AI for Better Ad Creative: 3 Steps to Better Results

Before we get started, I want to ask you something. Are you tired of feeling like AI is moving faster than you can keep up? Because I know what that feels like. And I hear it all the time. Marketers keep telling me they're drowning in all the new tools and updates and features and must-try frameworks. And, and they got to do all this while they're also doing their job and trying to run their business. And that's why the AI Business Society was created. It's your personal curation team for AI. We tell you what deserves your attention and what you can safely ignore. And inside you get live trainings from leading AI practitioners and interactive skill building sessions, bite-sized learning libraries when you just have a few minutes, and a community of incredible marketers who are just like you. Hundreds of them are on the inside, and you can join right now and lock in an incredible special sale we've got going on. Visit socialmediaexaminer.com/ai. Give it a shot. You got a 30-day money-back guarantee, so you've got nothing to lose. Go to socialmediaexaminer.com/ai and let's now jump back into today's show. Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your host, Michael Stelzner. Hello.

Hello.

Hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzer, and this is the podcast for marketers, creators, and business owners who want to know how to put AI to work. Partner with AI on your ad creative might be one of the best decisions you can make to improve your results. My special guest is an ad creative expert who leverages AI to improve his craft. He's the CEO of Fraggle, an agency that helps direct-to-consumer brands acquire customers with diverse ad creative across Meta and TikTok. His course is the Ad Creative Course. His newsletter is Nice Ads. Fraser Cottrell, welcome to the show. How are you doing today?

I'm good, thank you. Yeah, what about yourself?

I'm doing great. Today, Fraser and I are going to— I'm getting the Fraser and the Fraggle kind of thing mixed up in my head.

Yeah, I know. A lot of people do. I often— people end up calling the company just Fraser because it's just easier. It often happens.

Well, today we're going to explore ad creative and artificial intelligence. Intelligence. So let me start a little bit with your story. How did you get into AI?

So I think it kind of all started when, obviously, because I do creative, right, and everyone's always talking about how AI is going to steal, you know, creative people's jobs. And so, uh, I, I started using AI. ChatGPT was maybe like the first tool I kind of like properly used, and I think I was probably using it And I still kind of do actually use ChatGPT, even though I've like fully gone over to Claude, which I'm sure we'll talk about later. But like, I still use ChatGPT for search. I feel like that's it. It's like one of the best apps to just open up and just Google something, essentially. So I think that was probably my first touch point of AI. But I think maybe like 2 years ago or so, I really started to like use AI within the business to make my job as a, as a founder and a CEO a little bit easier. But then probably about like summer last year, I moved over to Claude, and that was when I really started to like quite heavily use AI in both my personal workflow, but then also like the agency's workflow as well. Yeah, so started with ChatGPT, eased myself in, and now, you know, a lot of parts of our businesses run with some kind of AI in there.

Well, talk to me a little bit about the ad creative side of it too, like the journey on that front as well.

Yeah, so obviously there's a couple of different ad creative types that we make. One of them is, uh, is video. We're a very video-heavy, uh, agency, but also images and statics. In my opinion, images, static ads are some of the easiest creative to make with AI right now because of how great image gen models are. Uh, you know, like as we're talking, what, about 2 days ago, ChatGPT just released their new image model, right? Which, which again is, yeah, kind of blowing people's minds a little bit. Yeah, so specifically when it comes to our creative, we've started to implement it a lot, whether it's just generating like the odd shot for a video that we filmed in our studio, all the way to, you know, static ads, which basically the entire image is AI generated and then we're just adding on text and stuff like that on top. But even when it comes to like the video stuff, often enough, like scripts, research, things like that, we can go really deep with, with AI with a lot of that.

Love it.

Okay.

We're going to dig pretty deep today, obviously, into using AI to ultimately create some really cool stuff for your ads. But I'm just curious because I know you've been doing probably a lot more with AI than you're letting on to, and you probably talk to plenty of your peers and colleagues that are doing what you're doing. And what are some of the misconceptions you see when it comes to AI?

So I think kind of the, the main misconceptions that a lot of people, not just colleagues and friends, but like everyone in general, and the first one is that it's lazy, right? You're letting, you know, you're letting AI do your job for you. So you're lazy for doing that. You're just outsourcing the work to a robot, right? I think that's probably the first one that most people think. I think the next one that goes hand in hand with creative in general is that it's not high quality. The only thing that it can make is AI slop. Because I think people instantly think of all those like vegetable videos that are going like viral on TikTok, right? They all, they saw when like Sora 2 launched and all like the crazy videos that people were making, which were kind of, you know, they weren't great, were they? Like, you know, they kind of existed for the sense of laughing at AI. And so that, is a massive misconception. So I'd probably say, yeah, lazy and not very high quality is like the main ones that most people think.

So dispel the misconception a little bit on both of those, if you don't mind.

