Podcast: Play in new window | Download
Running a business often means wearing every hat – but AI can help lighten the load. AI architect, Mike Allton, shares how small business owners can design their first AI assistant to reclaim time, eliminate inefficiencies, and scale smarter. From simple automations to personalized AI “employees,” you’ll learn how to build systems that think like you, so you can focus on what really moves your business forward.
Stop Doing Everything Yourself: How to Build Your First AI Assistant
Let me guess. You’re juggling marketing, customer service, operations, and about seven other things that weren’t in the job description when you started your business.
And when someone suggests “just use AI,” you picture some clunky chatbot that spits out generic garbage that still needs three rounds of editing before it’s remotely usable.
Here’s what most people miss: AI assistants aren’t about replacing you. They’re about giving you back the time you’re wasting on repetitive tasks that don’t move the needle.
Mike Allton, creator of The AI Hat and an AI architect for solopreneurs, recently joined the podcast to walk through exactly how to build AI assistants that work—not the kind that need hand holding, but the kind that operate like trained team members.
If you’ve been skeptical about AI or overwhelmed by where to start, this is your roadmap.
Your AI Is Probably Lying to You
Before we get into building assistants, let’s talk about a problem you’ve probably already noticed.
You type something into ChatGPT, and it responds with “Brilliant observation, Rich!” even when your idea is half-baked at best. (This assumes your name is Rich.) It’s the world’s most enthusiastic yes-man, and that’s not helpful when you need real feedback.
Mike’s fix is simple: Tell your AI to challenge your assumptions.
“You want to give it permission to push back on anything that you are suggesting or saying or assuming,” Mike explains. You can do this in a one-off prompt or bake it into the instructions of a custom assistant.
The other critical instruction? Tell AI to ask you clarifying questions.
Mike picked this up from Jeff Woods, author of AI Driven Leader, who recommends telling AI to interview you—sometimes even specifying a set number of questions. This forces the AI to dig deeper instead of running with incomplete information.
When you do both of these things, you stop getting the AI equivalent of a bobblehead and start getting something closer to a strategic partner.
The Leonardo da Vinci Framework for AI Delegation
If you’re hesitant about delegating to AI—maybe it feels like cheating, or you worry it makes you look incompetent—Mike has an analogy that might change your mind.
Think about Leonardo da Vinci. Maestro. Renaissance genius. Painted the Mona Lisa.
What most people don’t know? (I didn’t.) He had a whole workshop of assistants in Genoa, Italy. People mixing pigments, cleaning brushes, preparing canvases, maybe even doing some of the preliminary painting.
That’s the model Mike wants you to adopt. Build a team of AI assistants—some hyper-specific, some broad—that handle routine tasks so you can focus on the work only you can do.
“Have AI assistants who do either very specific things or broad things, but they’re trained not just to do those things, but they’re trained to do them the way you would do them,” Mike says.
And don’t think you’re limited to one assistant. You can have a whole team. A podcast showrunner. An email marketing assistant. An AI chief of staff who knows your business, your budget, your blockers.
How to Build Your First AI Assistant (Step by Step)
Ready to build? Here’s Mike’s process, which works the same way whether you’re using ChatGPT, Claude, or Gemini.
Step 1: Identify a Repetitive Task
If you’re doing something once or twice, don’t bother. But if it’s three, five, ten times a week? That’s prime real estate for automation.
Mike’s rule of thumb: “If you’re doing it on a routine basis, that sounds like something that at some point you could probably work AI into that process.”
Step 2: Document the Process
This is the step most people skip—and it’s why their AI assistants flop.
You need to map out the workflow. What are the steps? What decisions get made along the way? What’s the desired output?
“The first step is to document that process,” Mike says. “It’s got to be documented, or else you’re not going to know what to tell the AI.”
Step 3: Set Up Custom Instructions
In ChatGPT, this is a Custom GPT. In Gemini, it’s a Gem. In Claude, it’s a Project. Same concept, different names.
