A GEO audit shows how AI platforms like ChatGPT, Gemini, Perplexity and Google AI Overviews describe a brand, and Ann Stanley, Founder & CEO of Anicca Digital, has built nearly 100 of them in the past year using Claude Code. She tells Paris Childress what those audits revealed about big brands, why incumbent SEO agencies were the biggest sales barrier, and why prompt tracking needs far deeper sampling than most tools disclose. The second half is a practical guide to Claude Code for marketers: skills as instruction kits, chaining skills into workflows and agents, and building analytics tools like Armadello.
Ann Stanley has built nearly 100 GEO audits in the past year, and Claude Code did most of the heavy lifting. Her conclusion from all those conversations is that many big brands still don't understand how AI answers are built, and that GEO only lands when it is sold alongside SEO rather than against it. The second half of the conversation turns to how marketers, not just developers, can use Claude Code skills, workflows and agents to build tools of their own.
You train an AI coding agent on a manual process and give it real data. Ann's SEO team at Anicca Digital had already built a manual audit. She trained Claude Code on that version, then added data: AI visibility reports from Waikay, the tool built by Dixon Jones, plus Ahrefs data to check how each brand performed in SEO, because a brand that does poorly in GEO sometimes just has a weak site. The result became a 12-pillar GEO audit framework, with color-coded analysis and content clusters built from Waikay's recommendations.
The hard part was collection. Many AI platforms block bots, so the workflow split in two:
Once the system was trained, Ann was producing between 5 and 10 audits in about two hours. Her caveat: the learning curve was steep, so the effort only pays off when you know you will repeat the task many times.
They win attention more reliably than contracts. Anicca used Instantly and Clay to personalize cold outreach with real Semrush visibility data, and about 5% of recipients asked for a report. Big brands that would not normally answer a cold email took the meeting because the email was specific about them and their competitors. Ann's verdict is that the project worked as a branding exercise, and that to use it for lead generation she would pitch it differently and add more follow-up.
Most didn't understand the mechanics. Ann found that many of the brands she met did not know how LLMs work, the difference between training data and grounded search, or how chunking and vector databases differ from search engine indexing. The second barrier was commercial: they all already had an SEO agency or an in-house team, and didn't see GEO as different enough to switch.
"If you're gonna pitch a new service like GEO, you've really gotta combine it with the SEO story." Ann Stanley
The screen grabs also exposed brand problems directly: mentions that were not recommendations, wrong facts, and in one case a perfume brand confused with a Covent Garden street of the same name.
Because AI answers are probabilistic, and most tools sample too thinly to show it. Building on Rand Fishkin's research into answer variability, Paris argues that prompt tracking cannot work like keyword rank tracking. Ann adds that data is expensive, so many tools scrape once a day. Paris describes his own approach: every prompt runs 30 times, prompts update every seven days, and the reported line carries a margin of error of about 1.5 points either way. Ann raises two more problems: results from the API can differ from what users see in the interface, and there are reports of platforms showing bots different results from users.
That is part of why Ann now sees GEO as more of a branding exercise. Even so, Anicca has seen a big uplift in UK leads, many of whom mention ChatGPT, and they convert well even though that traffic is still under 1%.
Start with skills. Claude Code runs on your own machine, can access your files, connects to tools like email, Stripe, a CRM or Google Analytics through API keys in a .env file, and runs scripts such as Python and Playwright for you.
"A skill is like an instruction kit." Ann Stanley
Ann's rule is to keep each skill specific and string skills together. Her weekly content chain shows how:
The whole chain takes just over an hour across two days, against about four hours by hand. Linked together, skills become workflows, and agent-style "commanders" in Anicca's C4 platform choose which skills to run for a role such as head of social media. Paris describes a similar setup for this show, where one Grok Bot agent takes an episode from Descript through YouTube, Transistor, the episode page and a LinkedIn article.
Ann's closing point applies to GEO as much as to analytics: data without insights is useless, and insights only matter when they turn into action. Experienced marketers who know which questions to ask, and what story a chart should tell, are the ones best placed to build with these tools.
Paris: Hi, everybody. Welcome to The GEO Show, brought to you by GEOforge, full self-driving for AI visibility. Today, I'm thrilled to be with Ann Stanley. Ann founded Anicca Digital back in 2007. She's trained over 1,000 marketers through the Anicca Academy. She's a regular speaker at events like brightonSEO, where we first met each other. She has built now and delivered, nearly 100 GEO audits just in the past year based on a 12-pillar GEO audit framework.
Paris: She's got an interesting background that includes paid media, and she puts on something called the Thursday AI Club, among many other things. So Ann, welcome to the show.
Ann: Great to be here, Paris. Really, really nice to... Well, we've met now before, but, to actually get to talk
Paris: Yeah. So even just in that quick, bio, there are so many interesting things we can dive into. What stands out first to me is this, in the past year, delivering almost 100 GEO audits and you've, through that process, have developed a 12-pillar GEO audit framework. Can you tell us more about that?
Ann: Yeah. So basically, I don't know about you guys, but over the last couple of years, business, particularly in the UK, has been a bit rubbish. So we're all, we've been talking about, GEO... Well, I've been talking about GEO at conferences and things for four years now. And I noticed in, Google Trends that it was a really growing popular subject.