So I mean, it's definitely not lazy. You and I both know the amount of work you have to put in to make AI do the thing you want it to do. AI is only as good as the context and the instructions that you give it. AI is not at the stage where it knows everything. It can give it a go, but the chances are the output isn't going to be what you want. And so most people at that stage would write off AI and they say it's no good, it doesn't work, it doesn't do what I asked it to do. It's normally because you haven't asked it properly or it doesn't have any information. All right, so that's the first thing. So it's certainly not lazy because you do have to put a lot of work in. And in my opinion, what it allows you to do is get, get to where you need to be faster so you can, you know, perhaps do all the stuff at the same time, or just be able to build better, I think, as well. Um, in terms of it not being high quality, I mean, you know, like we said, the new ChatGPT, you know, uh, image model and stuff, the stuff that's coming out of this is unreal. I think video still has a way to go, and we'll talk about that, but I think in terms of images, like one still frame, it's becoming worrying how difficult it is going to be to distinguish between AI and real content. As someone that, you know, spent a lot of my teenage life taking photos, it is worrying that, that, you know, area of media is becoming, you know, very AI heavy, because this stuff looks really good.

And for folks that don't know what Fraser's talking about, ChatGPT Images 2.0 is what they're calling it. And as of this recording in April 2026, it just rolled out a couple of days ago. And everything you just said is what we've been what you're saying about Nanobanana also, right?

I mean, I'd say that Nanobanana is still one of the best image models. Like, I, I haven't played around with the ChatGPT one enough yet. Yeah, from what I'm seeing is a lot of people saying that it's like ChatGPT has finally caught up with Nano. Yeah, um, because my go-to is still Nano. It's really fast, it's great with prompts, you know, you, you can replicate a lot of stuff very well. I think the issue with like ChatGPT before was that it kind of gave it a go, but it was never usable. Yeah. Whereas now I think finally is.

Yeah. And these things are just going to incrementally get better and yeah, we'll get into that a little bit later. But yes, we are in a stage right now where images that are generated by these various different models are so good that they're indiscernible from an actual professional photograph, which is really powerful. And that's the upside. And maybe we're getting a little bit ahead of ourselves, but when AI and ad creative are brought together in the way that we're about to talk about today? What are the benefits? What is the upside that you're seeing that others will see if they, if they pay attention today?

Yeah, I think the main upside, right, is that it levels the playing field to some extent. I think beforehand, if you wanted— if we talk about, you know, e-com, which is what I do, if we talk about that, if you wanted really good product images, you basically had to have the money to do it, you know, whether it was to hire a freelancer or to hire a studio, whatever. You had to have the skill or the money to go and, to go and, you know, to, to go and take those photos. And so a lot of e-commerce brands were running with fairly low quality images for a long time because that's all that they could do, or they blow a huge amount of budget on it. Whereas now you can get product images which are very usable for a couple of cents. And, you know, that is a huge advantage and a massive way to level the playing field. Obviously, there are negatives to this as well, right? You know, the, the— that now the dropshippers can really generate anything they want and try and sell anything, you know, potentially something that, um, isn't real. But I think when it's done great, it levels the playing field, but also it allows you to reach volume levels that you weren't able to reach before, which is a massive thing, particularly within the ad space. Now more and more people are chasing volume. And in order to get those volume levels without AI, it's going to cost you a lot of money to hire a lot of different agencies or hire a massive team. But with AI, you can get to certain volume levels much faster because it frees up bandwidth massively for whether it's writing or it's, you know, the actual creation, whatever that looks like. It frees up the creator's bandwidth a lot because they can focus on other parts.

Well, and I've had a number of people across both of my shows specifically that are ads experts, and they have said with a lot of the updates, uh, coming from Meta in particular, I think it's the Andromeda update, they, they want lots of creative massively. Yeah. And that was out of the realms of possibility for a lot of advertisers, right? Because they just couldn't do that. I mean, this opens up a huge opportunity there, does it not?

Yeah, yeah. The Andromeda update kind of basically stopped— it's there to stop advertisers being lazy. People got into a really bad habit of almost gaming the system by creating a million different variations of basically the same creative with little bits changed so that Meta would treat it as the same ad. That doesn't exist anymore. And so instead of making 100 different variations, you need to make 100 different ads. Like you said, becomes impossible for people at a certain scale to do that. AI allows you to reach that scale at much lower budgets. And also like Meta in general are pushing AI in their own AI models. On advertisers more than ever as well, right? So they want you to use it whether you want to or not.

Yeah. Well, and by the way, I mean, I believe in split testing and I think it's really cool. And you could probably very easily now— AI can help find some real winners, right? And that could be very powerful, which means you could ultimately get more engagement with those ads and ultimately reduce your costs and all those things. So there's lots of benefits. Obviously, we're going to get into some of this right now. But before anyone considers using AI in their ad creative, is there any kind of fundamental things we need to address or things people need to be thinking about before they begin, you know, what we're about to talk about today?

