AI slop is content a model could produce unassisted from public internet research alone, with no proprietary knowledge, sourced evidence or original perspective added. On this episode of The GEO Show, Paris Childress and Simon Young work through where the AI slop line actually sits: how content length changes how much AI use is acceptable, why AI-generated images now carry a bigger penalty than AI-generated text, and why GEOforge blocks any content that is not grounded in a brand's own knowledge base. Their working rule is that the more proprietary knowledge feeds into AI, the less the output reads as slop.
AI slop is not defined by whether AI touched the keyboard. It is defined by whether the finished piece contains anything a model could not already produce on its own from public sources. On this LinkedIn Live conversation, Paris Childress and Simon Young work through where that line actually sits, because "just don't use AI" is not a workable rule for either of them, and the conversation keeps landing on the same test instead: does the content carry proprietary knowledge, a sourced call recording, a specific client case, something the model could not invent from internet research alone.
If it does, using AI to help write it does not make it slop. If it does not, no amount of editing, rewording or chaining the draft through three different models fixes that.
Paris defines it as any content that is purely derivative: research pulled from the internet, synthesized well, but with no unique perspective added. His test is direct: if a model could have produced the same piece on its own by researching the topic, the piece has no reason to exist, because the model will eventually just answer the question itself and skip citing anyone.
The more proprietary knowledge that you can bring to AI, the less that whatever your output from that is is going to be AI slop.
Simon's version of the same test comes from calling out a PR professional on LinkedIn who publicly warns against AI slop while posting content with the tell-tale spacing, paragraph breaks and em dashes of an unedited AI draft. Polish is not the giveaway. The absence of anything a human could only know from doing the work is.
Paris ties the answer to length. A short LinkedIn post is short enough that writing it yourself is barely more effort than editing an AI draft, so platforms treat heavy AI use on short-form content as the most egregious case. A 2,000 to 3,000 word research article is different: there is simply more writing to get onto the page, and leaning on AI to produce the words is acceptable as long as the underlying research, opinions and perspective are genuinely the author's and the draft still gets edited and proofread.
Simon pushes on the harder case: chaining a draft through ChatGPT, then Gemini, then Claude to merge the outputs into something "unique." Neither he nor Paris has a clean rule for how much reshuffling counts as sufficiently different, and both treat that as an open question rather than pretend otherwise.
Simon argues images are currently penalized more heavily than text, and the visible symptom is sameness: brand social posts that all look like they came from the same generator, with the same fonts and the same polish. His fix, built with Claude in about ten minutes, is a template tool that treats brand elements as SVG puzzle pieces, a logo, a pattern, a chart style, that get assembled into a new composite image with no AI watermark or signature, because the final PNG is a unique combination rather than a single model output. He demonstrated it live against GEOforge's own blog image generator and Stitch's blog page, where several existing images were visibly AI-generated and inconsistent with each other.
Here the two disagree. Paris's working assumption is that language model crawlers are built to read words, not images, so image quality or AI-generation is mostly irrelevant to how a page gets cited. Simon counters with a working YouTube ranking signal: putting readable words directly into a video thumbnail, matched to the words spoken in the first thirty seconds of the video, measurably helps ranking, which means at least one major platform's algorithm is reading text baked into an image. Neither claims certainty about ChatGPT specifically, and both agree it is cheap enough to test rather than assume.
Paris says no, based on roughly a dozen studies he has read comparing sites with and without one: none show a measurable difference in AI citation or share of voice. His recommendation is still to add it anyway, on the same logic as any zero-cost, zero-downside insurance policy: it doesn't hurt, it might help at the margin, and it takes little effort, so skipping it because it is "probably useless" is the wrong instinct even when the skepticism is correct.
Nobody on this call believes AI slop is binary. It is a gradient, and the only lever either of them found for moving down it consistently was feeding AI something it did not already know.
Paris: Okay, we are live. Simon, how are you doing, sir? Good to see you again on this hot Friday afternoon.
Simon: Well, it's not hot here. I don't know what it's like where you are. It is quite chilly here.