Ann: So we thought there must be people out there that are interested in this. So what we decided to do, was do some cold outreach using tools like Instantly and Clay, which allow you to personalize the outreach with, real data about that business. And to be honest, what we did with that was we used some Semrush data to see how their visibility was compared with some competitors.
Ann: Sent them the cold outreach, and about 5% of them responded saying they'd like a report. Now I'd already got going with Claude Code by this time. I don't think I could possibly have done that many, on my own. And the team, 'cause, you know, we've got a team of people that do SEO, they'd already built a manual version. So what I did was I trained Claude Code with the manual version to actually produce Version. But I was able to, add real data into it and we used Waikay. I don't know whether you know
Paris: I do, yeah. I actually met Dixon Jones, I think in, like in 2009 or 2010, at Search Engine Strategies London.
Ann: Well, he was he's
Paris: Yeah, when he was at Majestic back then. Mm-hmm.
Ann: Brilliant... Yeah, he's got a brilliant piece of software, called InLinks, which was about entity S-SEO. So he'd built this really So he was really early on before all the investment money went into, GEO measurement tools. So I'd already got, an account with him. So as these inquiries came in, what I was able to do was set up a project within Waikay.
Ann: I then, used a Chrome plugin that Claude provided me to go and scrape, seven, LLMs. Just asked a really simple question: What do you know about this brand? I then used Ahrefs data to see how they were doing in SEO because, of course, in some cases, if they're not doing very well in GEO it's because their site is rubbish. They've
Paris: Yeah, there's often a correlation. Yeah
Ann: So you need to understand that. And then from there, what I did was, we were already using, a Claude project to create really nice content clusters, which came out of the recommendations from the Waikay report. And so you ended up with a really nice, structure where effectively you explain at the beginning the difference between the training data and what's actually, when they go out and do the grounded res- search.
Ann: And then you show them what they look like, within each of the LLMs. So basically, Claude Code would grab the screenshots that I had, saved. It then did the analysis and gave me a color coding. You know, red was there was a problem, green was great. And then I, then, I... It used a tool called Playwright to go and get the information, from Waikay, and it used the API, which of course Dixon, I think Dixon's actually developed an MCP now.
Ann: But at that time, this was back in January, February time we were doing this. And then what it did was it then did the analysis, it put in the Ahrefs, and then it created the content clusters. So I think that adds
Paris: Wow.
Ann: 12
Paris: That was a lot. I was gonna try to repeat that back, but I, it sounded like at least 12. One quick question that, Before
Ann: So. I
Paris: It, you mentioned that there was a Chrome plugin that you used to, just to collect whatever AI knows about your brand, so you would just ask it, "Tell me everything you know about brand X." What was the name of that?
Ann: Yeah. So what I would do, one of the things about using, and one of the problems with a lot of the tools that we talk about, GEO, is that they're all blocked by all the bots. So Playwright is a tool that Claude Code will use a lot. It's free. It's actually a software testing tool which actually acts as an agent.
Ann: So it will go and it will log in, it will go through the cookie consent. So what I would do is I would set up my accounts, within, Claude Code, and it will go and log in and after 24 hours of getting the data, it would then go and grab the screenshots. That took a lot of training 'cause it was really bad at grabbing the wrong bit of
Paris: And with the screenshots and,
Ann: Of
Paris: Screenshots of what exactly?
Ann: Well, So Waikay's So the advantage of using something like Playwright is you can actually go to another tool online and go and grab results and put them into your report. So if I'd had to go and grab those screenshots, that would've taken ages. Plus it gave me then the analysis. But the thing it couldn't do is it couldn't go to ChatGPT, it couldn't go to J- AI Mode, it couldn't go to AI Overviews, Perplexity, to go and grab what, AI knew about that, was because they were So what it did in...
Ann: That make sense? So certain things it could go and grab, and certain things I had to go and grab. So it's a really good exercise in knowing the art of the possible of what, these tools can Can't. then what happened then was it gave me a Chrome, a Chrome extension, which I just refreshed, and then every time I added another batch of...
Ann: I was getting between 5 and 10 a In. So it would go and add those, into, the Chrome extension, and then it would create a file on my hard drive to save it, and I would basically go to ChatGPT, w- "What do you know about, brightonSEO? What do you know about Anicca Digital? What do you know about Next PLC?" Whatever it was, and then it would, grab the screenshot and save it in the right place, so then it knew where to It.
Ann: So if I'd had to do that job manually, it would've taken me somewhere between around four But I was actually doing between 5 and 10 in about two hours once I got it sorted. But it was a pretty steep learning curve, a lot of training, particularly to get the screen grabs. So it shows what you can do, but it's a lot of persistence.
Ann: And if you know you're gonna be doing a lot of something, then it's worth the effort. But if I was only doing five, it would've been quicker to But because I did so, because I did nearly 100 of them over a The effort that went in at the beginning, you Later on.
Paris: Absolutely, yeah. It sounds amazing. And in the end, were you able to save everything? Do you now have this, framework i- in a repeatable format?