I think, like I said at the beginning, AI is only as good as the context that you give it. And so how, how I tell all my team to use AI is to think of it as an intern. It is someone that is coming to your business or your life and knows very little, you know, maybe has like a basic understanding of, of your service, you know, a very basic understanding of what you do, but you need to train it to get to the level in which you probably want it to operate at. So everything that you do, you almost need to over-explain it. You need to give it a lot of context. I wrote an X article on using Obsidian, which is like a open source note-taking storage app in order to like build your knowledge base so that it's not just pulling from random fragments of memories. Instead, it's pulling from one core database, just the same way that you would treat an intern, right? You are giving them everything that they need so that they can, so that they can run with it. So you do need to remember that, uh, when, when you're aiming to use AI, you need to make sure that you give it the information to pull from.

Yeah. And I want to double down on this because by the time everyone is hearing this interview, I will have already given my keynote presentation at Social Media Marketing World about specifically AI. And in that presentation, I say that context is the most important thing because the reality is a lot of people think, and it's true, that AI has been trained on all the information of the world. But what it does not know is all the things that you as the human know. It doesn't know the intimates of your product, you know what I mean? It can gather information from the internet, but it doesn't know the things that you know and it doesn't have the insights that you know. So it's going to make broad sweeping generalization. It's kind of like hiring someone and saying, hey, go create me a campaign and that's it. Like no interaction at all. Like you got to give them everything, right? You would train them. Here's who our ideal customer is. Here's what our company is all about. Here's what our customers say. All that kind of stuff. Is no-brainer when we're dealing with a human, but for whatever reason, people forget it when they're interacting with AI.

Yeah, exactly. And I think that's what it is, right? People expect, oh, this is AI, it's just been trained on everything, right? It knows everything. Well, yeah, to an extent. It knows averages and surface knowledge of a lot of different things, but it doesn't know the granular details. And when you have an AI which is really well trained, it kind of blows your mind when it does certain things. Like, I use it quite a lot just to help me run my business. And so it knows all my staff's names, it knows how much I pay them, it knows what their job roles are. And so when it references that, it's like, oh yeah, why don't you ask this member of staff for your help? I'm like, what?

Right, that's a really good idea. I didn't even remember you knew all that, right?

But yeah, like, this is the thing is that you don't realize, and then it kind of blows your mind. You're like, wow, this is actually incredibly powerful. But that only happens when you train it properly, when you give it the information that it needs.

Love it. Okay, so folks, foundational for all AI, doesn't matter whether we're talking about AI and ad creative or any other thing, is context is king, really, in this particular situation. So now we're at the point where we want to start figuring out how to use AI to help us create better ad creative. So what, what's the next thing we need to be focusing on?

Okay, so I think when it comes to our creative, like we said, right, knowledge is— knowledge is key. And so the most important thing, and the first thing that I do when any brand comes to work with us, is what our, um, our creatives do here, is deep research. So AI is incredibly good at extracting information, incredibly good, and also very good at summarizing that information in bite-sized pieces et cetera. And so what I usually do is I normally always run a deep research prompt, which I get another AI to write, um, which essentially is asking the AI to go onto the internet and search for everything about my brand. I want to know who's buying my product, why they're buying my product, If they haven't bought it, why didn't they buy it? What people are saying about it on Reddit? You know, what are some common issues that people have with my product? Where, where does my customer live? All these things in which you might already have that information potentially, but we want the AI to go out and find this information. And we, we use a function within AI tools called deep research. I'm not sure what Gemini's is called, but I know that ChatGPT's is deep research, but Gemini is also deep research.

They're all deep research. Oh, is it? Okay. Yeah.

Yeah, so they all have something similar, which basically means you're telling the AI, take your time, like, go, go and find this stuff. Make sure that what you give me is formatted correctly, it is well researched, and you've done your due diligence on this piece of work. And so what you should get back is a, you know, probably substantial document. It might take like 15 minutes, whatever, but you will get a substantial document about your brand and what it's found. And that is, I'd argue, the basis for your knowledge building. At this point as well, you can kind of feed that back into the AI, you know, and you can maybe give it some information that it's missing to fill out that document, the information that only you know. But that's the key place that we want to start here, is using deep research to build a basis of knowledge about our brand.

Okay, so I know that Claude and ChatGPT and Gemini all offer a deep research capability. First of all, you said you get one system to write it and then you get another system to do it. So just give us the tip on which system you use to write it and then which one you're using to do the actual research, or are you using multiple ones?

