Brent Weaver: Hey, what’s up friends, welcome back to day two of Cloudways Agency Advantage. My name is Brent Weaver, I’m CEO of E2M, your number one white-label digital agency production team. What an incredible first day we had together. I’m seeing some high fives, some emojis starting to flow in. Let’s warm up those fingers and get tuned in to an amazing day of content we have coming up for you. Several of you I saw already in chat were hanging out with us yesterday. What did you all think of day one of the Agency Advantage Summit? Let’s see some of those comments in chat. Cidra, thank you. Karen says “Excellent.” Wendy says “Excellente.” We got Adrienne, Grand Rising crew, hello everyone. From Hani, “Invaluable info.” What’s up Landon, good to see you. Jessica Alvarez says “Fantastic.”

We’ve had thousands of people register, and hundreds of attendees over the last day, and we’ve got a huge lineup in store for today. Quick recap: we had sessions yesterday from James LePage, Tim Kilroy, Vito Peleg, and Kushboo Choudhary. We had amazing panelists and so much insightful content. I had so many people reach out to me on LinkedIn and Facebook asking for session recordings and recaps. So if you go into our event platform, you can actually start seeing recordings from yesterday’s sessions, so if you want to rewatch some of those amazing talks on AI, make sure you do that.

What to expect on the day ahead: we’ve got four insightful sessions, how AI is reshaping agency operations, building SOPs with AI agents, how AI is reshaping Cloudways and DigitalOcean, and reinventing outsourcing, and a panel discussion on AI versus human creativity to top it off. And we have an awesome workshop kicking things off with AI prompting. So if you want to level up your prompting abilities, we’re going to be doing a hands-on workshop in a few minutes.

Big shout out to our sponsors, of course, yours truly with the team at E2M. We’re also being supported by WP Umbrella, one of the most fantastic dashboards to manage and maintain your WordPress websites, Template Monster, as well as ProfileTree. If you haven’t yet participated in the Easter egg hunt, you might be wondering, isn’t it after Easter? It is, but we’re running an Easter egg hunt over the two days of the Agency Advantage Summit, and our deadline for submissions is 2 p.m. Eastern time today. So if you haven’t grabbed that worksheet and kept track of the Easter eggs you’re finding through our platform, presentations, and chat, make sure you grab that document and get those submitted by 2 p.m. today.

We also have another game today. We did a couple of games yesterday with our word find and our AI image creation challenge, but today we’re going to be doing a game called AI or Not AI, so I’m looking forward to that. We also have our leaderboard, so keep that chat and engagement going, keep those points counting. For our top participants we’re going to be giving away some gift cards and hosting credits, so stay tuned. And then at the end of today’s lineup, after all of our presentations, games, and workshops, we’re going to be announcing our winners for all of our games and contests, so stick with us until the end.

If you have questions for our speakers, make sure you use that Q&A panel, and time permitting we’ll make sure that each presentation gets a little Q&A time. So up next we’ve got our Prompt Engineering Lab, we’re going to learn to talk to AI. You might wonder if the quality of your prompts and how you’re talking to AI matters, well, the headline is yes, it does matter, garbage in garbage out, or quality in quality out. So we’ve got Ariel Leven joining us. He is a specialist in B2B pipeline building through cold outreach, he leverages AI agents, automation, and Clay to boost SDR efficiency and productivity. So without further ado, let’s welcome Mr. Ariel Leven to the Agency Advantage 2025 stage. Good to see you, man. Are we ready to jump into our prompt engineering workshop? I’m seeing lots of emojis. Dude, you’ve got the stage, take it from here.

Ariel Leven: Awesome. Good morning everyone. So I’m very excited to share something that’s completely changed how I run my agency. This isn’t theory, this is what I do every single day to scale my business using AI. By the end of the session you’ll have a systematic way to get professional results from AI, instead of just the hit-or-miss stuff most people are doing.