Ann: Yeah, I mean, I hadn't thought about giving it away. You have to do it on Claude Code 'cause it needs tools like Python. You'd obviously have to have an account with Waikay to get the data, although I guess you could use it with an alternative tool. I got the Ahrefs API 'cause we've got an account with Ahrefs, so I had all that elem- element.
Ann: We definitely could give the s- the cluster, tool, content cluster tool 'cause that was one of the things I showed at brightonSEO, how you could use... I did a workshop on how to use Claude Code, as part of your, SEO or your And actually the content cluster was one of the ones that, I illustrated. So I could definitely give you that skill to share with your audience,
Paris: Amazing, yeah
Ann: Because that's... But you'd have to, you'd have to load some information, whereas the other stuff, it was just that, if you can imagine you're going... When you set up Waikay, what you do is you set up the client and two competitors, and you choose two topics. And then for all of those combinations you end up with, like, four, so you end up with about 16 screen grabs. Well, you can
Paris: Six
Ann: You've gotta go and grab Screen grabs, do the analysis. If you were doing that manually it would be nearly So it was, really... And it was an early thing I did as well, so it was... So I probably could do it much faster now, but at the time it was one of the first things that I did. So it was a huge learning curve of how to make it look right, how to get your brand in, how to grab the content, what you could do with Claude Code, what you
Paris: Yeah. And in the end, was it successful? Did you win clients through this?
Ann: I think it was... I mean, it was a branding exercise rather than a lead generation exercise, 'cause we got in front of some really big brands that I don't think would normally have spoken to us. We spoke- we had, of the 70 to 90 that we did, we had meetings with about one in five. So one in five followed up with a face-to-face meeting, and of course, that meant they knew We were.
Ann: But we- I think we only got work out of one. So on that basis, if it was- if you're doing it purely for lead generation, I don't think I quite pitched it right, so I think I would need to, do it slightly But for a branding exercise, we got in front of some really big brands who normally you would think, "Well, why are they responding to a cold email?" And they said to us when we asked them, was because the cold email itself was so well-written, it had specific details about them and their competitors, that actually they thought, "Well, somebody's put some effort into this." 'Cause you know you get hundreds of rubbish Every day.
Ann: So in a way, there was as much work went into the cold outreach using AI to do that as there was in actually developing the report, and then doing the follow-up e- session, which was about an hour. So yeah, it was a big project, a big exercise, which worked from a branding point of view but, would definitely need tweaking and a bit more follow-up to make it into a
Paris: Yeah. And what's your sense of how these brands, especially the big brands that you get... You got their attention. My feeling is there is a lot of curiosity right now about GEO and its relationship to SEO. I tend to think that it can be part of the branding, the brand marketing and product marketing world just as much as the performance marketing world.
Paris: Did you sense that, through your conversations with the bigger brands and their interest that they're thinking about GEO in a larger sense than they have thought about SEO?
Ann: So we're going back to February now when there was a lot of interest. I mean, obviously there's a lot more interest now. I mean, in Brighton in October, nearly everybody was... That was all they were talking about. That's all the stands were really. Three-quarters of the stands I wasn't at Brighton in April this year. I missed it, unfortunately. There was two things that came across really strongly.
Ann: First of all, they hadn't got a clue. They didn't know how LLMs worked. They didn't know the difference between training data. They didn't know anything about chunking and the way that, you grab data from an LLM is completely different from the way that you know, SEO, search engine optimization indexes. They didn't know about vector database. They didn't know anything about that.
Ann: So the first five minutes I sort of had to explain some of that to The second big thing, so there was a really la- a very high level of ignorance, or lack of education, I guess, is the, a more polite way of, it. The second thing was they all had incumbent SEO, agencies, or they were already doing SEO in-house.
Ann: So that proved to be the biggest sales barrier. That they didn't see that it was different enough from SEO that they would kick out their current agency and employ a So that was the main difference the other feedback that we got was the clarity of the way we laid it out meant that they got it, which was really good.
Ann: And there were some people, there was a few There was one really weird conversation with, ... 'Cause we did UK and u- and US market. We wanted to test the US market as well, let's see what the American market was like. And there was one SaaS business who was They were an AI company. I don't know how we ended up with that, 'cause we started off with mainly commerce businesses, which is our strength, and then we ended up with a couple of SaaS businesses.
Ann: And he said that was impressive was the fact that we absolutely knew our stuff. And I got the impression that there was already a lot of, you know, BS out there, and there was a lot of people that were talking rubbish and didn't know what they were doing. And what he said about us when we I'm very technical, I was ...
Ann: And I, and I'd spent the last two years working on AI, that I understood how LLMs worked, and I understood, you know, what GEO was about. And he said that was quite rare, that there was a lot of people going in, and they didn't really know. They were just selling with a new So I thought that was really interesting.
Ann: So there's definitely some stuff around that which, I guess, you know, 20 meetings was still really helpful because I think getting to talk to customers at different stages and understanding that if you're gonna pitch, a new service like GEO, you really gotta, you've really gotta combine it with the SEO story, I definitely pivoted to quite early on. And secondly, you are displacing another agency that's or an in-house team.
