become commonplace.
We should all be recording every single meeting we're in, right? Just for the fact that you can translate them, the AI recording, into notes and action points and tasks and follow-up and everything else that comes out of it. Most people don't bother doing that as much as they could, but I don't know that they know the benefits of then pushing that into a knowledge base. As an example, I was with a client this morning talking about their Google Ads campaign and the messaging within it. We're worried that they're going to be too generic and fall into the trap everybody's falling into, because it's an industry where it's so hard to get away from what everybody else is saying. The example is, they're an HR software company, so it's people talking about time management, employee management, and all the ads on Google are exactly the same.
We talked around the subject, and one of the headlines that stuck out from that whole discussion, an hour-long meeting where you'd very easily lose one of those pieces of knowledge dropped into the meeting, someone said, "Well, could we use the headline, Sarah's late for work again?" Rather than, you know, HR software, this just stood out as something completely off the wall, and it's fantastic. It should work really well, we've got to test it of course, but I've sat in meetings with some huge businesses, 15 or 20 people around the boardroom table, people taking notes, people spewing information out to other people, and it's crossing all over the place. We can use AI to make sense of that and then hone it. Someone was talking to me about using that set of notes as a resource to then pick pieces from it, to then use AI as I believe it should be used properly, to give you other ideas on top of what you've already thought.
Yeah, I think I understand that concept. I did find that link, it's called Memo Gem. It's a wristband, or you can attach it to your Apple Watch if you have one, or get the lanyard and loop it around your neck. I guess you could actually just record your whole day and then send it to your LLM of choice and have it summarize or do whatever. It's $179 for this. I thought about getting it, but I haven't pulled the trigger yet, though I am kind of tempted. Well, maybe you have to tag them into this post and they'll have to send you one. Oh, that's a nice idea actually. By the way, this episode is not sponsored, this is just a genuine suggestion we're pulling up, but I could send this over to them and see what happens.
Look, this whole episode's about the concept that they want to help us use that knowledge, right, or hopefully not for the other purposes that some other people listening may be thinking it could be used for. But yeah, I'd love to have one of these to be able to just talk while I'm working away on my screen. I've got six screens randomly, I've grown into six screens, and I've got Claude running on this and three or four others, and I've got projects here and there. I have got one of the higher Claude subscriptions now, and I end up leaving it to do a task here and here and here. I'm sure that's how we're all going to work, but to have something you could talk to as well, and have it upload on the fly to your own knowledge base, would be amazing.
It looks like my new Fitbit. A couple of weeks ago when I was in the US I got this Fitbit, and it has this little sensor that pops in, so I couldn't have two wristbands, that would look pretty lame, but I would probably get the lanyard for around the neck. Here's the Apple Watch, you can stick it onto the back if you have one of those. I think these are going to start to become popular. People are going to start just recording their days, everything, and it may not only be listening at some point, it's also going to be watching as you walk around.
Well, this goes back to a conversation I was having years and years ago, I wrote a book about toasters. Basically the fact that every single device in your house is listening to you now, right? We did a load of experiments in the early days when Alexa was kicking off, around saying certain things to generate adverts. We know the biggest knowledge base on earth, for example, sits behind Google or Meta. I wonder why, well, the question is why did Meta spend $19 billion on WhatsApp? Because it is listening to you, or it is recording the messages you're sending backwards and forwards. We're probably right that we're going to move into a world where every piece of knowledge is collected, and the AI knows more about you than you probably do.
This thing is 0.4 ounces. As you were describing that, I thought about the movie The Truman Show, have you seen that, with Jim Carrey? I was going to mention Jim Carrey too, where we're all going to end up in that. Yeah, everything, our entire lives are being recorded, not for a live TV audience, but for the audience of AI. It's crazy to think about. I guess what it comes down to is whether people will trade off privacy for the benefits this brings, because it's clear the privacy is a cost. If something's recording me, there are times I might be at home and forget it's on, and get in a fight with my wife or something, this could put me in some very uncomfortable positions if I'm not careful. I can imagine the costs of this, the privacy costs, very clearly. For a lot of people it's not as easy to imagine all the benefits, so I think the immediate reaction would be no, because they can't do that cost-benefit analysis completely if they don't know how good LLMs can be, if they have basically everything you've said throughout the day, and they can tell you how your day went, or how much of your time you wasted, how much time you spent scrolling.