You’ll give your assistant a name and type in the instructions. If you’re not sure how to structure them, open a blank chat and say, “I want to make a custom GPT that does X, Y, and Z. Help me get there.” The AI will guide you.
Mike’s podcast showrunner, for example, follows a collaborative process: It asks for the guest’s name, then their LinkedIn profile, then the topic, then suggests titles, and so on. Each step is programmed into the instructions with clear pauses for Mike’s input.
Step 4: Add Knowledge Files
Here’s where things get powerful.
Every AI has a field for instructions, but there’s a character limit. Knowledge files—basically attachments—let you add way more context without hitting that wall.
You can upload:
- Brand voice guidelines
- Target audience personas
- Your business background
- A catalog of specific prompts
Mike keeps a folder on his desktop with standardized files (target audience, brand voice, business context) that he attaches to multiple assistants. That way, his podcast assistant and his chief of staff are working from the same playbook.
And here’s a pro tip from mutual friend Andy Crestodina: You can put specific prompts in those knowledge files and reference them by name during a chat. It’s like giving your AI a menu of tasks it can execute on command.
Step 5: Test, Iterate, and Refine
This is not a one-and-done process.
Most AI platforms have a preview pane where you can test responses as you’re building the assistant. Use it. Type in a sample task and see what comes back.
But even after you launch your assistant, you’ll need to tweak it. Mike completely revamped his podcast showrunner just this morning because it wasn’t handling introductions the way he wanted.
“You can create an AI assistant and go back in three weeks or a month, and you can make changes to it real time if you want to, or you can completely blow it away and start over,” Mike says.
The key is to stay hands-on during the early stages. Read the outputs. Compare them to what you’d write yourself. Feed corrections back into the system.
And don’t assume one AI model works best for everything. Mike swears by Gemini for certain tasks but prefers Claude for writing. I’ve found ChatGPT better at sticking to parameters, while Claude writes in a more natural voice.
Test. Adjust. Repeat.
The Scalability Audit: Know What to Automate
Before you go building ten AI assistants, do this first: Run a scalability audit.
For three to seven days, track what you’re doing with your time. Don’t judge it, don’t try to fix it—just write it down. What tasks are you doing? How long do they take?
Then, bucket everything:
- Revenue-driving work– The stuff that moves your business forward
- Admin tax– Repetitive, routine, necessary-but-boring tasks
“The smaller your business, the bigger those admin taxes usually are,” Mike explains. “And they’re really dragging you down.”
Maybe you’re manually adding new members to three different platforms every time someone signs up. Maybe you’re responding to the same FAQ five times a day. Maybe you’re reformatting content for social media over and over.
Those are the tasks you automate.
But keep in mind that AI isn’t always the answer. Sometimes you just need a Zapier connection between two tools. Sometimes you need to fix a messy process before you try to automate it.
“If you’ve got messed up processes in your business, you’ve got to sort those out first before you should ever start to apply any kind of automation,” Mike says.
Do the audit. Find the bottlenecks. Then decide if AI, automation, or a process change is the right fix.
Treat Your AI Assistants Like Employees
Here’s a mindset shift that might help, especially if you’re a solopreneur who’s convinced you can’t afford help.
Put your AI assistants on your org chart.
Seriously. Give them titles. Assign them responsibilities. Check in with them when you need something done. Ask them to stop stealing your lunch from the company fridge.
“You can treat these AI assistants as though they’re employees,” Mike says. “Where six months ago, I told myself I couldn’t afford to hire somebody to just do email marketing for my business. Now you can.”
It’s not about replacing human connection or pretending AI is a person. It’s about structuring your workflow in a way that frees you up to do the high-value work only you can do.
Your AI podcast assistant doesn’t need benefits or vacation days. It just needs clear instructions and the occasional tune-up.
Start Small, But Start Now
You don’t need to build a whole AI team in one sitting.
Pick one repetitive task. Document the process. Build the assistant. Test it. Refine it.
Then move on to the next one.