Yeah, so in terms of like, people might be a bit confused here that, hang on, you're using another tool to write a prompt for another tool. Yeah, it's a strange way to think about it, but LLMs are very good at writing prompts for themselves because they understand what the AI needs in order to give that information. So whenever I'm writing a prompt for anything, it can be for an image generation tool, it can be for whatever, I normally, you know, use like Whisperflow or something, you know, in order to dictate to my computer. And I will say, go ahead and create me a deep research prompt for, you know, whatever tool, make sure you cover XYZ. And then that will go away and it will format it and rewrite the prompt. In terms of what I use for that, I normally use Claude for most of my like prompt writing, or it's like my go-to LLM at this point. In terms of the deep research, I'm kind of in between. ChatGPT is really good. It is really good for this stuff, but it does tend to hallucinate a little bit, and, you know, it's not— it doesn't always give you the best results. Gemini, my thought process is, well, Gemini's owned by Google. Google basically has all the information that everyone needs ever. So my thought is, I'm guessing that Gemini is probably better at it. So I kind of divvy between the two, really.

Well, in Gemini, it's deeper reports, generally speaking. It's much bigger.

Yeah, yeah. I'm faster as well, like much faster. Um, so yeah, potentially just use Gemini, but then use Claude for the writing. That— and, and again, when you're getting Claude to write that prompt, say, write me a prompt for Google Gemini, a deep research prompt. Like, call out the platform that you want it to write the prompt for.

Got it. Okay, cool. So we're getting this great report as a result of this, and what, what's next? What do we do with this?

So I think the next thing after that is to begin to train a Claude project. So if you don't know what a Claude project is, um, it's essentially a section within Claude that has its own memory and its own context. So this is separate to your just normal chat. This is— think of it as, you know, it's completely forgotten everything about you and who you are. This is like a fresh AI model, and now you can import what you want into here. And when you're in this project, it will only ever reference that information and nothing else that you talk about it with. So that's great because it means that it forgets everything that it knows and you can almost retrain it. So all of our clients have a Claude project which is ever-growing, which is really important. Like, your work is never done on training a, um, a Claude project. So there's a lot of things that you want to put in there. I normally say the basis is your deep research prompt, uh, sorry, your deep research document. The output of that.

And I assume you review that and edit that document to make sure it's accurate, of course, right?

Yes. Yeah, yeah, yeah, yeah. So have a look through that. I again actually have found that Claude is very good at— if you input a document and say like, can you ask me questions based on this document to ensure that the information is correct, that's a really quite fast way to do it. But yes, 100%, like, review it, make sure that everything in there, or add to it as well. There might be some things that are missing out there, right? Add that into your cloud project. So that's your first thing that gives you a basis. After that, the other key things that I almost always put in is an export of your store reviews. So your store reviews are an absolute, like, gold mine for how your customer speaks, what they love about their product, what they hate about your product, everything like that. Really great stuff. So many great ad ideas have come out from, from just like reading the reviews.

Yeah. And just for people that are not e-commerce, you might save customer testimonials in a huge spreadsheet, which is what we've got, and you can export that as a CSV file and effectively have the exact same thing, right? So wherever you can grab testimonials from, they might be in your store, they might be in some sort of a third-party tool that where all these reviews list. I mean, they're all over the place, right? And, uh, getting that in there is super powerful. What else do we need in there?

Yeah, so after that, it depends on how deep you want to go, right? Like I said, this is ever-growing, so you don't have to put everything in there now and then that's it forever. Uh, but I'd say one thing that we put in, we have like an internal document which basically says, and again, treating this person as an intern, this is the type of stuff that I would give to any new starter. And it's essentially like a 10-page document which is like this is what Fraggle stands for, this is who we are, this is what we do, and this is what a good ad is. Because to every agency, to every brand, what they determine is a good ad is completely different. So for us, we write, you know, we have this document, we feed it to all of our models, and we say this is what a good ad is. So it gives the AI a bit of a benchmark, basically, I'd say, um, at this point. And then from there, you can give it past ad performance potentially. So again, go your Facebook Ad Manager, export your data from there, and put that into, uh, into your Claude project. The issue with that is that it doesn't have any context to what these ads are. All it sees is, you know, ad name and results.

Is it possible you could print out like, or take screenshots or something of some of the best performing ads so at least has some context there?

Yeah, yeah, yeah. So What we often do is instead of just exporting a review, what I'll often do is I will pick the top like 10 ads, for example, right? And I normally do this every quarter or so, and our creators will like update this on an ongoing basis. But you take the top ads from the past quarter, let's say, um, and then I use a software called Poppy. I'm not sure whether I don't know if you know of that, but one great thing about Poppy is that it watches short-form video. So unlike a lot of other AI platforms that will just transcribe video, this physically watches it. So it knows what the picture looks like, it knows what's going on. For video ad creative and even for image ads, that's a must. And so what I'll do is I'll feed it that video along with an export of the data from that video, and I'll say, this is the video, this is the data that went along with it, can you summarize the learnings and why this ad performed the way it did, and call out any core visual elements that might have made this ad get those results? That's a really great way to do it and to you know, to like link them together.