Paris: Well, it's 36 degrees in Sofia, Bulgaria at the moment.
Simon: Wow, okay. Yeah, we're sort of a mild 20-ish.
Paris: Oh, that's nice. Yeah, this has got to be the hottest day of the year. I think most of Southern Europe is experiencing the same stuff.
Simon: But we're going to have El Niño next time, so we're all going to get flooded this winter apparently, but we'll see. Work-wise, let's get into it.
Paris: Really cool. I'll just set the stage for today, because we have a tantalizing topic. I finally drop the marketing term of the year, probably, which is called AI slop. Our goal today is to discuss how to avoid getting the AI slop label. We're now in a world where the watermarks are coming, or they're already here.
Paris: Google has had their watermark for a long time and nobody even paid attention to that, but Anthropic's watermark is coming and people are now really worried that they're going to get penalized for AI slop, and how do you still do things the right way and generate content knowing that you need AI to help, because I'd be at a big disadvantage if I didn't use AI at all. But I don't want to get the AI slop label. So that's what we're going to talk about today. Simon, do you have any thoughts on that at a high level?
Simon: Well, obviously it's the buzzword of the month, or the quarter, I don't know, it moves so quickly now in the AI space. But I sat in a meeting last week with a board of people and talked about AI slop, and they all started laughing because they'd all been chatting about it. So everybody's talking slop. The problem fundamentally is people are lazy. Most marketing departments, or nearly all marketing departments, are probably generating a great deal of their content using any of the engines.
Simon: And it's so easy to get lazy. The trade-off is, I can make tons of content, okay, but is that all getting ignored, is the question I think we face.
Paris: Yeah, I think that AI slop, in my definition, is anything that is derivative content, derived just from internet research.
Paris: The reason I say that is because if you produce a piece of content and then you read through it and say, well, ChatGPT or Claude could have produced something just as good or better if I had just given it the topic and said go out and research this topic and create a piece of content on it, it will go out and pull in sources and synthesize the information in a brilliant way, but effectively there's no unique information gain, there's no really unique perspective, it's all been pulled from the internet, so it's derivative.
Paris: To me, that's the real litmus test for AI slop: if it's derivative, based on internet research that you or your AI tool of choice has done, that means AI can do it on its own, it will do it on its own, and it will provide that answer on its own and won't cite anybody, because it's already pre-trained on all that knowledge.
Paris: Our goal is to give the AI new pre-training data, information gain, so that the content is no longer derivative, but actually contains net new knowledge for the AI. In my opinion, even if that content is generated or assisted with AI, if it is high information gain and it's grounded in proprietary brand knowledge sources, then I think it's safe. I think it's totally safe, and not only safe, but I think it's going to succeed.
Simon: My question around that, and I don't necessarily know the answer, is probably something a lot of people want to know. If I do the research in ChatGPT and it comes back with a great piece of information, the benefit of using AI is obviously it can expand you into other areas of a topic you may well not have thought about. How far does somebody have to go to add information into that, or rewrite it, to make it unique enough that it isn't just slop? A lot of people will just copy and paste. I called somebody out on LinkedIn about it.
Simon: He's from a PR agency and talks a lot about not writing AI slop, and then one of his posts is blatantly written by AI. There's no spacing in it, the paragraphs aren't placed correctly, you can see the em dashes, and so on. To the trained eye you can tell that's AI, and for somebody who's preaching don't use it. Maybe it was sarcastic, like, I'm going to post it and see whether people call me out on it.
Simon: But the question is, how far or how much do you think you have to edit that content, or reorder it, or play with it, to make it sufficiently different?
Paris: Yeah, that's a very interesting point. I don't have a great answer for you, Simon. I think it also matters how long the content is. A social media post is a very short form. You have blog posts and other longer form stuff. I think it gets most egregious for the platforms with short form content.
Paris: If you have a simple idea you want to communicate in a LinkedIn post, LinkedIn would probably prefer you just write that in your own words, since it's relatively short anyway, instead of having it written in a very polished way by AI and then copy-pasting it.
