LLM's, Prompting, Automating & Cheating with AI | E88

In this episode, we’re ditching the basic bitch prompts and going deep on real AI power for marketers. Emma and Ruta unpack everything from prompt engineering to custom GPTs to automated workflows—so you can stop treating AI like a toy and start using it like a pro.

We get into:
 ✅ The AI lingo you actually need: models vs. providers vs. platforms vs. benchmarks
 ✅ Prompt engineering 101: zero-shot, few-shot, meta-prompts & when to use “reasoning”
 ✅ Building your own Custom GPTs—for content, strategy docs, coaching & more
 ✅ Automations that turn transcripts into blog posts, spreadsheets into insights
 ✅ Data cleaning, contract checks & process docs—all without lifting a finger
 ✅ Picking the right model: why cheaper LLMs sometimes outscore GPT-4 on copy
 ✅ AI as your personal CMO-mentor: role-plays, feedback loops & impulse checks
 ✅ Content audits, gap analyses & campaign plans—instant, AI-powered QA

If you’re tired of treating AI like a magic wand—or you’re ready to stop copy-pasting prompts and start building real AI-driven workflows—this one’s for you.

We’re Ruta and Emma, the marketing consultants behind Blame it on Marketing. 

If you’re in B2B SaaS or professional services and looking to do marketing that actually drives revenue and profit, we’re here for it.

Visit blameitonmarketing.com and let’s get this show on the road.

Transcript

0:00

welcome to another episode of Blame it on Marketing. It's just Emma and I today, we are going to be talking about AI. keep it not basic bitch when a marketer says to me like, I use AI for marketing. What I assume they mean is that they log in to chat, and then they just prompt and, use the interface there's a whole world called prompt engineering This is not about cheating or shortcutting or anything else. cause the next gen of marketing jobs are going to look like you're on your own. Hi everyone and welcome to another episode of Blame it on Marketing. It's just Emma and I today, or should I say it's the best episode of the season because it's Emma and I today. And we are going to be talking about AI. We're going to try and keep it not basic bitch because we know that there's a lot of stuff out there and we kind of want to go through what we think you should know. But in the spirit of all normal episodes, let's do some marketing confessions. I'm trying to think of something like very recent, like recent that I have messed up. I think maybe it's not a mess up, but maybe it's an attitude issue. I obviously, like most lots of marketers, my personality and doing marketing as a job is very closely connected for me. And so I will admit over the last few weeks, I've taken some marketing things maybe more personally than I should have done. Yeah, but I think taking things too personally, and I found it really hard. I could genuinely have been really stressed out and yeah, I've sort of realised over the last week or so that, you know, it's not life or death. And even though the stress is very real, like super, super real, it's probably not worth it. So yeah, that's my confession. It's like just trying not to take it all too personally. What about you? good. Very emotionally reflective. like it. My confession is I always think I'm going to finish things faster than I do. And I finish things fast. we are, you know, like we work fast, but sometimes I can underestimate how hard things are. So I've been working on this benchmark AI, like task benchmark for like three, four weeks now. I thought it'd be done within a week. And like every time I hit go, there's like a new error and I'm like, and I think, yeah, and it annoys me to no end and I get angry about it. it's like, if I, maybe if I accounted for that in the planning process and in the beginning, it wouldn't piss me off so much. So I think that'll be my marketing confession. I think I finish work way faster than I do, I think it's, I think, yeah, I think maybe planning in that Q &A time is, is, is like Q&A? QA? Come on, Emma. Yeah. Yeah. Yeah. But we all, we also assume, I think because we are more tech savvy, like in the tech savvy bracket, that we can just do this shit. It's like the audacity of thinking we can just do stuff like that. I'm in the same camp, honey. listen, the audacity, also we get it done and we actually get it done. know, like it's the audacity, it might be harder than it is, but like we'll get it done in the end. It's all good. It's all good, yeah. It's all right if it sometimes takes a little bit longer. So I want to talk about, because this is something that we were actually WhatsApp messaging just before we got on the call, which was to do with like when marketers say they're using AI, what do they, what do we think they actually mean? And so Ruta said to me, what did you, I can't remember what you said. You said something about models. Oh, yeah, I was giving you a title and then you changed the title and I was like, that just sounds like models. And you were like, I don't think marketers would even know that. And I was