A little bit about me. I’m an agency owner like many of you here. I have hands-on experience with digital marketing, I do pipeline generation and a lot of AI automation work and prompt generation. Probably the reason I’m here is I wrote a post a little while back with awesome prompts for ChatGPT, and it’s got 5.2 million views. Things really developed since then. I use generative AI very heavily, because I use it via API, so I do lots of queries. The main tool that I’m using AI through is a tool called Clay, you might have heard of it, it’s a go-to-market tool, but I also use it to create content. And I have real-world experience, I’m not in the lab all day, I’m actually using what I’m going to teach you.

The key thing, and why you should listen, is I think generative AI was actually built for agencies. It’s amazing the stuff you can do with it: content creation, lead generation, client communication, building websites, building apps. This mind map shows you a lot of the possibilities. But don’t try to do everything at once, pick one thing and master it, then move to the next one.

Why does prompt engineering matter? Here’s the thing, most people are using AI like Google. They type something like “Write me an email” and wonder why it sucks. Without structure you get different garbage every time, and AI has no clue about your business, your tone, your industry. It’s like hiring someone and never training them, and of course we don’t want to do that.

So what will you learn today? First, the three types of prompts and when to use each one, this alone will save you money. Second, my four-step process that turns anyone into a prompt engineer. And third, real examples you can steal and use today. And if you want to shoot some examples that you want to try, please do so, and I’ll try to follow the comment section.

Tool-wise, we’re keeping it simple today. We’ll use Perplexity for building the foundation, it gives the best guidelines, and you’re going to see what I’m talking about very soon. Perplexity is also free, and you’ve got about three searches a day on the pro version, so you can join now if you haven’t. From there you can pick your fighter, you can use ChatGPT, Claude, or Gemini for the actual work. Don’t overthink it, they all work fine for agency stuff. Personally, I’m going to use ChatGPT o3 because it follows instructions very well.

So we have three types of prompts we’re going to talk about today: non-reasoning for your daily grind, reasoning when you need it to be smart, and search prompts when you need current info. Most people don’t even know these exist. Let’s understand each one before we get into using them. Non-reasoning prompts are your bread and butter, we use them for content creation, emails, social posts, the stuff you do every day. They’re fast, cheap, and reliable, no overthinking, just solid results. This should be 80% of what you’re doing. Reasoning prompts are when you need AI to be your coach or strategic advisor, when you need it to think through a few stages, you can use it when you’re creating thought leadership content or need a detailed plan, but they are expensive and limited in use if you’re not a paying user, so use them sparingly. And search prompts are for research: prospect backgrounds, competitor analysis, current market trends, anything where you need fresh data. Perfect for qualifying leads before you reach out.

So when do you use each one? You use reasoning models when you need to solve a complex problem that requires logical thinking and step-by-step deduction. You use non-reasoning models for tasks that involve pattern recognition, generating creative content, or quickly retrieving information. And you use search prompts when you need to quickly find specific information from a database or the internet.

So here’s my four-step system that turns amateur prompts into professional ones. We start with guidelines, then we create the prompt, then we improve it, and then we implement it. But the good thing you’ll all be happy to hear is that we’re not going to write the prompt ourselves, we’re going to let AI do the prompt writing for us. So let me take you through a live demo, and then we’ll talk about the three types of prompts, create some prompts, and then talk about projects.

Our first stop is to get into Perplexity, and we want to use Pro. Perplexity is free up to a certain amount of usage. What we’re going to do is type this into the window: “Write the guidelines from a pro for a prompt engineer AI agent who writes specific types of prompts.” So let’s start with search prompts. The reason we use Perplexity is that it’s going to go search the internet and find the latest and best information, and then form its guidelines. So it’s doing some thinking, and it gives us the guidelines, and then shows us what it based it on. Now all I need to do is copy this, and move into my LLM of choice. I’m going to create a new chat using the o3 model, usually when you open ChatGPT it’s probably going to be on 4o, what you should do is choose o3 if you have the Pro version, because that’s going to follow orders the best. So I’m going to drop the guidelines here and bracket them so ChatGPT will know to reference them, and then put in another line and say, “Write the following prompt using these guidelines.” Now it’s going to know that I’m referring to these guidelines, and I’m going to put my specific description here. Let’s say I want to search specific companies, “Search for companies that were recently funded and provide detailed information that could benefit me in personalizing outreach to them.” So that’s all I need, I haven’t engineered the prompt myself.