Paris: If it's longer form content, let's say a 2,000 or 3,000 word research article, I think it's much more acceptable to lean on AI a whole lot more, because there's a lot of writing that has to get done and the words need to get on the page.
Paris: As long as you still bring your unique research, your unique thoughts and perspectives and opinions, and that's still reflected in the content, but you used AI to help get the words on the page and speed things up, and you're still going to edit and proofread it, to me that is totally above board. I really don't see why anyone should find fault, or why a platform should penalize, a long form piece of content that goes through that process. I do think length matters a little bit.
Simon: There is even an AI slop button on LinkedIn now, so you can supposedly report posts. But whether people are using that much, I don't know. I think that was just a marketing gimmick from LinkedIn to get people excited about the fact that we can push back against it. I've even seen people talking about deliberately putting spelling mistakes into their work because then it can't be AI. I'm like, wow, are you really going to do that?
Paris: Or maybe you're going to intentionally degrade your content.
Simon: Or maybe you go to ChatGPT, get it to write the initial draft, then take that draft over to Gemini and get it to rewrite it, then get that into Claude and try to get it to say, can you combine these three pieces into a unique piece? How far do you want to take this before you know whether it is or isn't getting flagged? I think with imagery it might be more important.
Simon: I was talking to Claude about this this morning for the Stitch projects we're running, and Claude actually said the more egregious penalty at the moment is against images rather than written content.
Simon: I think they are very worried about a lot of the social platforms coming to look very similar. I'm seeing loads of posts that you can just tell are AI generated. You can see the people who are using it. Brands can be very similar. People don't necessarily use different fonts, or say to Claude, can you please give me something that stands out differently to standard content?
Simon: For the Stitch project I was trying to build some imagery, and I think we'll probably go on to talk about this piece that I've done.
Simon: But when you're presenting your brand, you've got your brand guidelines, your fonts, your colors, your tone of voice, everything going into presenting, for example, a social media post or a blog header or a YouTube thumbnail, and you want it to be reasonably consistent. But there is a danger, and we do see it already, that a lot of people will use image generators to produce a load of images. They all look very similar. Yes, they're on brand, but it's almost too polished.
Simon: I then told Claude, give me some images for my blog posts, some of which we're going to publish through GEOforge. I want it to be a consistent theme, and I don't want an AI fingerprint or watermark in it. It came back and said, if we skin it differently and produce you a tool, then Claude went on to build me a tool that creates SVG images as templates. I can select multiple pieces, then tell it the words I want in it, and download it as a PNG.
Simon: It said that won't have any signature or watermark or pixels that flag it as AI. It will read as user generated content.
Paris: So what you're saying is that Claude is telling you that even if all of the puzzle pieces individually are AI generated, if you put the pieces of that puzzle together in a way that you constructed uniquely, then the entire image becomes unique, non-AI-generated, non-watermarked, non-tagged. That's interesting.
Simon: Yeah, think about it as a puzzle. I asked Claude, I want to be able to have five different structures. Can I share my screen here and show you? That would be easier.
Paris: Let's do that. Right at the bottom, the icon in the middle.
Simon: Share screen, and let's share this bottom one. Let me know when you see that. This is GEOforge, I've just actually published an article and I used this feature I'd created within Claude to create this image.
Paris: Just a second, I don't think your screen is up yet, Simon. Let me check the settings here. Oh, I'm sorry, here we go, I needed to add it to the stage, I just did that now. Okay, now it's up.
Simon: So we can see inside GEOforge. Here's one of the articles that's been written. The image I wasn't particularly happy with that the system generated, not because it wasn't a correct image, but because I want them to all be branded, on theme, and feel consistent when you get onto the customer's blog page. This is the blog page for Stitch, and at the moment you can probably tell that a bunch of these images are AI generated. I wanted to get away from that.
Simon: So I told Claude, well, Claude actually suggested this to me, I'm not going to claim a hundred percent credit for it because Claude told me what to do, and it said let's build you a tool. Within it, it's got the logo, and you see this stitch pattern down the side.