Ann: So, I think that's, really interesting. And- The stuff that Rand Fishkin was doing about, frequency of, variability in results had s- just started to hit the press. So some of the more knowledgeable ones were aware, well, you know, how often do you need to take a m- you've only done this for 24 hours, how realistic is this? A- and in some cases, the mentions versus recommendation really came out strongly just by going and doing those screen grabs.
Ann: Because there was one company, I won't mention the name, they were a perfume company, and there was a street in Covent Garden with exactly the same name. Well, it wasn't them that was being mentioned, it was the And in some cases the facts were wrong, and also they were, hallucinating, brand, but they were getting wrong information. But also, they weren't necessarily saying good things about them.
Ann: So I don't think people actually... A bit like your own SEO results, if you don't actually, search for yourself, what does it show compared with everybody else? Well, in some cases I don't think people are actually even checking whether their own, brand is, So I, so I think there's a branding exercise around that, but I do agree with you when it comes to measurement, it's more like a display advertising, measurement that we need to be looking at, and the user journey is more like Say
Paris: Yeah, there's a lot there. I think that what Rand Fishkin started to uncover months back, and I think it's starting to become more accepted, is that you can't approach prompt tracking in the same way that you do, keyword rank tracking for SEO because it's probabilistic in nature. If you run the same prompt 10 times in a row, you're likely to get 10 different answers, 10 different... Or even 100, yeah.
Ann: Or
Paris: And so you need to think of it more as a sampling exercise, not as a precise deterministic, checking of rankings. And also they're changing so often and they the citation, the list of citations is changing s- so often. That was never really true with rankings. They rankings tended to stick in SEO. I mean, you would move around a little bit and it was a big deal to go from position three to position one, and that might take you months and months, but then once you won it, you tend to stick there and with, a good d- defensive strategy.
Paris: But the, th- this is one thing that I think a lot of the current tools aren't doing well enough, is they're not disclosing their depth of sampling. And then the other point,
Ann: And
Paris: S- and the frequency. Yeah
Ann: And I think the reason is because it's so expensive because they're ... In a lot of cases they're buying data from similar places. If you want to try and scrape yourself, it's really expensive and you get blocked, you know, very easily. So there is a few sources of data. A lot of them are using Data, and in order to make the product affordable, they're usually only scraping once a day. They might be scraping a large number of prompts, Then, those prompts don't necessarily reflect what Really,
Paris: Yeah, that's a whole nother issue too. You have the prompt One issue is the prompt discovery process, and are you actually how do you come up with the prompts and the right prompts? And then the other thing is the depth. The depth and frequency of sampling is often very shallow. And then the tools are not really disclosing that they have they actually have low confidence in the results.
Paris: The margin of error is very big. One of the things we set out to do with GEOforge from the beginning was that we run, we update it, we update the prompts every seven days, but every prompt gets run 30 times in a row. And that incurs quite a lot of cost, especially from OpenAI's API, which forces us to charge higher prices than most of the other tracking tools.
Paris: But we report on the actual margin of error, and we even draw that in the graph. There, there's a buffer around the line that shows, okay, one and a half points above and below is our margin of error. But beyond that, this is trustworthy to this extent. But the market's not quite there yet in terms of its, education, sophistication. People just think it's gonna run once. It's gon- I'm either there or I'm not there, and that's the truth
Ann: Well, the other problem is there's two other big problems with what you just said. The first one is, lots of people say that when they do the comparison between the OpenAI API, Are very different. Secondly, Gemini has just started to show diff- I think this was Cyrus who said this the other week, last week, that what they show in the user and what they show in the bots are completely different. And in fact, they're faking and making mock data
Paris: Yeah, I've been reading about that
Ann: There's a whole pro- whole load of problems around that. So for me, I feel that, GEO is now more of a branding exercise. And you've got to go back to basics of branding, and therefore, in the same way that if you were measuring the impact of social media branding on them coming on and doing a sort of an organic SEO or going directly to the resi- the sites, that's the benefit.
Ann: But ironically, we found, in the UK, we've seen a big uplift in the number of leads. And a lot of them are coming from ChatGPT. Now, it could be just that that's what they think they searched on, but they are consistently mentioning it. And they're asking for all sorts of weird stuff. And secondly, the conversion rate seems to be, even though the traffic level's really low, sort of less than 1%, we're still getting good quality leads, so they're converting better 'cause obviously they've already done their research.
Ann: So I think that's all really interesting. But I mean, I think, I think if you go back to GEO and SEO generally, there's still a big overlap of what you need to do. And then if you then look at something like Claude Code as well, what you can do with Claude Code for your digital marketing and your search toolkit, again, it's sort of like, ever-increasing circles. There's so much more that you can do. We got into the weeds quite quickly, didn't we? I didn't expect us to do
Paris: No, that's great. I love getting in the weeds and I think our audience is able to handle it and technical enough. And let's pivot into Claude Code now, 'cause we talked about that in our, in our prep call quite a bit. Most of, most of the folks in my agency, we're all on Claude. But there's only one person, our head of product for GEOforge, he's the only one that really uses Claude Code day in and day out.
Paris: Everyone else is using Claude Cowork, which is merged with, Claude Chat. And there's this, I guess, kind of a fear of that Claude Code is just for developers, it's a coding tool, and, it they know it's more powerful, but they never go there and they don't know what's possible. But could you help us generally demystify this? 'Cause I know you've done a lot of great work that you just described earlier with your 12-pillar GEO audit framework.