like, no. Yeah, yeah, of course. We're talking majorities here, not individuals, yeah. think, like with all tech, there is that like adoption learning curve thing that goes on that is like, okay, we've got AI in our lives. We've only had it for a very short period of time in the grand scheme of our marketing careers, that is. And so yeah, I think some of the terminology. like with everything is like, okay, we're talking about models, there's platforms, there's, you know, like all of this extra terminology now, LLMs, people may not know what that means. So I think, yeah, I think maybe that's where we should start, Ruta. Why don't you tell us, give us a little upskilling in the AI lingo. Go on. Jesus Christ. Model or LLM, which is large language model is like the thing that you drop down and select. Like let's say if you're using ChatGPT, you pick 40Mini. So that's a model. So that's, if you could imagine it as like a language model, basically how it's going to understand you, how it's going to output and each model is built. to do something specific or achieve something specific. So it might be speed, it might be being good at math, it might be being good at thinking. Then there's providers. So ChatGPT is a platform of OpenAI who is the provider. So there's OpenAI, Google, DeepSeek, Perplexity, they're all providers of their own platform. So... You can use Chat GPT if you want open AI models, but you can also hit up the open AI models in other ways without using the interface of chat GPT So I think that's the difference between like interfaces, like platforms and providers. And then I think one that maybe is worth talking about is prompts. Mmm. So there's a whole world called prompt engineering and we'll get into it a little bit later. But when a, when a marketer says to me like, I use AI for marketing. What I assume they mean is that they log in to chat, they pick a model and then they just prompt and, use the interface to use it and to, you know, have a back and forth and get something out of it. So how you do those prompts, there's like a whole different. world of it. And then there's like loads and loads of ways to do prompts. Most marketers will be using what's called zero shot. So it's you say what you want and you expect it to give it back to you. I think those are the main ones. benchmarks. I did a little LinkedIn post a while back being like, Hey, do you know what a An AI benchmark as a marketer. Most literally 90 % of people were like, what so AI benchmarks are AI tests that try to determine the quality of that model, that LLM that has been built. So you can go and look at them. And if you want like a particularly mathsy one or a particularly thinky one, there's different benchmarks, different tests that people run to basically figure out which one's the best. think those are all the terms. There's automation, but that's like a separate thing we can get into. Any other terms that you can think of? No, I think those are the core ones that are like batted around in kind of general LinkedIn sphere conversations. And I think it is important for people to understand those. I like to think I'm kind of like, I'm definitely not in the expert AI user camp, but I'm definitely, I'm encroaching on the experts now because I'm, In fact, actually using AI to upskill myself in AI, which I think is a point we'll get to. oh Yeah, we'll get to that and how, you we think maybe marketers could use it better, but I think it is important to understand those very basic terms and the terminology Would you agree that most marketers, if they say they're using AI for marketing are in that camp of they go on a provider, go on a platform and have a chat with it kind of thing. Cool. um that, and I also didn't know about the world of prompt engineering, but I can imagine that that is a thing. I can imagine there is a whole science between actually being able to get the thing you need out of it. a lot of marketers, and I put myself in this camp, kind of just doing it naturally. you know, so yeah, think most of us are in that camp. I can share in the description. there's a prompt guide.ai website and they go through like all the different types of prompts and how, how and why you might want to use them. I guess we can get onto prompt engineering, right? So what is it and why do we need to learn it? I mean, now I fully kind of understand that, like I say, there's a whole science behind it. I think the reason you want to do it is because, you know, I mean, the basic principle is you put shit in, you get shit out. It's like, it's like the marketing mantra 101 of everything that we ever do. So if you, if you want generic content or you want super generic stuff, it's fine to log into ChatGPT and do that and just do your quick prompt. But if you want anything better than that. which we would hope that most do. I think we both find it funny when you know, like people on LinkedIn are like, I know you're using AI because of all the dashes. And it's like, you know, you're not like one bullshit and two, like you can engineer