Another great tool for this is Gemini, because Gemini is built with vision built into it. You could basically download the video and upload it into Gemini and you could effectively give it a script to say, watch this and describe in detail like the creative assets that are happening and like the hooks and all that kind of stuff. And it'll do a really great job with any kind of a video. And a lot of people don't realize that's what Google's strength is, the fact that it's been built on top of the YouTube, you know, ecosystem and all that fun stuff. Hey, before we get back to the show, I want to talk to you about something I've been thinking a lot about lately. You already use AI, and if you're like me, every single day you're using it, and you're not someone who needs to be convinced that it matters. But here's what I've learned from working with hundreds of marketers inside our AI Business Society over the last 2 years. The problem is almost never AI. It's trying to do it all on your own. You're testing tools that you're likely not using again. You're rewriting prompts because they don't sound right. You're probably watching tutorials that give you lots of ideas, but you have no idea what to do with it. And you're working harder with AI, not smarter. And that's exactly why I built the AI Business Society. Every live training is led by an expert practitioner who uses AI in their marketing all the time. And I'm personally there on every single one of these training sessions to make sure that nothing gets too crazy or abstract. You leave with something you can implement that very day. Plus, you're surrounded by a community of marketers who are just like you, and they're all there to support each other, and you get to be part of this awesome community. The investment is $497 for the full year, and if within the first 30 days you decide it's not for you and you join, then just ask for a refund. We're not gonna ask any questions, but at least you gave it a shot. Head over to socialmediaexaminer.com/ai. That's socialmediaexaminer.com/ai. And now let's get back to the show. So, okay, so we've got this Claude project, and effectively this project has been trained up on our business, right, or our client. And it's got, it's got the research, it's got the customer information, you know, the testimonials. It's got, it's got like, hey, here's what for us is a good ad, or here's our style, and then here's some past performance data. So now we're at the point, I would imagine, where we actually start getting creative, right? So what's the next step?

Yeah, yeah. So once you have that project kind of trained and ready to go, yeah, we can actually start making some ads. There's lots of different ways because there's lots of different types of ads which we can do here, but of course we need to decide what we actually want to create. So let's say we're going to make a static ad, right? That's like a nice simple piece of video, uh, piece of ad creative for us to use here.

So you said video, do you mean— you mean image, right?

I mean image. Yeah, sorry. Yeah, so, so let's say we're going to make, um, an image ad. So an image ad usually consists of an image and some text. To some style, basically. So the first thing we want to do is we want to go to our Claude project and we want to say, using the knowledge that you have, write me some static ad headlines. Ask me any questions that you need in order to complete this task. I normally always include something like that at the end of my prompts because then instead of it just going on at itself it can ask you things to kind of clear its mind a little bit. That's usually like where I'll start. It'll give me, you know, a load of headlines that is based on that knowledge. It might pull some customer, uh, content in there. It might put it in quotation marks, you know, like whatever. And then what I like to do is, once it's given you some, pick 2 that you like and 2 that you hate.

You're asking the AI to do this, or— right, or are you doing this?

I'm doing this. Okay, so once, once I, I have received those headlines that it's written, I will pick out two that I really like, which are in the direction I want to go, and two that are completely wrong. And I'll say to the AI, give me more like this and less like this. Because again, remember, all these chats compound, and so you're, you're constantly training it.

The memory is coming in handy here because it's learning what you don't like, right?

Exactly, exactly. Yeah, so it's— so in instead of just you reading the chat and going, oh, I don't like that one, like, tell it that you don't like that one and why as well. And so again, that's an ongoing process that will just continuously get better. And basically just work with the AI then in order to get a creative, like, the, the copy for your ad down. You can also ask it at this point to, you know, dream up some images to go with it. Like, should this go with a product image? Should it go with a, you know, an image of the customer? Whatever, right? Again, you can brainstorm. That's a great way to, to think of it. Think of this, that you're brainstorming with your creative team in order to come up with this idea. Give it some ideas, it'll give you some ideas. And then eventually, once, once you have your copy and maybe your image idea, you can move on to the next step, which is actually piecing it together.

Okay, so let's talk about this a little bit. I have a background as a copywriter and I use Claude all the time to write. And one of the critical things that you may have missed, folks that were listening, is you want to ask for multiple variations. So generally speaking, depending on what model you're using, if you're using the latest Opus models, they'll probably give you 6 iterations typically, at least when I'm— for the stuff I'm doing, which is emails, which are beefier. But I would imagine for headlines it's going to give you potentially more. And then what I've done eventually is I've asked it, what is your favorite? And I've let AI tell me what it thinks its favorite is. Now, this is because I'm having to generate 6 completely original emails based on a theme, and it's a lot of work to read all 6 of them, right? So what I'll do is I'll read its favorite and I can tell instantly whether it's good or not, you know what I mean? And if it's not, I'll go to the others. But asking it to create multiple variations is really, really critical. You said write me some static ad headlines. Are you simply having it write you headlines first or are you having it write you all the copy? Like, let's talk about that a little bit. And what creative direction are you giving it? Are you just letting it come up with the creative direction? Like, let's talk about that a little bit.