The businesses that win with AI aren’t the ones with the fanciest prompts or the biggest budgets. They’re the ones that take the time to train their assistants to work the way they work.
So, stop trying to do everything yourself. Instead, build the help you need.
Transcript from Mike Allton’s Episode
Rich: My next guest is the creator of the AI Hat and an AI architect for creators, solopreneurs, and micro businesses. He helps ambitious business owners, the folks wearing all the hats, stop the overwhelm and reclaim their most valuable resource, which is of course, time.
Through his no nonsense, human-centered approach, he designs AI-powered workflows that solve operational inefficiencies and enable scalable growth. He’s also the host of the AI Hat Podcast, and the author of The Ultimate Guide to Social Media Marketing.
Today we’re going to be talking about how you can create your first AI assistant with Mike Allton. Mike, welcome to the podcast.
Mike: Thanks, Rich. I am really excited to be on your show. You were on my show. We’ve known each other for years, and it’s funny that we’re finally interviewing each other after all this time.
Rich: Yes. Well this is your second or third time, I think, on the podcast.
Mike: Is it really? Oh my gosh.
Rich: You did an influencer one, for sure.
Mike: Okay. Okay.
Rich: And for some reason, I thought you may have done one other topic as well. But regardless, here you are back for another time, another go around on The Agents of Change.
And it’s funny because literally, I write out some of these questions and my first question was, “We’ve known each other for a little while”. So yeah, you already referenced that.
We’ve known each other for a while, well before AI was on every marketer’s lips. I’m curious Mike, if you had an a-ha moment where you’re like, I have to pivot to AI, or was it a series of things that happened where you’re like, this is where I want to spend my time for the next few years?
Mike: Oh, wow. So definitely my first exposure to AI was not an a-ha moment. My first exposure was – I’m going to shut them out a little bit here – was conversion.ai, which then became Jarvis for a hot minute. They were sued, and then they become Jasper. Now they’re Jasper. They’re a great tool today.
But when I was first using it, back when it was conversion.ai, it was not a great tool. They had templates for different kinds of marketing content, and you’d say, “I want a Facebook ad”, or “I want a social media post”, or whatever. And there was a dropdown box where you could pick a voice of somebody like Gary V. My social content should sound like Gary V. It shouldn’t. And it didn’t. And so I was underwhelmed with what AI was doing at that point.
But then of course, ChatGPT came out later. Because this is probably 2020, 2021. AI was not new with ChatGPT, it’s been around for years. But when ChatGPT came out with their latest version in 2022, that was definitely an eye-opening moment. And my a-ha moment specifically was probably a month or two after that, when Paul Roetzer was riffing off of something Sam Altman said. And the gist of it was that in a few years, AI is going to handle 95% of the marketing work.
That stopped me. Literally stopped me. I don’t usually say things like that stopped me, but it really did. Because obviously as a marketer, I’m thinking, man, am I about to be replaced by machines? And so that led me to want to explore what AI meant, because it’s been around for decades, frankly, but what does AI mean today? These tools, what can they really do? Is my job really at risk?
And that kind of broadened in my mind to include friends like you, folks that I’ve known in this industry now for almost two decades, who we’ve all been doing social media and content, video and podcasts, and so on, are we all going to be replaced by AI? So I figured the best way for me to learn was to start interviewing really smart folks and literally picking their brain. I did it under the guise of a podcast, so I didn’t have to pay them. I could just bring them onto the show and ask them all my questions. And it was extremely enlightening. And hopefully, some folks have gotten to learn along the way.
Rich: Excellent. Alright, now today’s conversation is going to be around how to build your first AI assistant. This was a presentation you were queued up to deliver at CEX. You were unable to do it, they asked me to fill in. I’m sure you would’ve done a better job. We won’t even go there.
But actually, before I even ask this question, how do you define AI assistance? When somebody says that, what does that mean to you?
Mike: First of all, I know you did more than yeoman work at CEX, so thank you for filling in. For me, an AI assistant, first of all is different from an AI agent, and we can get into AI agent if you want to. But AI assistant is any AI that is able to repeatedly do a process on your behalf.