your prompts not to have those things. Like if you find that when you're asking for copy, especially copy, it's always kind of off. It's probably in your prompt. Like your prompt isn't right. Yeah. Yeah. Yeah. Yeah. And I, I do think of myself as, like I said, not expert AI person in marketing, but I definitely am moving myself up in, in the, in the gears. but I, one of the things I do definitely pride myself on is my ability to prompt. So I, I always think of this from the point of view of like, okay, not, it's like, it's kind of like the core basics of, of marketing just in general. Like what's the purpose of the thing I'm trying to do? So be clear on that. Who are the, who's the audience, the ICP, you know, the subset of, your ICP that you're trying to target with the thing that you want to do. If, if you're using it, I guess, to support content, this is there are other uses of AI for boxers, but we'll get onto that, primarily content. and then yeah, like what are the key themes and you've got to feed it information, like the assumption that you can just be like, write me an article about how to create a great marketing campaign. and that it will do that and that that's okay is meh. If you can, if you go and you have sources and you have done some of the research and you give it that, you are adding that extra layer of depth in that will give you some of the detail that you need. you, but that's not to say that you can't use chat to find those sources, but then you look at the sources and say to the chat or whatever it is you're using, this is a good source. For example, we worked with a client recently and we know that they need really good, strong, solid sources. So we've gone in and we've said to the prompt, like, make sure you are quoting people like McKinsey and Gallup and, you know, those level of kind of research. So that level of research so that you know that it has that depth of knowledge in whatever it produces from a content point of view. Yeah. And I think as well, if you're sitting here and you're like, well, how do I know what do I need to give it? There's something called meta prompting. So you can actually open up, yeah, you can open up a chat and be like, Hey, I'm trying to write a prompt for an LLM that I needed to do X, Y, and Z. What do I need to include? Help me write this prompt. And then basically the end product of that is the prompt you're going to give to the LLM to then make, do your work. And I did a bunch of this for the benchmark because the benchmark is basically marketing tasks. And I was like, there is no way I'm handwriting out like all of these tasks prompts with like tone of voice and like length and technical like specifications. So I literally just did meta prompting and I was like, this is great. I'm never going to like off by heart write a long prompt because there's just, you're not going to get it right. And also if you tell the LLM it's for an LLM, it will like write it in a way that the LLM will basically digest the information. So it's, yeah, it's one of my favorite tricks. We've seen quite a few like marketing type influencer people on LinkedIn being like, hey, here's my prompt library. And I was like, like me and Emma were giggling. I was like, I just get chat to write my prompts. Like I don't need a prompt library to have stuff like you, just don't need that. And if you're relying on those, it's almost like. you're never going to learn how to write a good prompt and get what you need out of it. If you need other people's prompts and also they're never going to be as specific as what you need anyway. Yeah. The other part to this, which is like, I think it's, this is a usage point around prompts and this prompt engineering stuff is you don't need to open a new chat every single time. Like I, I think that again, is one of the things that sort of throws you under the bus. If you are producing, especially like I say, mainly if you're using it to produce content, is because basically what it's like, otherwise it's not learning. The whole point of the AI is that it's constantly learning and it's picking up. the way you want it to do stuff, the way you want things phrased or whatever positioned. And so if every single time you open up a new chat, all that knowledge is gone. um context is gone. So, I mean, there's two ways to handle this, right? And I am big fan of one of them. One way is that you, set up projects, you know, and then you can keep track of your chats. I mean, you should probably do that anyway. That's just good hygiene, good practice. And the second thing is obviously. the and if this you do have to have paid accounts to do this sort of thing which you should anyway we'll we can get onto that but to set up custom gpt's so and you can do that for many different things you can set up custom gpt's for like i don't know if you're a freelancer you can obviously have them set for clients if you have or if you have like specific pieces of marketing activity like email marketing social media posts, those sorts of You can set up separate