Brent Weaver: Sorry, go ahead, Ariel, can you zoom in on your prompting area just a smidge for our folks?

Ariel Leven: There we go, perfect. Thanks for that. So I told it, “Write the following prompt using the guidelines,” the guidelines are down here, and it knows I’m talking about these guidelines because I bracketed them, and then I put what I’m trying to do. Now it’s going to do some thinking, and you can see it’s actually sharing the process with us, it’s clarifying the prompt, finalizing it, and now it’s going to spit it out. Now I have a search prompt, but I’m not sure it’s the best prompt, so I’m going to ask it to suggest ways for me to improve this prompt.

And by the way, if you need these, I’m not sure how you guys are going to get the presentation, but I’m going to go a little bit off script. You can go to Linktree, linktr.ee/11ariel, and the first link is going to be this presentation as a web page, which you can use and copy the prompts and the process itself. So it created the prompt for me, and then it suggested some tweaks to make the prompt better. Now the next thing I’m going to do is ask it to rewrite the prompt implementing all the suggestions. Now I have a search prompt that I can use, and it spits it out. Now I can take this prompt and use it in Perplexity, Claude, Google Gemini, and also ChatGPT.

I just want to show you a quick trick. As you can see it called this “recently funded companies,” but what I actually want to do is rename this chat to “search prompt generator,” so next time I can use this to generate a prompt, I can just go in and edit this input, and it’s going to create a new prompt for me. Now in the background I’m going to do a new chat, choose 4o, and ask it to run this search, and it’s going to run the search for me. So what we did is we created a detailed search task and gave it a lot of references, and that means we’re going to get really good results that we can actually use in our business, find the right companies without needing to pay for a Crunchbase subscription or whatever.

Now, the prompts are not always going to hit the spot on the first go, so you might want to improve these a little bit. “Not all required fields were found,” okay, so you might want to improve this in a specific situation, or another option is we can ask it to simplify the prompt and then run it again, or use it with different models, I can take the same one and use it in Claude or Gemini.

Now the next thing, I’m going to go back to Perplexity and put a new query, again asking it to write the guidelines for a prompt engineer agent who writes reasoning prompts, could be for ChatGPT. So before I asked for search prompts, now I’m going to ask for reasoning prompts. It gives me all the data, I can copy it, same thing, we’re going to open a new chat, choose o3, and ask for a reasoning prompt. What I need to do is say “Write the following prompt using these guidelines,” and then paste the guidelines. I close them with brackets, and what I want to do now is use ChatGPT as my consultant, so, “Act as my business consultant and advise me on choosing the best path for my business to survive and thrive in the age of generative AI.” That’s relevant to a lot of us. Now it’s going to write the prompt for me. As you can see, ChatGPT, or whatever LLM you’re using, is going to do most of the heavy lifting for us. Again, I’m going to ask it to suggest ways to improve this prompt, then ask it to rewrite the prompt implementing those suggestions. And then it tells me, okay, you can copy-paste the improved prompt below into ChatGPT or another cutting-edge model, replace bracketed placeholders with your details, and run.

So I’ve engineered this prompt with the use of AI. I’m going to do a new chat, and I want to use reasoning, so I’m going to use o3, and then I’m going to look for the bracketed information and fill it in: company profile, SDR agency, digital marketing, core product B2B appointment setting, AI automation, USA/UK, competitive edge AI knowledge. So I still need to do some work here, because as mentioned before, garbage in garbage out, so we need to put the best information: automation level eight, data quality and access eight or nine, ask me questions, success metrics, market trends we’re watching, generative AI. So I do need to do a lot of work here, but that’s like any questions that you’re going to do, and now let it run. As you can see, it’s going to do some thinking and then give me some recommendations. Naturally, if I were to do it in a situation where I have more time, I’d be a lot more detailed and the outputs would be a lot better. It actually decided to search the web even though I didn’t ask it to, because it wants to know more information. So as the models become stronger and more powerful, they’re going to have this initiative.