Paris: Is that something that's branded for Stitch?
Simon: Well, that's a template piece, think of it as a puzzle piece. This was all built in Claude in about ten minutes. It's given me different styles of post, and all of these elements are just pulling in from SVG files. So you see this graph style, or...
Paris: Okay, got it. It's creating some sort of a mashup among all of these different images?
Simon: Yeah, it's making the mashup. So you choose which version you want, and you put on your title. If I go back here and get the title of this blog, UK GDPR and data rules, and I put that in my tool here, you can see now it's added that into the image. Claude is now telling me if I was to download this image, it's a unique PNG. It's not generated by AI because this is a templated generator that includes the puzzle pieces to make my image.
Simon: But as you can see, if I said I want it in a different theme, a different color, a different size, a square, then I can, and this was all built in Claude in ten minutes. So I've got one with bars on it, let's say that fits better, I want it darker, and let's go away from the square, let's say this is the blog image now.
Simon: I then said to it, well, what would happen if I imported my blog post? So I've put my blog post in here, because I care that my content is rewritten, so it's here. Obviously GEOforge has written the version which we're happy with anyway, but I said can you get it to write in the voice of my client, and give me different lengths of content.
Simon: Then I can say to it, write a post, and it's written me a LinkedIn post to go with this image, to link back to the unique piece of content that GEOforge is building.
Simon: The question then is, does that constitute non-AI slop? I don't know, but I've got to be a lot closer than us just copy-pasting straight out of ChatGPT.
Paris: I still think it's not AI slop, Simon, because ultimately you have derived a LinkedIn post from a blog post, and I don't think that's a problematic derivative step, because before that, the blog post was derived from original knowledge base source material. We saw that last time and we'll look at it again, that your draft in GEOforge is referencing specific sources. If you open up on the right, let's see, the sources should be labeled here, but I think it's because you've moved it out of draft mode.
Simon: Do you want me to go and look closely now?
Paris: When it's in a draft state, if we go to another one that's still in writing in progress, any of those, now you can see that. That's a good example, you can see the sources on the right. All the sources are labeled so you can see every specific passage that was cited from the knowledge base. You can trace each source. In fact, the citation markers are even in the draft itself. That goes away, by the way, once it's published, but it's there for the initial editor.
Paris: It says source one and source four. You can go to the legend on the right and see source one is a WhatsApp chat, source four is a recording of a sales call, and I can dig into those sources and see that those are the sources supporting the claims made in that first overview paragraph.
Simon: This is where I struggled a little bit, where I said generate image and then it doesn't...
Paris: We're not really strong here yet, I have to admit.
Simon: No, this is... you know, if I was to, how central lettings move from individual WhatsApp accounts to branch-wide inboxes, all I was talking about doing, I can then go into my little tool that I've built, and there you go, I've got an image.
Paris: Well, it'll just pop it into your... is your plan, Simon, to start having these kind of consistent branded hero images for all the posts in this framing?
Simon: Yeah, so in theory, if I then look at the blog page for Stitch, which I think is this one, this looks generic, right? I'm not saying mine are going to be world-class design, but it will be a lot better than what we've got. And again, in theory, the question really begs itself: a blog is copy-pasted, completely AI slop, and the images are AI slop. Do the crawlers just not pick it up? Does it hit the front of it and go?
Paris: I really don't think the crawlers pay much attention at all to the images. I don't think, because these are language models, they're there to gobble up the words, but not the images. The images are for humans.
Simon: You say that, but one of the things I do on YouTube a lot to rank videos is put the words into the thumbnail, and a hundred percent, YouTube reads the thumbnail and any words in the image, definitely. And we should test that in GEOforge, because putting the words into the... as long as they're readable in the thumbnail, it picks it up. One of the ranking techniques in YouTube is words in the thumbnail plus words spoken in the first thirty seconds. If they match, that's a good signal. So it could be something we could test.