custom GPTs for each one of those things. Feed it all the knowledge. Get your meta, chat up to write your prompts. Pull that together, save it. And then every time you kind of go into that chat, you're using it as that like machine, to kind of get the best quality out of it. so yeah. Yes again if you've never made a custom GPT or it's basically a system prompt for your chat. It's given normally when you click new chat, the only context it has is what you put in the message. When you give it an additional system prompt in the background, it knows what its job is. its job is it's a content writer for your company and this is the experience it has and this is the things that it wants to talk about. So it gives it all of this additional context. And again, with like the style of writing especially, M dashes, all of that stuff, you can shove into your system prompts, into your custom chat GPT, so you never have to deal with it spitting out content that you don't like. And when I say content, I don't know, it just means stuff you're gonna post, like just responses, basically that don't work for you. Yeah. yeah you're not gonna have to work as hard to get to the thing with all that said it is marketing content there should always be an edit a human edit always always always always so even if that is what you do there always has to be that check what do you think is some of the things that AI is particularly good at, but maybe marketers are still doing manually now? Ooh, I've got quite a lot of things actually. one of them, so I think, and this is the stuff that I think a lot of marketers don't love and that's the process side of marketing. So like, for example, creating processes around how to use your CRM. You can obviously get AI to write a lot of that stuff for you. can obviously, like, let's say you're using a HubSpot. Obviously HubSpot have a lot of that stuff on their website anyway, so it's easy for the AI to pull in what is maybe best practice usage. But then you can obviously create, you know, flows and things like that in your HubSpot. And then you can ask, you can ask chat to kind of document for you, give you thoughts on how to, you know, best use it, give, you know, then give sales like a breakdown of like a really nice light touch version of how to do it. Another thing that people are doing a lot manually are like, say for example, like cataloging content that you've got and quotes and things that you've got on your website, things that you've used. So you can obviously use, you know, your, your LLM of choice to catalog things like, can you create me a spreadsheet of all of the quotes on our website that we've used, the client's name. And they're, you know, which company they're from. it's, it's like done that work for you because, know, it's the kind of stuff that sales ask for all the time. Or can you actually do like a content audit and can you put the content into like two buckets for me, you know, like this theme and this thing and ask it to kind of like catalog everything for you. Cause that is again, that's a big manual job that people do, but it's something that marketing people are asked for like time and time again. and not maybe something that you actually, know, loads of people go, I can't do that because it's gonna take so long. I actually used it for one of those the other day, like content audit on someone's site and it was fantastic. Done within 20 minutes, guys. chef's kiss audits by hand. Nope. Yeah, I totally agree with this. Go, go. on those things is also finding gaps. finding gaps that you haven't thought about yourself. so like let's say for example your company has like three or four key themes or pillars of content that you want to create based on you know whatever your icp like and you're like i'm not sure like how many pieces of content i've got that fit in these buckets. i mean so quick and easy guys. Same as like, I think that goes for like, let's say you've planned a campaign, you can put your campaign plan in and be like, what am I, am I missing something? Like, you know, especially given if you've got a long history chat, when it has all the context, it's like, what am I missing? Like, tell me what I could add, what I should take away, you know, doing like almost like the consulting, checking bit. Yeah. Or like, I've come up with a campaign plan. The purpose of the campaign is to... it could be lead gen, it could be whatever. And I want to make sure that this actually is gonna hopefully deliver what I think it's gonna deliver. Do you think it will work? It's quite useful for being like, actually, I don't think you've got enough content in here that's gonna give you the lead gen or that's gonna create... or this is a lead gen or a brand awareness campaign and it... you want to do demand and here's a gap, like it's very good, it's very helpful. And that's not to say that it's taking all the thinking away from the marketer, but it's like, it's the equivalent of if you haven't got a senior marketer, it's your sense check. Yeah, yeah, absolutely. I