The last thing I want to do is repeat the same process for a non-reasoning prompt, because this is where I use it the most, and this is where you can really leverage AI a lot. The game today, it used to be very challenging to create a piece of content or come up with ideas, but today it’s a lot easier and there are lots of channels we can leverage it on. So I really believe in content. It’s actually going to create some sort of dashboard for me, and I’m not happy with this, so I’m going to say, “Can you provide a textual response?” So if you’re not happy, don’t give up, just ask it to provide the information differently, and if you’re still not happy, go back and tweak the prompt. And how you tweak the prompt is you go back to this conversation, prompt engineering, and say, “I actually got these responses,” and you can copy-paste them or screenshot them, and say, “I’m not happy with it, what I meant was I need something else.” Another 30 or 40 seconds of work are going to give you a huge benefit.

A quick shortcut: if you’re doing very specific things like vibe coding, if you’re not familiar with vibe coding, all those apps like Lovable and Base44, that was just sold, where you prompt and you get a website, a landing page, a working app, sometimes a mobile app, sometimes they would already have a prompt generator built for them. For example, if I was interested in engineering a prompt to build a website with Lovable, I don’t know how to write a prompt for Lovable, it’s going to be different, there’s going to be a lot of questions. But luckily somebody already did the hard work for us, so instead of me sitting down for a couple of hours reading all their documentation, I can use a pre-made GPT. I’ve already installed this one, and I can say, “I want to create an AI agency website,” and it’s going to ask me some questions and spit out a very detailed prompt for Lovable based on best practices.

So let’s do the last exercise, and then let’s put it into practice. I’m going to go into Perplexity again and write the guidelines for a prompt engineer agent who writes non-reasoning prompts. Something I’ve done a lot recently is taking an existing piece of content, not necessarily mine, and then repurposing it. So I’ve got a new prompt for a non-reasoning prompt, and I’m going to open a new chat in o3 with the guidelines. “Write the following prompt using these guidelines. I am going to input a YouTube transcript, and the output I want is a blog post, a LinkedIn post, and a Twitter thread. The original video is not mine, so make sure to make the new content unique.” Now this one I’m going to take from end to end and show you how it creates it and how we can use ChatGPT even more. We’re also going to use ChatGPT to create an image and content for an image. Remember, we want to ask it to suggest ways to improve this, then implement it.

While it’s thinking, I can go into YouTube and choose a video, I’m going to choose one that’s long enough, “Generative AI in a Nutshell.” I’ve got this tool here, and I’m going to make sure it’s picking up the right video, and copy the transcript. I come back to ChatGPT, copy the prompt, open a new chat, this time I’m going to use 4.5 which is good for writing, put the prompt there, look for where I should paste the transcript, and let it run. It’s going to think through it and give me some content. I can actually ask for this content as HTML if I wanted to. So this is pretty much how we generate prompts using AI. But what I want to teach you now is how to get very lazy in using prompts.

Can I just, what type of agencies are you guys running? If you can put it in the chat, I can give you some real data. Web dev, SEO, PPC, okay. So as you can see it provided the information, but the last thing I want to do is say, “Create a prompt for an image generator that would suit this social media content, make it 1024 pixels by 1024,” which is going to look okay on most platforms. So now I have the thread, the LinkedIn post, an outline for the blog which I can later ask it to expand, and now it’s also creating, it created an image prompt because it knows what I’m talking about. If I’m happy with it I can say “Create it please,” and now it’s going to start creating it. Oh, that’s 4.5, my bad, 4.5 doesn’t know how to generate images, 4o does. A website build, funnel, CRM, awesome. Public relations, great, interesting.

So whilst it’s creating the image, I want to talk to you about projects. Projects are, as I said, for lazy people, and I mean that as a compliment. We set them up once with your company info, and AI remembers everything, no more explaining who you are in every conversation. There are three components to a project: the knowledge base with your background and examples, guidelines for different types of content, and instructions that tell AI how to behave. So for example, if I want a project that knows about me, I’ll put my CV, an executive summary about me and my company, and guidelines.