Paris: That is interesting, and I think that's a very good point, the words in the image. I don't really know the answer for AI crawlers, but let's say Googlebot or OpenAI's crawler hits a page and sees the image, I guess there's alt text, there's going to be alt text in the image that it can read, or...
Simon: No, which shows up? That's not put in. So when I code an image for a thumbnail on YouTube, I will embed all of the schema and everything, tell it the creation, who created it, the author, a description, tags, all sorts of stuff in it. But the basic thing is, as long as those words are in that thumbnail, it works.
Simon: And we all know we could take a PDF now, a flattened document, and give it to AI, and it could read it all, transcribe it, and put it into a blog. So we know to an extent they're reading those words.
Paris: I am very confident that Google's crawlers can read images and process them. I'm not so sure about ChatGPT as much, but it's an interesting thing to test, and why not do it anyway, because it also makes for a good-looking image. I don't think it hurts the user experience, and if there's a chance it helps the SEO and GEO, then I think it should be done. It's kind of like the LLMS.txt file.
Simon: But most people have...
Paris: Not that... this is another thing, it's kind of the equivalent of the robots.txt file, but it's LLMS.txt, and some people think you should have a file where you give specific instructions to AI crawlers through it, but it's been more or less proven that it doesn't have any impact at all.
Simon: There's people selling that as a service though.
Paris: Oh yeah, "we'll create you your LLMS file," it sounds great, it sounds like of course it makes perfect sense, we have to do it. But the real studies that have been done, and I've read about probably at least a dozen different studies that have measured the impact of sites that have an LLMS.txt file versus those that don't, all consistently say there is no measurable difference or impact in the results, in the share of voice they're going after. So it's basically been proven that this file has no real impact.
Paris: But my point was that if something is easy to do and there's a chance it helps, then the obvious decision is to just do it if it doesn't hurt. It definitely doesn't hurt, it might help, and it's easy to do. If it's in those three categories, go for it. It's a one percent insurance policy.
Simon: The one percent rule, right, you can be... why not, and all of those one percents add up to a lot, because most people can't be bothered to do the basics.
Paris: Yeah, Simon, now I see you've got the content pipeline pulled up on the screen, and you've been in this now for a few days, thanks for your patience with all the setup stuff, because I know it's not all smooth all the time. There's the CMS integrations, Google Analytics, Google Search Console, DNS records for your site. But now that's all done and you're really ready to rock and roll, and you've got ten topic recommendations on the board. You have twelve pieces, is that twelve? You have twelve in progress, I think, in drafts.
Simon: Yeah, yeah.
Paris: The second column, writing in progress, and then you've got ready to review, which means you've teed those up for your client to approve. Tell me, what's been your first reaction to the content? Is it good? Just from your first impression, what do you think?
Simon: The structure of it, and the detail, and from the fact that we've served it, we've given it a lot of information, and I think that's going to be key as people go forward. I was talking to a client this morning about recording his calls, and there was a little device, you showed me one that was a watch, didn't you?
Paris: Yeah, MemoCat, what was it?
Simon: Well, this client showed me one called Pocket. It's like a little credit-card-sized device you can put on the back of your phone, and it's recording everything if you want. The cool thing was it's AI enabled, and it's going to take all the main action points out of meetings, tee stuff into your diary, draft emails up for you, and send confirmations and quotes. It's going to do all of that before, by the time you've got back to the office, it's all done.
Paris: Amazing, isn't it? So you have a meeting out of the office, this thing records, and let's take it a step further, because I've been using Grok this week, you have a bunch of agents standing by ready to do the work and take the tasks and perform them.
Paris: Imagine you get back from the office and the work is already in progress or done by your AI agents already, and you get some kind of message that says, hey, congratulations on the good meeting and the nice lunch you had, and welcome back to the office. By the way, we had fourteen action items from that meeting, and seven of them are done, and the rest are in progress by this, this, and this AI agent, and they'll all be done by 5PM today. I think we're going to see that world this year.
Simon: Yeah, a hundred percent. Whether everyone adopts it, that's the...
Paris: The early adopters are going to be there.