That's all going to change now presumably, with the big court case going off in the UK against Meta and so on. But that said, and it's true whatever anybody says, we're all carrying a mobile phone around, and that is harvesting data. It absolutely is, and it's listening. I can't tell you how many times I've seen a Facebook ad in my feed based on something I had recently spoken out loud and not done any type of internet research on, not had any chat about anywhere, I'm sure of it, just something I've said. So I know the phone is listening, and it's mining that conversation for ad opportunities. It doesn't surprise me that much, I think Facebook has always been pushing the envelope in this way really hard.
We actually had that data, we tested it many times, if there are people watching this, if you've got a cat or a dog, or maybe you haven't got a pet, start talking about cat food or pet products, or if you've got a cat, start talking about dog-related products, and you will see the ad 100%. Within minutes as well, that's the scary thing.
Well, with that, Simon, let's transition into GEOforge's knowledge base, which is called BaseForge. I think you set the record, I'm pretty sure you set the single-day record for most BaseForge knowledge documents uploaded in 24 hours, working with Stitch. Apparently Stitch has done a great job recording sales calls and customer calls. You were able to dump 1,335 files into the knowledge base, which is incredible. I want to take a look at these files. When I look at the titles, I can see Zoom calls, meeting recordings, a lot of WhatsApp Stitch demos in the title, and then some of them are just timestamped.
Yeah, there's a ton of conversations with customers in there, but luckily they've been recording for a fair amount of time. A lot of these guys aren't using the system I tend to recommend, a piece of software called WhatConverts. I've been using it for years, recommending it to hundreds and hundreds of customers, just because it'll record all the calls. I wasn't recommending it for a knowledge-base reason, I was recommending it so people could go through and understand where the trigger points were, what customers were discussing, how they felt, sentiment basically. But you don't have to do that manually anymore, we can just get the AI to run through it and tell us what's working and what's not working. That's probably the fundamental thing we're talking about here: getting to the point where we understand far more about our customer base, or ourselves, than we ever probably could before.
There's another angle here, which is beyond training, just providing this broad pool of semantic understanding about a brand. It's also about setting the record straight and being able to capture the voice of the customer. When we were in our last episode we looked at a competitor, what were they called, Wati. Wati had 500 G2 reviews, and we saw that if we scraped all those reviews we were capturing the voice of the customer really well. We're doing the same thing here, but even better, because in 1,300 files there have to be several thousand unique questions that have been asked about their product, or about this industry, by customers or by prospects. This is incredible, because as long as the answers to those questions in those calls were correct, and I assume most of them were, what we have is the largest Q&A index that could possibly exist.
If we feed this as Q&A, or FAQ posts, publishing FAQ posts reflecting these questions and answers, we're effectively just training the AI on how to answer correctly. So beyond getting recommended and chosen and cited by AI when a customer or prospect doesn't know your brand, what about when they do know your brand, and they say, what's the difference between Wati and Stitch, how is Stitch better when it comes to CRM integration with Pipedrive? If somebody asked that question at any point and it surfaces through this knowledge base, then ChatGPT can give the perfect, accurate answer. So ChatGPT can become your best salesperson, when people are prompting with knowledge of your brand, looking for questions about you or to compare you to somebody else.
Where do we go from here, now we've got this massive knowledge base? Next, obviously, we've got to take this and find where the gaps in the knowledge already out there are. The old-school way of doing that: let's ask a question in the browser, any question you like, and I'll show you the old-school way I used to do this piece of work. Let's ask something relevant to Stitch, such as, is it possible to integrate WhatsApp with my CRM system? I haven't practiced this one, we'll see whether it brings anything up. We get some ads, which we would, and if we scroll down we get to the AI overview, then a few videos, and then the "people also ask" section.
I used to go to the drop-down arrow next to the top question, and if you click it, and keep clicking it, it opens and closes, opens and closes, five or ten times. You can see below it building a whole raft of questions, more and more coming out of Google, all or most of which are supposed to be relevant to what you asked in the first place. Each of these will have structured schema, or Q&A tags, within the sites they live in. Some of them may be video, most of them are probably listicles or Q&A. I used to pull all of this, give the list to a customer, and say, which of these are relevant, which ones do you think you can answer better, and then build that into what would become a blog or a video. That was a piece of knowledge.