What are we trying to solve with projects? AI has a specific context window, which means that as conversations get longer and longer, it’s going to forget the original information we gave it. But a project maintains all that information, and we can have projects for different things. Like if we create content for specific clients that have specific guidelines and tones, we can have a project for each specific client, and specific projects for our public relations and different marketing collateral. So I have the image now, I can download it, I have the thread, I have everything.

So how do we create a project? Let’s say I want to create viral LinkedIn posts. I know there are swipe files of hooks and copy for posts, so templates. I’ll create a new project and call it “LinkedIn post generator.” Now in the instructions I can say, “Whenever I input an idea, consult the knowledge base, find the top three post templates to use, and output the three posts.” The instructions should be a little better, so I can write them in Perplexity, I go to Perplexity and say, “Write the guidelines for a prompt engineer agent who writes instruction prompts for AI projects,” and I go through the whole process again and put it there. And then I add files, the files are going to be the swipe files, let’s see if I can find something suitable. I already have here 160 LinkedIn post templates, and now it’s going to upload it. Now whenever I have an idea for a post, I can go into here and say, “adoption of generative AI in web dev agencies,” and it already knows what I want to do, I don’t have to write a long post because it has everything there. Before-and-after format.

So you can have projects for different parts of the funnel, for website build, software architecture. You can have a project for every part of your business that is repeatable, that the AI needs a lot of context for, but you can’t be bothered to input the content into it every time.

Brent Weaver: Hey Ariel, we’re coming pretty close on time here for the workshop. We can definitely complete what you’re working on, and I don’t know if your camera overheated or something, but that’s okay.

Ariel Leven: Yeah, let me change the camera. It doesn’t seem to work all of a sudden. So with projects, make sure you give it all the information. The one you want to do today, if you’re creating content for your agency, and if you’re creating content for LinkedIn, you definitely want to create a viral content generator, and you want to have one for your company information as well. So as you can see, you’ve got the prompts here, this is in Perplexity, this is in the LLM, as I said all three. Just make sure that you use projects, use the four-step professional process, and match the prompt type to the task.

The first thing I would do to implement this is go and write everything that we do in our agency, especially that is generative, that is writing-based, that involves creation of things, then choose the one that’s the most costly or most annoying, or you might say, I just want to start with something simple and get some success. Choose that, create a prompt for it, perfect it, teach your team to use it, use the pro versions if you can spend the $20 or so, and then set up your first project as well. So that’s that. Do you have any questions? You can find this as a website on Linktree, 11ariel.

Brent Weaver: Amazing. Ariel, if we can get a link to that, maybe just throw that in chat. A session’s like on fire when people keep asking, is this recorded, can I get the resources? So there was definitely a lot of meat in this one. All right folks, let’s get some questions, we probably have time for a few questions for Ariel on how to build out these prompts. This is coming from Jill: which version of ChatGPT does everyone like?

Ariel Leven: What a great question. I’m not sure everyone likes, but I do want to recommend you guys visit a site called Chat Metrics, which my friend built with vibe coding, it’s called chatmetrics.live, I’m going to put it in the chat. So what it does, I’ll share my screen. So this is going to help you more in finding the best model for what I’m looking to do. For example, if it has a little crown it means it’s the best in the category. If it’s blue, you only get it in the paid version, if it’s green you get it in any version, even the free version with limitations. Personally I use Gemini and Claude and ChatGPT, I pay for all of them. Perplexity I haven’t paid for so far, but I might in the future. It’s really a matter of taste. For coding, naturally there are some winners there, but for writing I think ChatGPT and Claude are very good.

Brent Weaver: Really depends on your use case, and probably to some level personal preference too, just in terms of the style. So I know you’re doing a lot of prospect research, SDR stuff, are you finding that people in that role who are doing outreach have been leveraging tools like this? Is this kind of the new standard for people in the SDR space, or are there other positions within a sales team doing this work and handing the research off to SDRs?