Yeah. I mean, it depends whether you have any creative direction already. You might have an idea of what you want to create and you might have got that from looking at other brands' ads, whatever. You know, you could even find an ad that you really like and paste it into Claude and say, Rewrite this for my brand, and it will do it right there. There's a lot of different ways that you can go about this depending on how creative you are as a person. You know, we hire people that are creative, and so often enough they will go to Claude with some kind of idea that, that they want to create, and they will tell Claude that I'm planning on making a static ad for XYZ brand and I want it to focus around the messaging that, you know, you need to be well hydrated before you run, and so you need to use this product.

Yeah.

So they have like a, you know, and we also want to make sure that we hit the persona of marathon runners. So you might go to it with that kind of information, or you might just go and be like, let's brainstorm some static ideas based on the data that you know, and let's see what we come up with. The more ads that you create, the more data you have, the more direction you're probably going to give your AI.

Yeah. And Claude is really good at images, as are all the models. So you can take screenshots and plop them in there if you've got a creative inspiration from some picture you took somewhere or something you saw. So you're writing— okay, so you're having it write the ad creative and you're going back and forth with it. And let's say you've got your ad creative down and you're really happy with it. And let's just say it's a headline and a couple of sentences or something like that. Um, what's the next part of the process? Because Claude is not good at creating images. No, but you know, there's probably something here, so let's keep talking this through, you know.

Yeah, so at this point, this depends on how you want to build your creative. You can go down two routes— well, three, I guess. First one is fully AI. The entire image, text, everything is going to be generated by AI. The next one is like a hybrid approach, which is what we usually use at Fragle. The next one is completely manual. Okay, so I would advise hybrid or manual, in my opinion, as someone that makes a lot of ad creative, deals with clients that want changes. That's the important part. So you want to decide on an image model to use. Uh, this can be, uh, Nanobanana, or it can be ChatGPT, whatever you want to use here.

Um, and by the way, Nanobanana is part of Gemini, folks. So if you have Gemini, you have Nanobanana.

Yeah, correct. Um, so it's such a stupid name, isn't it? I don't know why they called it that. Yeah, it really confuses me every time.

So I'm gonna imagine prior to— I mean, Nanabanana is probably the one, or Nanabanana 2 is probably the one you guys are using the most right now. Is that accurate?

Nanabanana 2 Pro, yeah, correct. Yeah, yeah, yeah, that's— yeah, that's what all of our, all of our team use to make basically any type of image stuff. It's very cheap.

So how do you, how do you get the great quality images? Let's talk about that. Like, is Claude helping you create the prompt for that, or let's Yes, yeah, yeah.

So, okay, so let's say that we go in the hybrid approach, right? So in this approach, we are going to generate the image with AI, but we're going to add the text. That's like the easiest way to think about it.

Oh, okay, so you're going to pull it into Illustrator or Photoshop or something like that and add—

yeah, okay, yeah, yeah. So I mean, you know, for the sense of like trying to do this cheap, let's say Canva, right? Canva is such a, a great tool. Okay, so, so what I do is you either have an image idea that Claude has already given you, or you want to dream up an image. So let's say that my idea for this image is I want the product on a purple background, and I want it like a nice studio, studio shot. So I would tell Claude that. I would say, write me a prompt for Gemini.

Yeah, for—

yeah, yeah, yeah. And drag in a photo of your product so it knows what it's dealing with, and say, write me a, uh, a prompt. It— I, I want a professional studio product shot of the, uh, of the product which I've given you. It needs to be on a purple background, soft lighting, really crisp, needs to look real, whatever, right? You can be fairly brief here. And again, ask me any questions that you need. To make sure you do this correctly. It will then ask you some questions if required, or it might just fire you back a prompt. Read through the prompt.

These prompts are not dimensional, right? So you can make vertical, square, 16:9, all that stuff, right? It's more about what it's gonna compose when it makes the image, right?

Yeah, so read through the prompt to make sure it lines up with roughly what the direction you want to go. At this point, yeah, you can say, you know, make it square, make it vertical, whatever. And then the best thing to do then head over to your image gen of choice, input the prompt, input your product. Also forgot to mention here, but when you are generating that prompt inside Claude, make sure that you tell it that you are going to attach an image of your product.

I see, so it references it in the prompt.

Got it. Yeah, yeah, yeah, yeah. Otherwise it's just going to generate something completely random. So we dump the prompt, into, um, into Google. We give it our product image, we click go, we see what comes out.

