Loading summary
A
Okay. We all know building ads is a critical part of marketing. Replit4 just released a brand new ad creation feature and I'm taking it for a test drive on today's show. It does some things really well and it does some other things that I think it really needs to improve. I'm going to walk you through my secret step in Claude to make sure that you save yourself a bunch of time and money before you actually go into replit and build the ads. Then I'm going to show you how to iterate and work on those ads in ReKit to get the best results for the product that you're trying to promote. Let's get to today's show. Every sales rep knows the feeling pipeline looks full activity, looks good, but you're not sure which leads matter, which ones are wasting your week, and which call you should have made three days ago. HubSpot's prospecting agent gives you a clear path forward. It shows you who's actually ready, gives you the details your reps need, and points them towards the conversations worth having so reps know exactly where to focus and what they're walking into. Check out HubSpot.com to learn more about HubSpot's prospecting agent. On today's show, we're talking about something that's getting more and more important and that's ads has tools like Claude code and replit and lovable have become really prevalent. And it's easier than ever to like build an app or a product and get it out in the world to actually test, validate and grow that you need to advertise. And so how is AI doing it, actually generating ads? We've had a few shows in the past, but there's a brand new tool from Replit that I wanted to test live with you on the show. And so today we're going to dive into what the product is. We're then going to actually generate some killer prompts in Claude to actually put into replit to try to build an awesome ad. Okay, so literally hot off the presses here. This is a fun promo tweet from Ripple. Sometimes you don't have a marketing team or the money. And this is what I'm saying, sometimes if you're just testing out an idea, whether you work at a larger company or you're a solo entrepreneur, you just want to like get an idea out there and validate it. You know, Kieran and I are writing this book. We tested titles with ads to kind of understand what title might resonate best with with our audience. And so the easier you can create and iterate on ads, the faster you can learn. So Replit has introduced this. Let's watch the video real quick for a second here. All right. What if you can make scroll stopping ads? Replit is making ad creation fast and easy. Okay. High end luxury fragrance brand ads built your ads ready in minutes. Your entire ad pipeline. One prompt. Oh, so the value prop they're talking about here is you go from a small number of ads to a lot of ads, which I do think is interesting, but in this case, sometime it's going to be just going from no ads to an ad or a couple of ads. So the value prop here is simple as they're trying to reduce the friction of creating ads. The question is, are they really reducing the friction? Are the ads actually good and would we run them? That is a big, big question. The first thing I'm going to want to do though is I'm going to want to make sure that my prompt and my input to create those ads is actually good. And normally I really like to prompt in Claude. You can prompt wherever you want. You could write your own prompt and put it directly into Replit. But I find that Claude Opus 4.6 is a really good model for prompting. It helps think things through with me. So let's give it a try. I'm using Willow Voice as my voice to text so that you can hear what the prompt is versus me typing it awkwardly. And so here's my prompt. Hi. Yes, I want your help building an amazing prompt for Repl dot4's new ad making skill and feature. This prompt should be to create three ads each for LinkedIn, Instagram and Google AdWords. For HubSpot's customer agent product, I want you to go and do deep research on that customer agent product. Understand the core selling benefits, understand how you should position it, understand the best practices for advertising on those three platforms that I gave you and come back to me with a deep quality prompt as if you were an ads master, a leading marketer who spent all day doing ads and give me that full prompt that I could go and use in Replit and Willow then takes that and turns that into text. And I'm just going to run that in Claude and see what Claude comes back with. We just dropped a free AI ad prompt kit inside you get a complete worked prompt so you know exactly what good looks like. An adaptation guide to make it your own platform specific creative direction for LinkedIn, Instagram and Google and a troubleshooting guide.