So now we're talking about a scenario where, with what we've described with Stitch, we know ten times, a hundred times more than we ever would have done. We used the AI within BaseForge to go and examine all of this, and other parts of GEOforge, to mash it together and give us this new knowledge base. That's fascinating actually, I'm just absorbing the "people also ask" trick, because I've been aware of it for a long time but never really been clicking it the way you showed me. If you just keep clicking repeatedly on the first arrow, open, close, open, close, it's going to go down the rabbit hole into that topic, and you end up with all of the questions in theory around the topic. It's a sort of early AI, right?
Yeah, it is. Because if I know you're a master of video marketing and YouTube video, if I wanted to create a script that would rank my video number one for integrating WhatsApp with my CRM, I would probably need to give all these questions to AI and say, give me a script where I essentially address all these questions. Would that probably win it? Oh yeah. Let's drop down a few of these and see if there are any videos in them. So it is starting to reference some videos there, and what's also interesting is they're now dropping the AI overview into the "people also ask" section itself.
So talking about making a video to get it to rank: if you gave a customer a hundred questions that might be asked around their topic, and said, pick me the top 20, but let's try and keep them in a theme, so you might have a group of five, another group of five, and you're building a story around a topic, and then get the person to make a video answering the questions. Within the video you can timestamp the answers, and within the answer you've got the Q&A schema, you're just following Google's rules. I know we're talking about Google here, not necessarily any of the other LLMs, but the logic still follows: if you create a piece of content correctly, and you're answering the questions, you should, in theory, get cited, because you built the knowledge base.
This is a new discovery for me as well, let me check this video really quick. It's basically a talking-head video with some animation in the back. When was it published, three months ago? This guy actually has a pretty decent channel, 17,000 subscribers. This video, published three months ago, has 164,000 views, and here's the arc: most of these views happened in the first 30 days, or even the first 20 days. It may well have been manipulated with ads, which again works. This really underscores the importance of video when it comes to AI overviews and GEO in general, but AI overviews mostly, because as we've seen here, if we collapse the sponsored results, get rid of the ads, the AI overview starts with this thumbnail, with these videos here that I can scroll. If I view all here on the right, these citations are video, video, video. I'd say 25% of this real estate is video.
And then, I can scroll down a bit more, there's a Reddit thread, and now here's a video section with video thumbnails, there's our guy Lakshit Ukani who we just saw, and here's another video we saw earlier. If I view all I'll see all the video results themselves. And then there's "people also ask" again, with AI overviews that contain video thumbnails. Google said, I think back in 2017 or 2018, somebody will correct me if I'm wrong, that they were going to make the entirety of Google into video by 2020. It never transpired, but I don't see why it wouldn't happen eventually.
The difference is, obviously, with traditional search you cannot rank at the top of normal Google inside a month or two, whatever you do, unless you've got a hugely authoritative website. If you start a new YouTube channel today and make a video about any topic and do the right things, you can get it to the top of YouTube inside 20 minutes. Whether it stays there or not is a whole other ball game, because it then depends on whether it's consumed, watched, and useful, and I believe more so whether it's useful, and that's talking to your knowledge base. What we're doing with AI now is asking, is that content useful, and can it therefore be referenced, and do people want to digest it? If it's interesting and true and referenced across other sites, it will rank, which is, believe it or not, what you're supposed to do with traditional SEO. We're in a world where this will be incredibly important.
Yeah, great stuff, Simon. By the way, our 30 minutes is up, I feel like we could go on a lot more, but to recap: what we really wanted to emphasize in today's discussion is that if you want to make this transition from SEO to GEO, which so many people are trying to make today, and do it the right way, content generation, high-quality content generation, is still a cornerstone of the strategy, but you cannot just turn out the content. It's got to be grounded in something proprietary that you own. And if you don't have anything proprietary that you own, you better start talking and recording some of your strongest thoughts about your business, your niche.
That's where we are now. On the next episode, Simon, which is going to be on the same time, the next thing I want to do is start looking at what kind of content topics we can create out of this enormous knowledge base, and then how we think about content strategy and start to ship content and analyze the quality of that content. So that's what we have to look forward to next time. Cool, no, I've enjoyed it, and it goes too quick. Yeah, it sure does. All right everybody, well hopefully you'll all join us again next time, same time, we're going to be airing Tuesdays and Fridays at 2pm UK, 4pm Bulgaria time. So please tune in again on Friday, and we're going to look at some content recommendations from GEOforge, and start talking about content strategy for Stitch, which is now grounded in this robust knowledge base. All right, thanks Simon for another great discussion, see you again on Friday. See you soon.