think a lot of marketers are still doing like chat to chat. And I don't think a lot of marketers are in like the automation phase of like combining AI with there's Zapier workflows. So I would say there's a ton of stuff there. For example, One of my favorite things that I set up was a workflow in Zapier that took our transcripts of our podcasts and then turned them into blogs and then put them onto HubSpot for me to review and publish. There was like 80 plus podcast episodes. That would have taken me three years to do manually or like to copy and paste out of chat, even in that sense, but you could just hit go and it did that for you. So I think it's like connecting automations with AI. Mm. is still in its early infancy days. And I think like Zapier is probably one of the most easy and accessible ways to start because they've got some really great inbuilt tools. The other thing that I use AI for a lot is for data cleaning. So you can give it a spreadsheet and be like, hey, all of these numbers, like phone numbers are not formatted in the right way. Can you go through and format them? So I really like to use it for data cleaning and data validation because that takes a lot of manual work off my hands. I think if you've got a lot of data as well, and I think this is one of the things that a lot of people are scared of doing, right? Which is taking your spreadsheet of doom, shoving it into something, and then being like, what does this all mean? But actually that can be quite helpful because again, it's like, if you are, let's say you're a mid-level marketer you're being asked to produce board reports, and this is your first crack at understanding data, it's a way that's going to help you get confident with data. And honestly, I, my biggest thing about all of this now is that this is a way for you to learn. This is not about cheating or shortcutting or anything else. This is a way for you to learn. And to upskill yourself for the next, for the neck, what the next gen of marketing jobs look like. Cause then, cause the next gen of marketing jobs are going to look like you're on your own. You are on your own. We are now not paying for anyone else and you're going to be, have to be able to do all this shit on your own. that's my kind of like doomsday prediction about all this. So we did an episode with Tim. And one of the things that was my huge takeaway is like, instead of now starting work and being like, OK, got to go through my spreadsheets. It's like, can I build something with AI or can I use AI to get this work done? And when you start processing those things and you might not be able to do in the moment, you might not have time, but if it's on your list of things you could build and you start slowly building them out, you can take so much actual work away from yourself and then have time to do fun stuff or not do stuff at all. That's the, yeah, you sort of said that to me a few times. It's like that not starting from a blank page and going, Oh God, where do I go now? It's the same as like, I was actually thinking about this before we got on the call, which was to do with, should you use AI to write marketing strategies? And actually, Yeah. Yeah. right. And I'm not trying to do us out of a job as, you know, fractional CMOs. But the reason I think yes is that it's better to have one and to then validate it than to have nothing and to, and to like be flying in the dark, you know, so. the hardest bit of anything is starting from scratch, right? Like getting those first few lines down. if you're in an editing job and you know you've inputted the right information and you did all that, you're gonna, the product at the end is not only gonna have taken you less time, it's gonna be a better product in the end. And that's what it's about, isn't it? like you can find great templates out there. I'm going to say this because I wrote one. So you can, you know, you can use my marketing strategy template. can use HubSpots, you can use whoever, and then you can input that into your AI. can say, these are the business goals for this year. This is the sort of stuff we want to achieve. This is the data that we collected last year or over the last few years that shows us that what we're doing is working. This is the stuff that doesn't work. You know, help me. bring all this together and you've got a starting point. my god. so strategies don't need to take fucking four weeks six months to write anymore, you know? Once you have the goals and you have some data, that should be a day job. Like, as in one day's job. um Not a day job for someone. the time that you are, I think this is maybe the thing that scares marketers is it's like, well, then what is my job? Right? So if I'm not doing this bit anymore, what's my job? Well, your job is now validating it, building relationships with people internally and still doing the creative thinking and the creative part of the job and still collecting the data and auditing the data and understanding it. So. but the heavy lifting in the middle, the bullshit bit of the job that