B
It's Built around replit's ad creation tool,
A
you can, you can go from a blank page to ads that are actually worth running in minutes. Get it right now. Click the link in the description. I find to really do generative AI projects. So whether that's images, that's video, things that are broader than just plain text, getting that prompt right up front is hugely critical. I want to underline that it's really important to take this step of working to refine and iterate your prompt before you go and actually create the assets because it takes a while and it takes a decent bit of credits and usage that you're paying for to go and do this. I'm on the $20 a month Replit plan. I'll burn through that pretty quick if I'm trying to just iterate on these ads over and over again in replit with bad prompts. So I'm going to save a lot of time and I'm going to save a lot of money if I iterate on the prompt prompt in text before I actually go and create the ads themselves. The other thing that I find really good about AI, it's really good at helping you go from 0 to 1, meaning you might not know that much about advertising. You might just have an idea. You have an app you want to test, you just have something you want to get feedback on. You want to spend a hundred bucks in ads to go and do that, that's all totally fine. That's why this prompting step again is so important, because I can have it go and research those advertising best practices and, and I might not be an advertising expert, but it can give me a good baseline and then I can iterate on them based on what I see in the actual ads themselves. So that being said, it is given me a prompt. It's a massive prompt. You are an elite performance creative strategist who manages $50 million in B2B SaaS ad spend across LinkedIn, Instagram and Google Ads promoting HubSpot's Breeze customer agent. It gives full context on the product, competitive positioning, brand voice guidelines. It's giving you the type of ads to create and it's actually giving you the ad angles. And so what you want to do here, and this is like the really important step that nobody does, you want to go through and look at each of these actual ad concepts. In a perfect world, we weren't doing this on the show. I'd probably spend a good 30 minutes just going through all of this. And so for LinkedIn, it's saying, hey, we Want to lead with an angle for AD1? A lead with a jaw dropping custom result that makes the buyer stop and think. Wait. Really use one specific stat from a named customer or HubSpot's aggregate data pain point problem agitation. Okay, let's say I want to do something different. I can just highlight this and say for this instead, I want to make this a a complete screen stopper. I want it to look very anti linked in and more homemade. Update this part of the prompt with that direction. Okay. And so what you want to do is exactly what I'm doing here, which is pick the parts that you don't like. And you know it's AI, so it's going to give you the overhyped. You're a genius. You did a great job. Love that angle. But it's like, look, this is the taste you have. And one of the things I know about LinkedIn is that LinkedIn can sometimes feel a little boring, feel a little slow, feel a little too polished. So what you see here, it added some critical creative direction. Add musk, a complete scroll stopper basically takes what I said and gave the instructions and you would go through and do this with each one of those ads. And what's really great is that it's giving best practices and skills and quality standards. All really good. And again, like I said, you would go through section by section and do some edits. Let's say that we've had the time to do that. Then we're going to go over to repl it. And one of the things you're going to want to do is use a skill because when they say I've got a new feature, it often means they've added a new skill to do work for a very specific task. You see a few marketing skills that it is put together here. One of them is Ad Creative, which is what we're talking about here. And so I've clicked on that skill so it has the full context in addition to my prompt. And I've now pasted in my prompt from Claude. I've picked the ad creation skill so it knows that this is what we're working on and has a unique skill there. And I'm just going to run this guy. And now we're going, we're off to the races. You're now on economy mode. We've automatically switched you to our new economy mode which is up to three times more cost effective. Great. One of the things with these tools is it does cost real money. If you're iterating on these Again, which is why you iterate on the prompts. Prompts are much cheaper. Iterating on the ads themselves through AI is expensive. Sometimes you might get some ads that are close and then go manipulate them in Canva or your design tool of choice. And so why it's thinking here, what is replit? Replit is what's called a vibe coding tool. It's a tool that allows you to take an English language or whatever your native language is, natural language, and input that. And it's going to actually take that, think about what you're trying to do and write code, computer programming code to actually deliver the outcome you want. And sometimes that output is research, sometimes it's a website, it's in this case it is ads. And what's really interesting about Replit, their new version, Replit 4 has a lot more design and creative elements to it. And you're seeing this happen right now with this Replit Canvas feature where it's actually showing you that you can look at and manage and edit things here in the Canvas tool. What it's trying to do is merge, instructing it in your natural language add, as well as being able to do what's called a graphical user interface where you can go and manipulate and move around and do spot edits and everything as part of that. And so repl.itlovable base 44. There are a lot of companies out there whose whole goal is to help you go from an idea to build something very quickly. In this case, we're seeing how they do it. Ads. I've used several AI ads tools, super scale, et cetera. I'm excited to see how Replit does now that it's got Replit 4 with this new ad skills feature.