Ann: Yeah, that's one of I mean, we're even building full SaaS products now, but, which we'll come onto. Yeah, so my journey with Claude started over a year ago, probably 18 months ago, when I started to prefer output from custom GPTs and ChatGPTs we were building. So, the first thing we really got into was Claude Projects, which are still available within chat, and actually Claude Projects is being developed much further, which will, you know, you'll see a lot of news about that over the next month.
Ann: I think it's got some really exciting things that are happening. And also about a year ago, skills started to develop, so you started to see things like, "I'm going to invoke the PowerPoint skill or the Excel skill." So, the language of skills started to be introduced, and Claude was the first one to use that. Interestingly, you... With a bit of technical setup, you could use Claude Code on your laptop or your desktop over a year ago, but it was very much in the terminal, you ha- or you had to create a server for it, and it could access your files and nobody was really using it apart from developers.
Ann: Then around, October, it became a little bit more obvious that you didn't have to be a developer, but you could use something like Cursor or Visual Studio and create these chats and start to use it without having to use what we call the terminal, which is that little black box that your IT team comes along when they wanna talk Laptop.
Ann: that was when we... And I was already vibe coding with Cursor anyway, so I'd been vibe coding, first of all with Bolt, and then with Cursor last year. I was already pretty familiar with using Cursor, so that step wasn't a big step. But the big thing that changed was in May when they put Claude Code into the desktop, and then they made it available so everybody could download it, and you didn't have to touch anything over Or Cursor or anything like that.
Ann: And that really then opened the floodgates for suddenly everybody being able to use it. If I explain how we got addicted to it back in February. So, we were... There was a few things that triggered it actually. We were already heavily using n8n to do automations, and we knew we needed to do something else. And, Tony Hancock, who's now joined by Anicca AI & Insights, which is my AI spinoff, he'd started to play around with it within the more term- the terminal to...
Ann: He decided he wanted to rebuild every single Shopify app that he was spending money on. So he just With that. And then I think OpenClaw was announced in January, so there was this big hype. So we knew he was playing around with that. Our team were already doing quite a lot of stuff with Google Scripts, which meant that we could create some dashboards, and we'd already got this website that we were building called Secret Agents, which is sort of semi-retired now.
Ann: But, we went on a course with Mike Rhodes, who talks a lot about the brain and how to create your own agency brain or your own business brain. And we did a couple of things with him, and then we just It. and so the main thing about Claude Code, which you need to be aware of, is that, now easy to access through your browser, or you can even access it on your phone, but it accesses your desktop, and that's the big dis- differentiator between something like Cowork.
Ann: You can also use it to link to all your other softwares through API keys, and the way you do that is you have something called .env file, environment file, on your local machine, which means if you want to link to your email or Stripe or your CRM system or Google Analytics, in some cases, you link through that API key.
Ann: Some cases, you just log on. Sometimes there's an MCP server. But basically, it gives you access to everything Do. It gives it access to all your files, but it also uses tools. So it will run things like Python and JavaScript to go and do, you know, those scrapes or Playwright. So suddenly, you can talk to it as a human, it instructions, and it will go and do You.
Ann: So it suddenly is a general... So it's not just you ask it a prompt, and it gives you an answer. You of course can do that, but it's now got all this extra functionality For me, the big thing that made the difference was le- was skills. Now, skills I think by this time next year will probably have gone, but at the moment they're really fundamental.
Ann: So a skill is like an instruction kit. Now, if you've ever created a custom GPT, or a Claude project, you've normally got a prompt, which is your instruction. You've got some files, which is your template. So a skill is similar to that, but it also can have those sort of JavaScript and stuff as well. So what I do is two things.
Ann: I either go and get it to do a search and find a skill that somebody else has built, so I borrow that skill and then I will customize it and make it work, I build my own. So when we did the GEO reports, I was working with it to create the different elements, and then it would create that as a Skills.
Ann: So the first thing was the skill about creating the Chrome extension. Another thing would be about creating the Another one would be about creating, going off to Waikay to go and get that data or doing the A- And then that would all be all together, you could call it
Paris: Can I jump in with a quick question about the skills? Because we, We had a period where there was a big explosion of skill building, and we were always wondering how narrow a focus should a skill have versus giving a skill a multi-step workflow? What's your experience there? 'Cause what you just described sounded like you were giving them pretty narrow focus areas, and then you string them together
Ann: The same. Yeah. So in exactly the same way that when you use Claude Code, you have one chat or project per task, and that normally relates to one file in your laptop, and there'll be a memory.md file which follows that through. You can hand over it to a new chat, particularly if you, if you s- if you work on it for more than an hour and a half and you're, and particularly if you're doing something like building a function on a...
Ann: Something like Armadello. I've been working on that since February, so I, you know, I'll be handing it over a couple of times a day. But every skill should be very specific, and then you can use them and string them together. So the best example is what I do every weekend, and every Monday morning and Tuesday morning. So on a Sunday, I've got a skill that goes and scrapes 25, 30 sites and brings back, a load of new blog articles on everything to do with AI in marketing, AI news, etc.