Simon: The more important point we need to make throughout this session is we're trying to get away from AI slop. The point being, you'll be able to use these voice recorders rather than having to go on to Claude or whatever, press record and talk to it, and so on.
Simon: That particular device has a direct integration to ChatGPT, Claude, and so on, so it can go away and look at your skill set, whether it should be building new web pages for this particular project while Simon's in his meeting, or creating the quotes, and add everything he's talking about, and his client's talking about, continually to the project, so uploading that to my knowledge base to instruct further, so the next meeting you have, it's probably laid most of it out for you.
Paris: Yeah, it's like in the medical world, the analogy would be taking pills versus being hooked up to an IV. If your body is the AI, you might as well hook it up to an IV instead of taking pills or getting shots, just have it constantly flowing in.
Simon: Take your vitamins. Well, this is just all day, every day.
Paris: Yeah, you have it in the bloodstream constantly. I think that's inevitably where we're headed. I had a conversation over lunch with Del Cho, who's our head of product for GEOforge, and we were talking about integrations for BaseForge, because right now we only have one, which is Fireflies notetaker, and you have to manually sync all of the meetings, then select which ones, then upload them manually. So it's still a little bit of a clunky process.
Paris: I was telling him about this new category of devices, the one you mentioned, the one I talked about last time, that will be with us, and with one quick press of a button at any point we can just record our surroundings wherever we are, and it's going to take notes, think in real time, and send it into our AI of choice, whatever that is.
Paris: I was thinking it would be awesome if BaseForge could sync into this whole class of devices, and you could say, "MemoCat, I want to sync my MemoCat into BaseForge," so that when you come back from that meeting, ContentForge, what we're looking at right here, has already drafted a blog post about something it took out of that meeting you just had, which would be amazing.
Simon: We're in a whole world of possibilities, and we're at a point where those types of devices will be used all day long. I even emailed the guys at Pocket, I felt a bit... I sent an email to the Pocket guys saying, would you like me to review the product and see if I can plug it into workflows for customers in terms of generating knowledge bases, which hopefully is interesting.
Simon: There's one I think called Whisper, that's another one, but that's just an app you can put on your iPhone, or... WhatsApp, maybe WhatsApp should share all the recordings they're making of us and allow that, imagine that knowledge base that already exists.
Paris: Oh yeah, and it definitely does exist, no question about it, the ads are proving it, we talked about it last time. You just have a conversation and suddenly you see an ad for something, and you haven't done a search, you haven't clicked anywhere, it's simply just a conversation and then there's an ad for that.
Simon: So I think one of our further episodes will need to be around, and you just said it, no click. You said click an ad, or click onto a website. We are entering a world where it is no click. The conversation I had with a customer earlier was, the issue, if you're going to get cited in AI at the moment, it's a no click universe.
Simon: You get the answer, and you might then go and search the website you've just been recommended, but as I understand it, and correct me if I'm wrong, there's no attribution between you and a brand that's mentioned in AI or cited, it doesn't normally give you a link you can click, and you definitely can't attach a UTM tracking link or a pixel to follow that through.
Paris: No, not generally. The citation links are clickable, and those typically will lead to that brand, whatever the source page is, but a lot of times when ChatGPT just gives you a list of recommendations of brands, it's just a list, and often there are no links. You can copy-paste that and go to Google, which I think a lot of people do.
Paris: But in many cases, the first actual click is from a navigational Google search, meaning ChatGPT has already convinced them you're one of the top options and they're not doing a research query, they're searching for your brand with the full intent to find your website as quickly as possible and go there and call you or fill out the form.
Paris: I think that's why I really believe if that's the future of human visits to websites, basically get me there and get me the fastest path to convert because I've already been convinced through the chat, then we really are now designing these websites and publishing content for AI training more than for human experience. Maybe it's a little bit sad because it's a very different new chapter of the web as we know it, but I think that's the truth now.
Simon: I published some articles probably more than ten years ago now that were about how does Google defend itself. We're in a world where we are going away from traditional search, and you've got scenarios where, I'm a big Claude user, I talk about it, I use it all the time, and other platforms are catching up really quickly, you can't be switching between all of them.