Ariel Leven: In the last year or two there’s been a rise to a new role called the GTM engineer, the go-to-market engineer. It’s somebody who’s going to do most of the work, set up most of the automation, and just drip-feed the leads to the SDR team. But that’s in larger teams where you have five, 10, 20 SDRs or more. Smaller teams are still relying very much on the SDRs’ work. People are using it, but not enough. Those who are using it are winning, because you can personalize messages, and score and qualify your leads with AI, things that we used to hire people from other countries to do, and now AI does a lot of it tirelessly, 24/7.

Brent Weaver: I find that it requires such a different way of thinking. I know you and I beforehand were talking about whether AI is helping us become more or less intelligent. I do feel like there’s a certain level of systems, abstract concept building, that’s required in order to get the results you’re talking about. You have to think about things in terms of frameworks and what you’re ultimately trying to get. But you’ve got these extraordinarily powerful tools that are picking up some of the more mundane things. Have you found that overall having these tools has improved your go-to-market results?

Ariel Leven: You can also allow yourself to think, what would I do if cost was not an issue? Now you can run different prompts doing advanced research on every prospect, getting some amazing data on them, writing a super personalized message that otherwise would cost you dollars per each lead, but now it’s going to cost you a few cents. So you can expand your thinking and be a lot more effective. But it’s making the go-to-market game a lot more challenging, because if some people are doing amazing work, then you’ve got to level up. If you’re not using these methods, all of a sudden your outreach is going to start looking not so personalized.

Brent Weaver: I feel like I’ve noticed my inbox seems more full of personalized outreach, people that know lots of things about me all of a sudden. Well, it never replaces a great offer at the right timing, right?

Ariel Leven: Exactly. You can be very personalized and interesting, but if the offer doesn’t resonate and the person doesn’t have a need, it’s not going to move the needle. So it made us more efficient, it made us reach more people, quicker and better, but it’s not a night-and-day difference, it’s a good difference. A lot of it is just keeping up with things.

Brent Weaver: And it sounds like maybe forgoing some additional software costs, or subscriptions to databases, or workflows you might have outsourced to an overseas SDR team, your team is able to do that work without that dependency.

Ariel Leven: Yeah, I’ve got a smaller team than I ever had, but we can do the same amount of work with less complexity. We need to go through the first stage of learning, understanding, and leveraging AI, getting some consultations, consulting with friends and experts, but once we have that, we can be a lot more efficient and quick, and we can do a lot more. I’m still to this day, I’ve been using this for the last years, and I’m still like, wow, this is amazing. There are definitely still some magic elements to using AI on a daily basis.

Brent Weaver: Any gotchas, things you thought were going to work really well and then the data was incorrect, or some of the personalization came back with hallucinations? Have you ever had things you moved too quickly on in terms of implementation with your team or clients?

Ariel Leven: I think the biggest one is we tried to find managed service providers with a tool called Clay, a Clay agent, which is an AI agent that uses essentially the brain of ChatGPT, but ultimately it wasn’t accurate enough to do that. It would require building something a little bit more complex with multiple agents, and I just decided I hired like five people in different countries and they do that work manually. So sometimes a human solution is still viable.

Brent Weaver: Yeah, and you still, with all this content, you want to inject some personality into it.

Ariel Leven: There’s a very good piece by John Oliver about AI content as well, it’s worth watching, but I’m going off topic.

Brent Weaver: There’s some bonus homework for us tonight. Ariel, this has been absolutely fantastic. Folks out there, let’s give Ariel some virtual love here on the Agency Advantage 2025 stage, let’s see some of those comments and emojis. I know you’re taking time out, Ariel, so thank you so much for being here.

Ariel Leven: Absolutely. Reach out on LinkedIn if you want to connect.

Brent Weaver: Post your LinkedIn before you bounce if you want to put that in chat, so folks have it easy and handy. We’ve got some amazing content coming up, we’re going to be talking about the agency of tomorrow, how AI is reshaping the way we manage businesses. We’re going to be welcoming back Oz Nimrod and Miriam Schwab to the stage, and my co-host Moeez is going to be jumping into the host hot seat. So stick with us, keep the engagement coming, be on the lookout for those Easter eggs, and you all have been a fantastic audience kicking off day two of the Agency Advantage 2025 Summit. See you right back.