is, you know, I don't want to write content management processes. No, yeah, yeah. And you can, yeah, it's just expanding your capacity basically. Like that's what it's doing and it's it's probably outputting work that is more likely to be successful, meaning that you're gonna look better because you're gonna perform better. I think that's the other thing. People are looking for perfection. they want the AI to spit out exactly what's in their head and exactly how they wanted it and exactly with the right words. No, you're gonna have to edit, but that process is so much faster. like, it's not it's spitting out perfection, it spits out a version for you to perfect. And like, you would have never got there anyway on your own. it's billions and billions of neural connections that data scientists have fed in billions and billions of pieces of data to train. You cannot outthink it. Like you might be able to like in a creative way, but that's like creativity for the sense of creativity, you know, like that's not strategic. That's not data-led. Yeah. Yeah. And I think that's it. There's a difference because like in marketing, there's a time for art and there's a time for science and it's science. Yeah, exactly. Exactly that. And I think a lot of people may not like that, but it is the truth. the other side to all this, so I want to talk about the models in a minute, but there is also a side from the, they're like how you use it in your day to day. You and I are big fans of using. AI, as like coaches, coaches slash therapists. So I dunno, like maybe give us an example, Ruta, of how you would use that, like in a marketing context, not just, that your boyfriend's annoyed you or vice versa, you know? Yeah, I do that too, but you're never going to find someone that is like more empathetic than your AI, by the way. So it's a good, it's a good thing. what's coming to mind is I think marketing teams are going to be shrinking and marketers are going to be coming more junior because no one wants to pay for senior people full time. So. And I've been in this point in my career too. You're kind of the most senior marketer in your company, but you're not that senior and there's no one else to like actually coach you or deal with you. I would absolutely be setting up custom GPTs, with the prompts of like, you're a, this type of, you can even like use a specific person with their like achievements if you want like your... I don't know, blah, blah from company, blah, blah. You're the CMO and you are mentoring people now. Expect mentee questions to be coming through and I would use it to validate things. So like if I don't know something, if I need like a backboard, if I'm finding it particularly challenging to get something signed off, for example, because it's like a people management skill, all of those things you can use chat to coach you on and to give you actions on. without any kind of judgment, without having to like, you know, open your cards up potentially, maybe to somebody that's not got your best intentions in mind at work. And then the other thing that I like it for is for specific niche marketing subjects. So if you know, you don't know shit about SEO, you can get chat to teach you everything you need to know about SEO and you can get it to put together curriculum for you if you want to and take you through it. and put like diary meetings in for when you're like, you know, learning with chat about SEO, et cetera. So like for different marketing niches, I would also have like probably a separate custom GPT if I wanted to. If it's like an area I really want to work on to basically coach me and teach me how something works. So like that's in the marketing world, like the people management and the like backboarding and then the specific hard skill learning is like the coaching types I would use. What about you? Yeah, no, absolutely those things. I have, like I was saying at the start, I've used AI, I've used my custom CMO GPT that I have set up to coach me to be a better CMO. I'm like, you're a CMO, I'm a CMO. oh Yeah. What's that? What's some of the things that I should think about? Like even things that I think I'm quite good at, like I'm actually, I do think I'm quite good at handling difficult relationships because I kind of know where the boundaries are with people and I feel like I can handle someone if they're a little bit tough, you know? But even I sometimes just need to check that I haven't, we joke, because I'm slithering, you know, that I haven't been too slithering. Mm. or that if I'm going to send an email, that it's not too slither in, that it's still like, I'm not going to be offensive or pissing anyone off by, you know, so yeah, just like, yeah, setting myself up for failure by, by saying something or doing something or, you know. and I do think sometimes with those difficult conversations, like, you know, it's good to prep and it's good to have like notes and those sorts of things. And I think it's really useful for, for those sorts of moments that are challenging. And then even the