Ann: So that is the raw material. The next skill takes that data and then writes a blog in a way that I've specifically trained it to do. So I spent a lot of time, I call it building it and bottling it. So I build it, and then I will save it as a skill, and then I'll use it again, and if it's not quite right, I'll keep on tweaking it until it's perfect.
Ann: And then that way you get that skill. It then goes and creates me an image for the blog, and then it... That, Skill, and then it's got another skill which it loads it up into WordPress, and then I go and human in the loop will go and check it on a Monday morning, make sure it's fine. I then publish it, and then it gives me the LinkedIn post based on my LinkedIn skill, which then means I can then, post Then the following morning what I do is I get it to write me a LinkedIn newsletter because LinkedIn algorithm's so crap now you don't get So at least if you've got a newsletter, you get it into people's inbox,
Paris: I'm with you on the newsletter thing. I'm, I definitely understand that, yeah
Ann: And then there's, there was a different image. I used to create a carousel with it, 'cause you can use it to create images as well, which most people I don't do that now, I just create... I use actually a same- I've ended up using the same image as I use with a blog with the bullet points for the main content and then it- as a shorter lin- LinkedIn post.
Ann: And the whole process from end to end, although I do it over two days, takes just over an hour. And if I had to do that manually, it would be four hours just to do the So, and I do that every week, and I've been doing that now for six months. Now, I would have given up because it was just no way.
Ann: So that's an example of several skills that I've Together. and if you link them together, and I have done that, I've got a tool on one of my SaaS products which actually links these together into an agent. And then that will... You would call that an AI agent because you trigger it. In our case, it's every Sunday. And then you've got human in the But all the elements are created for you.
Ann: So I think a skill should be really specific because then you can work on that skill, but you can have a chat where you evoke several skills and, you know, so I just have a chat that I use every Monday and Tuesday, which has got everything in it. But each of the individual skills are separate
Paris: That's amazing. Yeah, it sounds, a little similar to my podcast workflow now. I'm able to publish, a daily episode of this show, entirely with the help of Grok Bot. I actually have moved a lot of my workflows into Grok Bot, and, they have these routines. But basically I shoot an episode just like we're doing right now in Descript. Once we stop recording, I hand it over to Grok Bot and say, "Go," and ep- episode 67 is ready in Descript.
Paris: So it has... It also has the, it has the connectors. Actually I use a tool called, Composio
Ann: Yeah, yeah. Yeah, that's, good for
Paris: Yeah, so I've got all, I've got all of my connectors in Composio so I can take it from agent to agent across different LLMs. And, so it has Descript, it has, it has connectors with YouTube and everywhere. So it goes through this elaborate workflow, I don't know how many steps, but it produce it edits the podcast in Descript, so you can do things like you can remove the filler words and the gaps.
Paris: You can apply Studio Sound, which is an AI feature, and then it even makes corrections in Descript. So it uses Descript as a real human user, produces a final cut, publishes it to YouTube with a detailed description, which is keyword optimized with timestamps. It creates a thumbnail image specifically for YouTube. Then it goes to Transistor and publishes there with show notes, and that's what pushes it out to Apple and Spotify, RSS feeds for the podcast.
Paris: It also creates a LinkedIn article for the LinkedIn newsletter, which I agree with you. LinkedIn newsletters are the way to go because you're getting into the inbox and you have much more guaranteed reach there. And also GEO, tends to favor LinkedIn articles as citations we've seen for B2B. And in addition, it creates a page, in the We have a particular, it's not the blog, but we have a whole section on our website for all of our show episodes, and it creates an episodes page with the entire transcript.
Paris: So just in case if AI bots wanna come and crawl that whole transcript, it's there. And that all happens in about... Usually that happens all in under an hour. And so we've got a whole kit of repurposed assets, and I didn't even mention the shorts. It usually creates three to five shorts for YouTube and TikTok, and it also...
Ann: How long did it take you to train all that? Did you already have
Paris: I had done it all with Skills in Claude, and then I discovered
Ann: And then did you give them all to Grok Bot
Paris: I did. I handed it all over to Grok Bot, and I mean, I really just trained Grok Bot on all the skills. I was feeding the skills that I had created in Claude to Grok Bot to learn how to do it. And the difference is that with Grok Bot, the workflow isn't broken up into a... Well, it is broken up into a bunch of steps, but there's one Grok Bot agent that does all of it from end to end
Ann: I was g- I was, yeah, I was gonna talk about that. So, what I was describing is the difference between y- if you think of a skill as your building blocks, what you can do is you can build, you can build a house with them all in a certain order. So that's a workflow. So I- on our, we've got this product called C4, which is Company Command and Control, and Grok Bot was the nearest thing that I'd seen, which I've been designing since February to, C4.
Ann: So C4 works, Has got two, it's got three ways of working. You can work with individual skills, which have been also, which we provide, but then you can customize them based on your brain, so it's got a company brain. And then you've got, what I call AI ABC, which is an agent, a building canvas, so you can link skills together So that's a bit like a workflow that we would've done last year with n8n, where it's pr- it's in a particular order.