B
Before we continue, let me tell you about a podcast I love. Marketing School. Marketing legends Neil Patel and Eric Su bring you daily actionable digital marketing lessons learned from years in the trenches. Marketing School delivers bite sized marketing wisdom you can implement immediately. Whether you have a new website or you're an established business. You'll learn the latest in SEO, content marketing, social media, email marketing, conversion optimization and general online marketing strategies that work. Today they just did a great episode called 10 SEO Lessons that Still Work in the AI Era. Listen to Marketing School wherever you get your podcasts.
A
So while it's building on our ads, let's check back in on what we've done so far. So far we've looked at Replit's new ad feature. Then we went into Claude and we worked on creating an awesome prompt for our Ads and Claude went and researched and wrote out our prompt. We then made some edits and gave it some creative direction to make our ads even better. And what was really great is we used Willow Voice to talk to Claude to make that process really fast. Then we took our prompt, we came over to Replit, we did a really important step. We turned on to use the ad creative skill so it knew exactly what task we were doing. And then we put in our prompt. Now it's building our ads assets and we're going to look through them to understand what it did well, what it didn't do well. And we'll talk about how we would iterate and make them better. And I want to remind everybody, we had to do a pretty complex task which we wanted to create three different types of ad for three different ad networks. And again, all these ads are for the same exact product. Product HubSpot's customer agent and Claude did the research on that product and told Replit where to focus, what the key benefits and background of that product are. If you're going to do great ads, you have to deeply understand the product so that it understands what the value proposition to the audience could be. So We've got this LinkedIn ad. 77% fewer tickets 0 new hires NutriBee customer result. I like the data. The graphic drives me insane. That is a terrible, terrible image. Right. In theory, the ad concept is good. In terms of copywriting, the creative is not. And what I find with these AI tools is that the imagery and the creative and getting that right, you really have to iterate on. And it's not just a one and done where these tools have gotten very good at the copywriting side, the creative side. This is like a state of the art application is still not quite there. Like for example, the next ad. First of all, this isn't the HubSpot logo. You heard me tell Claude, hey, I need you to use the HubSpot style guide. And so clearly HubSpot's a sprocket, not a steering wheel. So. Well, that's a whole thing. 39% faster resolution breeze Customer agent Okay. By the way, I think this is what it was trying to get with scroll stopping was with this like real life whiteboard. But the problem is it still made it look like a very commercial ad and it faded in. If it had taken a different design approach, this ad could have worked. This ad's not very good. I wouldn't run it. It's pretty bad. If your mental model of an AI chatbot is still press 1 for billing. Then you need to see what's changed. Oh, you have the classic text on image issues. Let's give this a rev. This ad is really bad. The image has nothing to do with the product. The copy isn't as strong as the other two. You have an image overlap over the word support that makes it illegible. The CRM context and value propositions are in this weird blue text and like just hanging out at the bottom. This is one of the worst ads I've ever seen. Please redo it. And I'm not trying to be mean. This is just a bad ad. So again, what I do like is that you can on an ad or different components, you can work on iterating it there. Again, this is not a HubSpot logo. Responsive search ads. Now that's interesting. This because it's Google and it doesn't require all the visual imagery is going to be a lot easier because it's more of a copy writing problem. And this is actually really good. I like the fact that it's not just giving me like plain text copy. It designed what it would actually look like, how it may appear in Google search. And I find that especially when you're doing these type of search ads, getting that visualization is really helpful, especially if you're early on and haven't done a lot of search advertising. And this is a very different thing because it's copy only. HubSpot customer agent resolves 65% of your tickets automatically. Set up in minutes, not months. AI support agent that works across chat, email, WhatsApp and voice. 39% faster resolution, no code required. Start for free. Sign me up. That's good copy. It's very much like if you match that with an intent, somebody who's actually looking for this type of agent, you're going to be really successful. And that is much better. And what's great is it's giving me different headline options to test and it scored them right. Winning is too many tickets, too few staff, team support. 