And then do you just get one at a time, or what are you doing, multiples, or what?

It depends. So if you're— it depends on how you're accessing Gemini. If you're accessing it via like the API, you can get it to do like 3 or 4 at a time. Uh, you know, there's platforms like Arcades or Higgsfield, or, you know, like all these different places that you can access access it. If you're just using the chat in Gemini, there's, you know, whatever, you're just going to get one shot. But yeah, what I normally do is just do one, and then I will save that. If there's things that I want to change, obviously that might be it, might be perfect. If there's things you want to change, paste it back into Claude and say, this was the output, here's what I want to change.

Paste the image back into Claude. Okay, got it.

Yeah, yeah. So you're showing Claude what the output to that prompt was, and you're saying, this was the output, really good, 'But the product's too big,' whatever, right? So you're, you're the middleman between the, uh, between the AI.

Fascinating. Now you can do the same thing with these other models, but I like the idea that you're training your original project to understand how to generate the best possible prompt so that next time it does it, it kind of knows it needs to add a little bit more description here or there in order to get what you want, right?

Yeah, yeah. And, you know, we can, we can go down a rabbit hole here doing using Claude skills and stuff like that as well, right, in order to generate you the right prompts every single time. But, right, you know, that's, uh, that's a whole other can of worms, right? But the important part here is that once you have an image that you're happy with, my hybrid approach is that put that inside Canva, add your text, because that way if you want to change the copy, whether it's for an iteration or you're just not happy with it, you don't have to regenerate the whole image.

Yeah, and some other tips here, folks, is that when you are inside of Google Gemini, there's a little download arrow which will download the full 4K version of the image instead of right-clicking and downloading it. Yeah. And another tip about Gemini is the more you have Gemini iterate it, it's almost like it's making a photocopy and it gets more rasterized over time. Yeah. So there's probably benefits to using these third-party tools that just send the whole thing over the API and then get a perfect output every time. But you can also do things like you can tell it to zoom out change the camera angle, all this kind of stuff. Do you have any tips that you've learned as far as like when you have something really good but maybe the aspect ratio or the context of it is off or anything like that?

Yeah, I mean, again, it depends on how you're using these models. I know that in Higgsfield they have a load of stuff like that, right? Like zoom out and all this, and it will essentially like do all the— do it all for you. If you don't have access to those, like I said, the easiest thing is to, you know, print screen it, save it, whatever, put it into Claude and say, I want to keep everything the exact same, but just like, you know, zoom out. Or like you said, just in the chat be like, no, can we just zoom out here? Once I have an image that I'm happy with, I don't want to touch it because I'm scared that it's going to mess it up. Yeah. So usually at that point, if it is something like, you know, change the aspect ratio or something like that, I might then use like another AI tool like Photoshop to like, you know, generative fill it or something like that, something that can, you know, so something that won't regenerate the entire thing, because I'm always a little bit worried that the AI is just going to go a bit crazy.

Okay, what about video? Where do we use AI when it comes to video? Because I mean, we've done other interviews about actually creating AI video. We don't have to go there, but are you using AI to assist you in any of your video ads that you're creating with creators or yourself or anything along those lines?

So when it comes to video, we don't really do much fully generation, like, actual video. But where we do use it is in the scripting process, the ideation process. That's where I think it really shines, because like, going back to what we spoke about before, right, it's really good at copywriting. Claude is very good at copywriting. It's essentially what script writing is. So again, using that Claude project, instead of asking it to create us some headlines, instead we're saying, I have a video idea where I want a UGC creator to be, you know, running a marathon while they're speaking to the camera, and they're talking about these new shoes that they have on that makes them run 10 times faster, whatever, right? And so you're giving the AI a general idea of what you want, who you want in the video, what you want the target length to be, everything like that. And then say, you know, based on your knowledge, go ahead and write me a 30-second UGC script for this. And it will— no doubt the output will probably be something which will be timestamped. It'll, you know, say like, between 1 and 2 seconds, the UGC crew is going to say this.

We'll tell you how to art direct the scenes too.

We'll actually— yeah, yeah, yeah. Well, yeah, yeah. I mean, yeah, it can go into like great detail. It can go into as much detail as you want. Usually what I say inside the prompt, I say, just give me the script. I just want, you know, a paragraph of script. I can figure it out from there. So That's where we're using AI quite a bit there, is by, you know, using it to give us a first draft of a script the same way again, like you would with an intern, like getting it to write you that first draft. And then you then, as the expert, will then go in and fine-tune it and get it to the place where you want it to be.

You know, you've run your agency for a long time, and obviously you didn't have these AI tools before. Hmm. Are they actually not just saving you time, but are they increasing conversions? Are they allowing you to get way more creative than what you guys could do on your own? Oh yeah, like, because a lot of, a lot of people listening might have access to creative people and they're wondering, you know, like, can it really outdo my creative staff?