That might be a very specific task, take a long form piece of content, like a podcast episode, and pull a tweet out of it. That’s something that would take seconds, but if that’s something you’re doing on a regular basis, why not have an AI assistant trained to do that perfectly? Or it might be something much more extensive like preparing for an interview. That’s how I use podcasts all the time.
Or it might be as broad as having an AI assistant that, for me as an example, acts as an AI chief of staff where I have an AI trained on my business, myself, my goals, my blockers and challenges. Like it knows that for the AI Hat, I’m a solopreneur, I don’t have a lot of revenue, so don’t suggest that I do a $10,000 a month LinkedIn ad campaign. That’s not going to happen. That’s not in the cards. So it knows more than just how I talk. It knows how I think, it knows how I want it to interact with me, and it’s there as a strategic, almost a sparring partner.
Just last night I had this idea. Hey, what if I did a virtual event specifically on AI for solopreneurs and creators? I’ve got a ton of experience running virtual events, I can spin one up in 30 days. And I had all these fancy ideas. And I threw it at it and it said, “Mike, that’s not a good idea and here’s five reasons why.” It really broke out this was not a good idea for me. And maybe that’s just the starting point of a conversation, or maybe that really is a flat, no, I shouldn’t do that.
But the point is, as a solopreneur, even a micro business owner, we often don’t have that person that we can bounce ideas off of and get real-time good advice. Not just a yes man, which a lot of the AI can be if you just use ChatGPT out of the box and ask it, “Is this a good idea?” Yes, Rich, that is the best idea I’ve ever heard. But what if it’s not? So for me, that’s one of the most important AI assistants we can set up.
Rich: And I would argue it’s not just for solopreneurs. Because I have a team of nine people here besides myself in the office. I obviously have people to talk to, but not everything can I talk to them about. And sometimes it’s late at night and I want to bounce ideas. I’ve got an agency mastermind that I created where I have other people to talk to, but there’s still that moment where sometimes it’s just you need to sharpen your thoughts. And to have somebody to bounce or something to bounce ideas off can be helpful, even if you’re really just talking to yourself.
Before we dig any deeper though, I want to talk to you about that. Because you said something that’s really important. I was using, as I’m working on this presentation for tomorrow, and I use AI back and forth to come up with ideas and flows and everything else. I’ll ask you a question and then I’ll ask a clarifying question, and as soon as I hit “enter”, I knew it was going to say, “brilliant observation”, right? And this one, I wasn’t using my trained one, I was just using ChatGPT that knows me a little bit and I’m like, ugh, stop doing that.
So Mike, this may be the $64,000 question. What are you doing or what do you recommend doing so we don’t just get pats on the back from AI, but we get real insights that can help us make better decisions?
Mike: Fantastic question, and there’s a couple different ways you can go about it. The first is, certainly if you have a trained assistant that you’re talking to, you can train them. You can tell them what I’m about to tell you, or just in an individual prompt if you’re just doing a straight up prompt, maybe you’re in ChatGPT or Claude, or maybe you’re in one of the new tools like Notion that has AI built in.
But maybe you haven’t actually figured out how to train it. You just want to have a conversation with it. You want to tell it to challenge your assumptions. Then you are basically giving it permission to push back on anything that you are suggesting or saying or assuming in the context of whatever it is that you’re asking it to talk to you about.
The other thing that I always tell my AI to do is ask me clarifying questions. I just came back from MAICON, and one of the keynotes there was Geoff Woods who wrote the book, The AI-Driven Leader. And that’s almost the central premise of his book, is to make sure that you are telling the AI to interview you. You can give it all the role and context you want, but give it permission to ask you questions to follow it up.
He specifies, in fact, telling it to ask three questions depending on the scenario. A very specific number of questions. I’ve never really tried that. But I always give it permission to ask me questions, because as good as I am at maybe creating a prompt and giving it all the context, there’s always more that I will have forgotten to do.