other thing that I have started doing, which I really love is putting contracts in and asking it to summarize anything that it thinks is problematic in the contract. So like, good with vendors. Am I signing myself up for something that is gonna screw me later? Yep. yeah, just another little extra thing on top. tell, infinite ways to use it. We love using it. You should too. Um, equally, think, you know, I would love to see more, I can see more people reaching out to people like us and being like, we've got a one person marketing team. Can you help them get the AI stack set up? And, and that kind of thing. And I'm like, yeah, no, that's good. That's something that I think is valuable, um, to do, even if it's like, well, then have I done myself out of a job as a consultant? No. Our job is always to make the marketer look good. Yeah. And it's what's needed, right? Like, and it's where it's going. We're not, or maybe it's just Emma and I, like, we are under the assumption that our work will change. Like we're not under the assumption that we will be what we are. And yeah, like forever work as fractional CMOs and do the strategy bit and you know, whatever. Like it might not be that it's going to, it's not, it might not be that it won't be that it's going to change. And like, yeah. Yeah. is changing you can't stop people using chat like it's just not gonna happen. But our job is probably gonna become checking whether the stuff that came out is all right, supporting people to do the chats properly. That's gonna be the world of work for our, yeah, yeah, yeah. And I'm okay with that. So question for you, model queen, how much do you think it actually matters which model you use? Yeah. So again, models are the things that you drop down and choose like in chat GPT might be 4.1 or 04 or whatever. So up until when I've done the benchmark and I'm running like the benchmark is marketing tasks and then it gets judged by separate other LLMs to say how well it did that task. I was under the impression that it didn't matter that much for marketing work, but So like, let's take a look at copywriting for example. Mathsy LLMs like for example, OpenAI 4.1, do much, much worse at writing than for example, what you would consider a fairly small and cheap LLM like O4 Mini. Hmm? scored like 96 out of 100 on copywriting, whilst 4.1 scored 88 out of 100. So I think it does matter. think it also matters in the terms of limits and costs. So don't use expensive models and models that maybe have like usage limitations for simple shit. Like I know that's self-explanatory, but like use cheap models to do like the preload thinking and then give the expensive models your huge complicated prompts that they need to go away and figure out and reason with. of. one of my tests, which I haven't run yet, but it's going to be hilarious, like, it's basically like a statistical model based on your ROI, like how all of the channels integrate with each other. And like, if you pulled one, what would it mean for all the others? And I know some of the models are going to just crap out. Like, I know there's going to be models that just also respond with just anything and be like, yeah, yeah, that's it. And I've got like other models checking their answers. So I'm really excited for that one. gonna be really interesting. I love that. think, again, like think of the power of that. Think of the amount of times you've had conversations with senior people and they're like, we're just gonna stop doing this thing now because I don't think it works. But to be able to show the impact of pulling it and not, you not have to try and navigate that in your head, but you know that's a problem. mean, that's powerful shit, man. Really powerful. Yep. I was gonna ask you. Oh, I know what I was gonna ask you, which was to do with reasoning. So like, I would really love it. Like obviously, you know, for you to give us a little like, like tip bit on why, know, when would you use reasoning in marketing? I mean, I've got what I would do, but you give us an example of what you would do. So reasoning is basically thinking. like, instead of you putting in the prompt and the AI just basically spitting out what it thinks is best, it has an intermediate bit where it tries to understand what you actually asked for. And I know it sounds like it should do that anyway, but it doesn't. They're just statistical models of what's most likely to be correct. So when it does the reasoning bit, the things that I use reasoning for are things that are not yes or no answers, basically. So like, if you are asking it to take in a bunch of data and to come up with insights, use reasoning. If you're asking it to do a multi-step task of some sort, and it needs to like do this and then go away and do this and then bring it back, use reasoning because it needs to actually understand the stages instead of just basically what NA.I does is it reads your prompts top to bottom and it just starts spitting out tokens to to respond to whatever was top to bottom and it won't understand that there is