Ann: It's still got brains in it, and you still can inter- interrupt it and do things with it. But it's, it- it's a predictable sequence 'cause you know you're always gonna do it. Then we've got something that's similar to Grok Bot called Commanders, and in... That's how Muse is working. That's how, Dots is working, and that's what the new Claude Projects is going to do as well.
Ann: And basically what you're doing with that is you o- and to be honest, it's been around for a lot longer. Relevance, which has been around for at least 18
Paris: Oh yeah, Relevance AI.
Ann: What you
Paris: That one. Mm-hmm
Ann: Yeah, an, yeah, an MCP servers on things like N- on n8n. So what you do is you, y- you, if you imagine it, you've got a head of social media. You talk to your head of social media, and then they will invoke any sub-agents or skills that do each of the individual jobs, but they'll choose the one that's right for that job at one time, and exact- that's exact...
Ann: And that's what the commanders do. Now, there is software out there c- called, I'm not gonna be able to remember any of the name. Marblism, I think it's one. Sana. There's a load of them. They've had l- loads of investment, and they call them AI employees, and they work very much on that basis, where you've got a head of, HR, or you've got a finan- head of finance.
Ann: You've got a head of social media. You've got a head of content creation. You've got a head of, you know, YouTube or creative or whatever it is. A head of paid media. I've created something like 23 of these commanders, to do specific job titles, but then underneath it's got a load of skills, but it's also got the ability to go and work out what to do as well.
Ann: So I think that's what's gonna happen. So Grok Bot, Muse, and OpenAI Dots, I think is the way that it's gonna But, there's a couple of issues with that, is if you haven't already built skills, think you can not gonna be as well developed. It's a bit like when we started using generative AI, if you were rubbish at prompting, you got a rubbish result.
Ann: I think the advantage of working with the stuff on your machine, and actually developing the skills is that you might use them in a different way in the future, but you're training your own AI to really understand how you work, the content, the tone of voice, and it's got access to your machine. Whereas something like Grok Bot is in the cloud, and I don't even know if it has access
Paris: It has a virtual machine. Gr- each, th- they have a vir- a virtual computer, and you can see it and watch it, and you can take control over it. One of the slightly annoying...
Ann: But does it have access to the files
Paris: Does. Yeah, it has both. Mm-hmm. Yeah
Ann: Okay. Yeah, so I think that's the way it's
Paris: Yeah, I think so too
Ann: But I think if you actually want to build a dashboard or you wanna build some SaaS products, then I still think at the moment Claude Code is the That. and we can sort of s- describe why that's You know, you can bring in your Google Ads data on But actually if you wanna create really good... I mean, and Claude Code's got a new functionality called da- So you can bring that, as long as you've got the API connected, you can bring that data in.
Ann: You actually want to build a product or build something that you're gonna reuse, then you probably need to start using slightly
Paris: Yeah. Yeah. The We've had a lot of fun already with, the new dashboards auto... It These are artifacts, I guess, types of artifacts, dashboard, dashboards and motion we've had a lot of fun with too, which is, pretty awesome videos now. But, gosh, I mean, Claude is just eating major pillars of the SaaS world, video editing, and th- that dashboards are beautiful.
Paris: I mean, we used to use Google Data Studio or Looker Data Studio for all of our client reporting, and we're in the process of moving that entirely into Claude's dashboards. They're just better. You can they're so much easier to use, and they're more beautiful. And then motion, the motion graphics as well. We're looking at, workflows where we repurpose written content, and this is a, this is kind of a GEO play as well, where we wanna repurpose every piece of content that we're publishing, we wanna repurpose that into a video because we see that YouTube gets the lion's share of citations in AI Overviews.
Paris: And that's probably one of the lowest-hanging fruits that we see in specifically for Google AI Overviews is just over-index on YouTube because the... It doesn't matter if you have very few views on the video, if your channel isn't popular. If you have the video that answers the question, it could be a one-week-old video with no views, and you're gonna get, you're gonna get there
Ann: Yeah, I have a feeling that we might be doing quite well in GEO because we've got so many videos from all Because of course all the transcriptions And it's really interesting 'cause two things that we've discussed in the last five minutes is my biggest fear about developing a SaaS product, is that the LLMs themselves will produce features that SaaS.
Ann: So I'm gonna talk about Armadello in a second, but sort of explain what we do differently compared with other tools, and that's the fact that, the problem with something like dashboard, it's not dissimilar to Looker Studio, is that it's pulling the And therefore, some of the dashboards that you produce, or the artifacts that you produce, aren't necessarily as fast or as good as what you can produce if you go a different route.
Ann: So the route that we've gone is, I mean, I- always, in my role, done a lot of audits, 'cause my main role is business development and consultancy, and understanding enough about all the channels that I can go into a client and f- and understand their business and advise them on what to do. So it's really about getting underneath a client and understanding the data that we See.
Ann: So it could be a GA4, it could be a PPC audit or SEO audit, or whatever it is. So I've been doing that for 25 years and, what that means in practice is that I'm... I've got a lot of gray hair, a lot of experience, and I know what I wanna see that will give me the results. And one of the problems with GA4, for example, is you can never get the data you wanted 'cause it was never in And it used to be really frustrating, 'cause if you'd used, if you're old enough to have used Universal Analytics or
Paris: Yeah, I remember. Mm-hmm.