Simon: Google would obviously want you to use, for example, Gemini or their AI search, and what you said was really interesting about the fact that people want to get to the result as quickly as possible, the right result.
Simon: If you look back at traditional SEO, you judged content by how good it was, and did it answer somebody's question and give them what they wanted as quickly as possible, because Google knew if it put seven or eight very good results on the front page and a couple of spurious ones, and people click on a result that isn't great, the overall experience of using Google drops.
Simon: So what you're talking about, getting the right result as quickly as possible, is where the competition will sit for all of the AI agents: how many of my people get exactly what they want within X number of seconds.
Paris: You know what wouldn't surprise me, Simon, is that in the same way Google was measuring these click-backs to search results, the pogo-sticking effect, meaning I didn't get what I wanted from that answer and I'm going to click on another link or rephrase my search, was a negative signal for Google that whatever page we ranked wasn't good enough.
Paris: I'm wondering if AI, like ChatGPT, is doing something similar, where if I write a prompt and get a response, is my next prompt a natural follow-on that takes the conversation further, or do I rephrase my prompt because I didn't like the answer?
Paris: Being able to analyze whether it's a rephrase, meaning the first answer was a bad experience or incomplete, or a natural extension of the conversation, which is good because it means we're bringing them down that funnel or deeper into the conversation, I'm sure they probably have a way to do that.
Simon: Absolutely fascinating conversation for nerds, right? We love that bit, but I don't know that your average website or marketing person is going to try to optimize for that experience yet. But it may well come a time where you are trying to fall into all of those following questions around the topic.
Paris: Which also raises another interesting point, which is everything we're doing now in GEO is optimizing for the first prompt, the opening line in the conversation, but most conversations go on and on, and what about the follow-up prompt, and the follow-up to the follow-up? I think it's hard for people to grasp that right now.
Simon: Or which part of that conversation actually leads to the piece of business. Starting the conversation and giving people a bit of education, telling someone a story, in marketing we talk a lot about telling a story and having multiple touch points to get to the point where somebody wants to convert. You won't do that with AI slop.
Paris: No, you won't. I think that's a great way to bring it back to the beginning, because we do need to wrap up in a minute here. We all know now that AI slop is a bad word, we're all trying to figure it out. It's a gray area for sure, Simon, it's not a black and white issue, it's not either slop or no slop, there are very varying degrees of sloppiness, so we're trying to be as anti-slop as possible here.
Paris: I think the formula I strongly believe in is that the more proprietary knowledge you can bring to AI, the less whatever your output is going to read as AI slop. I think that's the golden rule, and that's really the golden rule of GEOforge as well: no content is even really allowed to be published which is purely derivative. Everything has to be grounded in that knowledge base.
Paris: If you don't build a knowledge base in GEOforge, you cannot produce content, you're unable to, because then it would be slop.
Simon: And I have seen examples already, even though Stitch has got such a huge knowledge base of topics that GEOforge has suggested, and then I try to generate and it will say we don't hold enough knowledge on this particular subject just yet. That's then encouraging me to help produce that. Maybe one tip would be if you can add in the prompt that then says to me, it doesn't exist at the moment, go and find it, or can you write me something right now?
Paris: Yeah, there's a gap, there's a clear gap that needs to be filled, so our job is to race to fill it.
Paris: Great. Well, Simon, it's so easy, I feel like we're just starting to get rolling now, but for the sake of the audience, we've always tried to cap this at about thirty minutes and we're approaching forty, so why don't we wrap this up? We can start thinking about our weekend ahead. Thanks again, and just so our audience knows, we're now committed to Tuesdays and Fridays, so for those who have listened and enjoyed this, keep tuning in.
Paris: We're going to be back on Tuesday of next week, same time, 2PM UK, 4PM in Bulgaria, and hope to see you all then. Alright, sign off.
Simon: Bye. Cheers.
Paris: One second, there we go, got the right button.