like a staging thing in the middle. So I would use reasoning for that. What I wouldn't use reasoning for is like something like, take this blog post and turn it into social posts for me. Reasoning is not gonna do anything unless you're like creating something new at the end that isn't within the blog posts. So I wouldn't use reasoning for that. like if you're, I don't know, using it to like spell check or check like a technical aspect of it. You typically won't need reasoning. Reasoning is for like multi-step complex tasks that it needs to actually understand the context of, ah which is a hard thing to explain when it feels like chat understands what you want from it, but it's just, it's calculating the next like letter to spit out basically with everything it reads instead of like understanding the prompt. Well, it's like a conversation. It's more of a conversation like you would have with a person or like if someone's briefing you, it's like that. It's like that ability to brief and then listen to what you've said and try and understand it back. And I think that, yeah, yeah. Cause like, so for me, reasoning again, like you're going to use chat, like a chat, whatever, to do your strategy and marketing strategy. That is the kind of task we want to use reasoning with because you want it to be like, have I understood this correctly? Do you want that? Is it this? Is it that? Is it that? Because you otherwise you do miss all that nuance. Otherwise, otherwise it will just give you the... And it can kind of ignore the middle bits sometimes if you don't use reasoning. And it's like, if it's like a super chunky prompt where you've added a bunch of data in, use reasoning, because it will actually read through the middle instead of just like statistically skipping to the answer because it thinks that that's the best answer. Yeah. It's hard to explain. Like it doesn't think like it's just maths and like, just predicts the next best token to give you. Like it doesn't actually think even though it feels like it thinks. Yeah, but yeah, you could, you can, I guess you can easily treat it like it does, can't you? So we're coming to the end of the AI episode. Obviously Ruta is, we're going to do our gossip, but Ruta is developing an AI benchmark tool. So when that is available, we will obviously make it available to everybody. To play with. going out on Thursday, which is tomorrow. It should be up next week unless something dire happens. So we'll post about it. Yeah. It's basically going to be like a, what you should be using, what models you should be using and what's good at what. And also I'm going to like have all my prompts open. So all of those crazy like regression prompts for like what means what you can go away and play with them with your own data, which is the fun bit, I think. Yeah. watch this space for that. Before we go, any marketing gossip? Any marketing gossip? There's always marketing gossip. I feel like we live in marketing gossip. Be mindful of what you comment on LinkedIn is what I'm going to say. Because for the most part, comments kind of get prioritized in the feed. So if I know someone, I'm connected on someone on LinkedIn and they comment on something, it prioritizes it and I can see what they commented. And sometimes it can be very showing of where you're at in terms of your skills and your marketing thoughts and all of that stuff. just be mindful of that is my marketing gossip. I know what you mean. I know what she's getting at. Yeah, I think I agree. Because I do think, yeah, especially if you are, I don't know, maybe it's more if you're consultant or freelancer agency side and you're commenting on things that are a bit like, basically, like not admitting you need help, but yeah, like Yeah. it, that's publicly visible, your clients can see it. So there is an element of just thinking that through. And I think mine is, mine is also LinkedIn based, but mine is, is just, I'm loving the like, the comedy on LinkedIn of people just not using it like a, a professional network and just being like, here's a meme. And everyone's like, ha ha ha ha ha. And those people are getting loads of engagement without engagement pods. And so I feel like, I mean, I've said it from the get go that I think it's the, the, the people are treating it bit more like Twitter. But yeah, again, obviously it is still public. People can see it's you. Be a bit careful. But I do love, I do love the funny stuff. I'm just, I love it. I say be my full comments, my comments are so deranged that you don't even understand what it means unless you know what it means. time people don't know that you're being, we're being sarcastic. uh yeah, but yeah, it's something to think about. Yeah, well, this has been really fun. We're hoping that we gave you like both good information but practical stuff that you can take away. And if something's weird or unclear or you want to see like the stuff we do, just DM us and we'll chat. We love chatting about the show. So yeah, thank you, thank you Ems and we'll see you in the next season.