65% auto resolved. And so that's like a very cool iteration and I love that it built me that in this webpage that I could see it all together. And AI is really good at scoring and grading and helping you iterate. So whatever search ad headlines I picked, I would want to kind of continue to say like, hey, I want the best grade possible. Help me work and iterate through those and give it creative direction. So it also didn't do any Instagram ads. One of the things that's nice is if you select a given element, you can see in this text editor that you know, it still has that ad that I'm trying to fix selected. So what I would do is say, you forgot the Instagram ads. Can you do those for me? So, redesigned our LinkedIn ad. Oh, and now it's got these Instagram ads over here. This canvas feature is kind of weird. So if we want to zoom in here, this redone ad is a lot better. Not a chat bot, your AI support teammate. This is a much better message, Much better copy. The data proof points are down here. The logo's still bad. This is what I'm saying. I think we could get to a really good LinkedIn ad here. It's also just like a bad image. I need to give it an image. I need it to give it a copy of the HubSpot logo. I probably need to do 10 to 20 versions of this ad and rev through it before I get something really good. And this is one thing I'll tell you. There's a lot of people out there that are like, oh, you could do this so easy. It's five minutes, it's free. No, you're going to have to iterate on this. And even iterating on it, you're probably going to run through a good amount of credits to iterate on this might cost you 20, 30, 40 bucks in credits just for this ad. Right. And even on economy mode, depending on how much you have to iterate on it so you can get there. These tools are really great, especially if you don't have any design skills or anything like that. Really, really opens things up. If you're a really good designer, they might be slower than actually just ripping and doing some of this stuff in Canva. So that is where we're at in like the current state of the models and the tools. We got some cool Instagram ads here. The before and after. I like that. It's very Instagram. Yeah, AI loves to do these weird, like post apocalyptic illustrations. One of the things that I should have done that I would advise you all to do is give it a not do list. Like, don't do these style of illustrations. Don't use any colors that are not on the brand style guide. Like this weird blue. Like, where's this weird blue coming from? And that will help you get better first revs. And so what's my take on the new add Skill feature in Replit 4? It's a mixed bag. Is it going to get you good ads? Eventually it is, but you're Going to need to iterate and work on that creative. You're not going to get the perfect outcomes one shot unless you're giving it a ton of existing sample creative to build off of and you're really continuing to iterate your prompt. What I suspect is going to happen is to get a set of ads like these nine ads to get really good. Probably going to take you a couple hours. Now, once you have them and you test them and you have some advertising data, the next batch is probably going to happen much slower. And as you know, Kieran, I talk about loop marketing as you run the loop from Express to Taylor to amplify to evolve. The evolve is like how you actually go and think through the creative, take the learnings and make it better the next time. It will get faster and better every time. But if you're just like, hey, I'm trying to get some ads up for this project I'm doing, expect to spend a few hours to get ads that are really, really good that would be at the level of standard that I would want them to be. So shout out to the replit team for innovating, for making ad creation easier and easier. I find it to be similar and in line with other tools like Super Scale and Canva and Base44 and others that do a similar thing. So I think a lot of it's going to depend on where you prefer to work and how you feel about, like this Canvas feature and some of the what I'd call user experience and like subscription software features around the tools themselves. But I'd love to hear from you drop a comment as to the ads you're building, what you're building them and what's working, what's not working, Please hit like and subscribe. Remember that we have done everything you need to do to get ads created and we've run that through from start to finish. We started in Claude. We built out an amazing prompt, we edited that prompt, we put that prompt in replit, and then we showed you how to iterate on those ads in replit to keep making them better and better till there's something that you would want to pay money to run. That was a lot of fun. I really appreciate you joining us on this episode of Marketing. It's the Grain. We'll see you next time. Hey everyone. You know, Kieran and I have been doing the podcast for a while now. We've been at this for a couple years. We love it. We could not be happier to be doing this, but we wanted to take things to the next level. We want to level up the impact we're having with Marketing against the Grain. So the next step of our journey is something we're really, really excited about. We're going to launch the Marketing against the Grain newsletter. And Marketing against the Grain newsletter is going to be amazing. If you are a marketing leader practitioner, you're in the trenches doing marketing every day. This is for you. We're going to deliver right to your email inbox and you're going to get all the behind the scenes frameworks, practices, tutorials from us, from guests we have on the show, and from people even beyond the podcast that we think are going to be helpful and really have an impact on your day to day, week to week doing marketing. You. You're going to love it. It is something we've been talking about for a while. We're really excited to have it out in the world. We've already got a hundred thousand marketers who are on this newsletter. Please join. It's completely free. We'd love to have you as part of the Marketing against the Grain community. And it's easy. You can click the link in the description below or you can head to marketingagainsthegrain.com subscribe.