Ann: Which I think was... I think it swapped over about five years ago. Even though the data might not have been as accurate, at least you could get the combinations, or you can have four or more metrics on the same graph, or you could create segments, or you could, you know, even till recently you couldn't even put the data in weeks and months until quite recently.
Ann: What we did with Armadello is we basically designed something like 100 pre-built, reports based on the fact that we brought the data into something called Supabase on a So every day we use GitHub, which is where, the rep- repository or the repo where all our data is saved. So all the work we do on anything to do with Claude Code, we back up into GitHub, which is a big step 'cause you don't, you know, you don't wanna lose work you've On.
Ann: So we bring the data in using something called GitHub Actions, and that brings it in overnight. We link into their e-commerce, or their CRM as well, and bring in across m- all the different platforms. Then, we've designed the tables so that all the data is there, and then we've got a whole load of pre-built, charts. But we've also got a chart builder and a dashboard, and we've also got AI Arnie.
Ann: So I don't know whether you can see, I've got an Armadello T-shirt. So, Armadello was named after a girl that we were working, called Lucia Dello, who, was building an e-commerce, a Shopify app, on analytics. And then we've sort of taken... They've sort of retired from that and we've taken it over. But, so Arnie is our Armadello helper and Because data's great, but without the insights, without the ability to run audits or know what the checkpoints are, it's useless really and doesn't really- add the value.
Ann: So what we've done is we've, And there's a demo store that you can go in, and you can go and sign up. It's free until, basically until the end of December to use it. But I don't think that dashboards will be as good Because it's the ability to know what data and how data you need to present. So it's the brains It of b- running m- a digital marketing agency for 20 years, and, you know, doing that as my main job.
Ann: And the fact that you can blend data from multiple points, which I think makes Armadello, really powerful, and then the insights which then turn into recommendations. But also, once you do that, you can do other things. So, you know, if you've got an e-commerce website, we can go and do price checking, once a week for you, or daily if you wanted to.
Ann: We can create product descriptions. So it's all the things that you can do as a result of that. And then that's how it fits into C4, because if you have the actions and then you integrate the two, you get a whole, AI operating system. But I think that later one is, i- the C4 side of it, I think if anything is gonna be overtaken by the L- LLMs, you know, what I've seen in the last month, I think that's gonna be the hardest bit With.
Ann: you know, you still need the, you still need the company brain. You still need all that
Paris: Yeah. I think that marketers, I mean, the experienced marketing wisdom and the taste as they say a lot, I think this still matters. You need to know which questions to ask first and understand the audience of whatever dashboard you're gonna be presenting. Is it a C-level audience where they re- really wanna just get the high level view, or is it, does it need to be more granular?
Paris: What story... Every graph needs to tell a certain story, and that's still where the human, I think the human input is critical in helping to design those types of charts. Yeah.
Ann: Yeah, and I think probably agencies will be the, one of the biggest users because they can... You know, it's multi-tenancy, so they can see all of their clients in one place, and it's so... You know, it's 100 times better than Looker Studio. I'm sure Claude Artifact will get there, and I know Gemini, Google itself is trying to create its own data, because they've renamed Looker Studio Data Studio now, aren't they?
Ann: 'Cause they're trying to make them more graphical. So there will always be people that are not prepared to pay, but I think our USP is gonna be around the fact that we're gonna be bringing in all the marketplaces, and retail ads, or we've got the CRM data, and we've been bringing in financial data. So it'd be a much more of a business suite, where the real IP is how you connect to everything, and also, how you use the AI, insights to then create the actions to make the d- Better.
Ann: You know, it's all very well having data, but data without insights is useless, and insights without action, you know, just wasting
Paris: Yeah, absolutely. Wow, there's so much that we can cover, Ann. This is, this has been fantastic. I my main takeaway is that you are a builder first, and you've built some amazing things, and diving headfirst into Claude Code, I wish that more marketers did that. I think it's not as intimidating as it might seem. You can... Of course, you can just chat with it and do amazing things.
Paris: But thank you for these great insights. And, would you can you tell people where they can find... Is Armadello available now? Is that something that people can come and access?
Ann: Armadello is, There's a demo-store.armadello.com, which I'm sure you'll If you go on the anicca.co.uk site, there's also a pop-up that comes up with a Claude It does need updating. I updated it in May, but to be honest, in the last month there's been so many updates. But if you're brand new to it, you definitely... And if you fill that form, you go in and there's a load more resources that you can access as well.
Ann: So, anicca.co.uk has got quite a lot of the resources. And then, if you were interested in the, Armadello and you want a demo, then you can go there and fill in a form. And the demo store's really good 'cause it's got nearly all the features, and you get fresh data in it every day. So Really good idea to
Paris: Great. Yeah, we'll share that in the show notes. Thank you so much for being with me today, Ann. It was a great conversation. I think we could go on for hours comparing notes and geeking out. But I hope this has inspired some people to roll up their sleeves and get into Claude Code and really adopt a builder's mindset now.
Paris: I think, marketers now have superpowers to build things that we never have had before, and I encourage people to do that. And thank you again, Ann, and good luck with everything
Ann: Thank you so much.