I think—

or is it collaborative?

Yeah, I think it is more like you— in my opinion, nothing is going to beat a script that's written by someone that is in this every day, right? Like, you know, because often— well, you are, you're writing scripts for humans to consume. We as humans the best way to say something. We, we understand the nuances of saying things. We understand context clues, all these things which an AI really struggles with. And so I don't think anything is, is ever going to be a, you know, a human-written script. But like I said, it can get you a first draft, which often enough, as you know, is the hardest part. Like, you know, yeah.

And then you can edit it, right, and refine it.

Exactly, exactly right. You can, you, you know You can take a script that AI's written and only use 2 lines out of it, right? But the important thing here is that it got you 30% of the way there 100 times faster than it would previously. So that's what the important thing here to know is that if you arm creative people with tools like Claude and you help them understand how it works, and the chances are they're just going to use it to make more content faster and it's just going to help them get to where they need to be much, much easier.

Yeah. And this is a big deal for anybody. Like, I used to own a creative agency and I call myself an on-demand creative. When there's a problem, I get creative. But it's really hard for a creative person to be creative all the time because it kind of comes in waves, you know what I mean? Yeah. And getting started is always the hardest part or finishing. It's one or the other. It's either getting started or it's putting a critical eye to it and actually taking it over the line. And this is where I think AI can be super helpful because a creative person can deliver their ideas to it and ask AI to critically analyze it, and it will see things that the creative person cannot see. Because I'm telling you right now, most creative people think their ideas are the coolest thing they've ever come up with.

That's so true. Yeah, yeah, yeah, yeah, yeah, definitely. And I think as someone, you know, that has staff, my hardest part of the job is trying to get creative people to hit deadlines. Yeah, right. Because as creative people, we want to spend as long as we possibly can on something. Exactly. That isn't how, you know, the world works, unfortunately. So, and, but AI tools can allow us to get there much faster. Like you said, it can get us a first draft faster. It can critique our work to get us the results that we want with knowledge that we might not have.

And it's very consistent and it's not fickle. Meaning, generally speaking, it knows when it's been trained well. I found Claude challenges me back. It doesn't, like, pull any punches, you know what I mean? Because I programmed it to do that. It's going to tell you, I'm going to have to push back on that. Yeah, right.

Yeah.

It's, you know, and I love that about that. Fraser, we have just scratched the surface. This has been such a fun conversation. It's my hope that everybody who's listening has at least a really— especially folks that are creatively minded, understands that there's a way to partner with AI in a very powerful way that can make you so much better at your craft regardless of what you do. And hopefully some people that are listening right now that maybe aren't super creative but want to be creative could leverage what we've talked about to actually create, you know, really great output that they could give to someone to film or that they could actually do themselves. Yeah, but I know we've just scratched the surface of what you've got available, so why don't you you tell everybody where they can follow you on socials? I know you have a YouTube channel where you're doing some cool stuff, and if they want to maybe work with your agency, where do you want to send them?

Yeah, of course. Yeah, so, um, for a lot of free content, the best place to go is my YouTube. If you just search my name, Fraser Cottrell, just search that on YouTube. There's a ton of videos on there. I post every single week. You, you know, it's free advice for anyone who wants to make ads. Uh, Twitter, I'm super active on there as well, it's just @SourFrazer. Um, and yeah, if you want to work with my agency, the best place to go, uh, it's just fraggle.com, F-R-A-G-G-E-L-L.com. And yeah, if you're an e-commerce brand and you're looking to scale up with some better ad creative, we'd absolutely love to have a conversation with you.

Fraser, thank you so much for coming on the show and sharing your wisdom with us today.

No worries, man. Thank you so much for having me. Appreciate it.

Hey, if you missed anything, we took all the notes for you over at Social socialmediaexaminer.com/a108. Be sure to follow this show on your favorite podcasting app. And if you've been a listener for a while, I'd love a review on whatever platform you're listening on. And also consider sharing this show with your friends. I'm on Facebook, LinkedIn, and X. Do check out my other show, The Social Media Marketing Podcast. This brings us to the end of the AI Explored Podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI AI help you become more successful. The AI Explored podcast is a production of Social Media Examiner. Before you go, if you're tired of figuring out AI all on your own and testing tools that don't seem to work the way you want them to work and constantly rewriting scripts and prompts because they don't sound like you, the AI Business Society was built for you. Every week we cut through all the noise. Every training is led by expert practitioners, and you're surrounded by marketers who are just like you. And right now you can join for $497, which is our special sale price, and your rate is locked in for life. And you got 30 days to ask for a refund if it's not for you. So what have you got to lose? Check it out at socialmediaexaminer.com/ai, and I'll catch you on the next episode.