In this episode, Kipp Bodnar takes listeners on a hands-on, critical exploration of Replit4’s new AI-powered ad creation feature. The main theme centers on evaluating how effective AI is in actually delivering "good" ads, dissecting not only the technical process but also the real-world results and limitations. The journey includes live prompt engineering in Claude, practical walkthroughs of generating and iterating ads, and a frank review of the creative output—giving marketers actionable insight into where AI shines and where it still lags behind.
[01:29–03:20]
[03:21–07:40]
Kipp walks through his "secret step": using Anthropic’s Claude Opus 4.6 to craft and refine comprehensive prompts before feeding them to Replit.
Emphasizes the huge impact of a well-researched and tailored prompt on final ad quality and cost efficiency:
“It takes a while and it takes a decent bit of credits and usage… I’m going to save a lot of time and I’m going to save a lot of money if I iterate on the prompt in text before I actually go and create the ads themselves.” (A, 05:39)
Key prompt elements:
[07:41–11:38]
"A lot more design and creative elements … they're trying to merge instructing it in natural language as well as being able to do what’s called a graphical user interface." (A, 10:24)
[12:23–20:47]
LinkedIn Ads – Copy: Good, Creative: Awful
Example:
“77% fewer tickets, 0 new hires, NutriBee customer result. I like the data. The graphic drives me insane. That is a terrible, terrible image. Right. In theory, the ad concept is good. In terms of copywriting, the creative is not.” (A, 13:53)
Issue: AI-generated images/logos are off-brand, visually weak, and often irrelevant (e.g., incorrect HubSpot logo).
Critical note:
“This is one of the worst ads I’ve ever seen. Please redo it. And I’m not trying to be mean. This is just a bad ad.” (A, 15:48)
Google Search Ads – Where AI Excels
“This is actually really good. I like the fact that it’s not just giving me like plain text copy. It designed what it would actually look like … that visualization is really helpful.” (A, 17:13)
Instagram Ads – Needs Directives to Avoid Weirdness
[21:05–24:50]
“A lot of people out there are like, oh, you can do this so easy. It’s five minutes, it’s free. No, you’re going to have to iterate on this. And even iterating on it, you’re probably going to run through a good amount of credits.” (A, 20:24)
Prompt Engineering Wisdom:
“Getting that prompt right up front is hugely critical … it’s really important to take this step of working to refine and iterate your prompt before you go and actually create the assets, because it takes a while and it takes a decent bit of credits and usage that you’re paying for to go and do this.” (A, 05:15)
Brutally Honest Creative Review:
“The image has nothing to do with the product. The copy isn’t as strong as the other two. You have an image overlap over the word support that makes it illegible ... This is one of the worst ads I’ve ever seen.” (A, 15:41–15:50)
Caveat for Marketers:
“You’re not going to get the perfect outcomes one shot unless you’re giving it a ton of existing sample creative to build off and you’re really continuing to iterate your prompt.” (A, 22:09)
Pro Tip: The ‘Not Do’ List:
“One of the things that I should have done … is give it a not do list. Like, don’t do these style of illustrations. Don’t use any color that’s not on the brand style guide.” (A, 20:19)
Replit4's ad maker delivers on its promise—eventually. For marketers without design resources, AI tools like Replit can jump-start your ad pipeline, especially on copy-focused platforms. But plan for multiple rounds of prompt and creative iteration, be precise with brand constraints, and expect to spend credits and time to reach results you’d be proud to ship. AI isn’t a set-and-forget solution—yet.
For more hands-on tactics and real marketing talk, drop your feedback, check out the newly announced Marketing Against the Grain newsletter, and tune in next week!