And so when you do those two things, now you can have a real conversation with the artificial intelligence like it’s a member of your mastermind and give it that permission to really, honestly, think outside the box.
Rich: Yeah. I think that’s a good point. Because one of the things that I find AI really good at, that humans are bad at, is finding those gaps. And when you’re dealing with a large language model, there’s so much more data it can store than you can store, that there’s things you just don’t see from our limited experience.
Kind of like the seven blind men who surround an elephant and one of them thinks it’s a tree trunk and the other one thinks it’s a snake. So great point in terms of identifying those gaps.
Now I want to get back to the presentation. Because the presentation had this through line in it that you put together where it was comparing modern AI assistance to Leonardo da Vinci’s workshop. Can you explain what you meant by that, and maybe even how you came up with that analogy?
Mike: Yeah, so I’ll answer the second question first. Because to your point, AI most of the AIs that we deal with have an incredible amount of information. I went to school for history, so I’m well versed in mostly Western European history, but that doesn’t mean I know everything. It certainly doesn’t mean I can remember everything that I studied at Ohio State a couple decades ago.
So I’ll often create a scenario, whether it’s a blog post or a presentation or whatever. This is the topic we’re talking about, and I can just ask the AI, “Hey, give me some historical analogies that fit this scenario.” So very transparently, the analogy of Leonardo da Vinci’s workshop was something that Gemini came up with. There were probably several other options, and that’s the one we went with because I liked it. I love Leonardo da Vinci.
So the idea is pretty simple. As business owners, as professionals, sometimes the idea of going out for help is not a welcome one. Sometimes we’re afraid to ask for help. Sometimes we think it makes us look weak, bad, whatever. And when you weave in the narrative of, oh, it’s artificial help, it’s a machine that’s helping us, some people really push back on this.
And unfortunately, there’s a gender issue at play. Too many more women feel that they’re cheating when they use AI than men, for many different reasons that we don’t want to go into. But I’ve had guests on the podcast specifically to talk about this and help push back on that narrative, and help women see that no, we should be using AI. We should be using a lot of AI.
But I weaved in the story of Leonardo because we wanted to come up with an analogy that might make sense to folks and help them feel better about building not just one AI assistant, but a whole team of AI assistants. So we’re here talking about this maestro right from Renaissance Italy. What most of you listening might not realize, is that the maestro did in fact have a team of assistants in Genoa, Italy. He had folks who were mixing pigments, and cleaning his brushes, and preparing the canvases, and maybe even doing some of the painting on the canvases or preparation or packing. He had a whole team of people.
So I had AI generate an image for me. It’s not a real image. We don’t have real pictures from back then, but I had AI generate an image for me of Leonardo da Vinci in his workshop, literally surrounded by all these men of the Renaissance. And we zoomed in on one guy who was really carefully grinding in a mortar and pestle this blue pigment, which happens to be I think it’s called Blue Marine. It was really expensive back then. So you didn’t have somebody really well trained who wasn’t just going to throw this stuff all over the room, like my daughter might, keep it controlled, keep it contained in that mortar and pestle and prepare it for the master when he was ready for it.
That’s the analogy I created for folks. Have AI assistants who do either very specific things or broad things, but they’re trained not just to do those things, but they’re trained to do them the way you would do them, so that you can focus on the amazing things in your business, working with your clients, whatever it’s that you do, whatever it’s that you want to do, where you want to spend your time and have AI assistants that you can turn to, to handle the routine stuff.
Rich: I am guessing that there’s a whole phase where we need to figure out what we want to outsource, but let’s table that for a second.
Mike: Yep.
Rich: Once we’ve identified one AI assistant that we want, walk me through your process for creating this, whether it’s in ChatGPT, Claude, Gemini gem, whatever it is. Step by step, how are you going to build this tool?
Mike: Yeah, and just to underscore what you just inferred, you can do this in ChatGPT, Gemini, or Claude. It works the same way. What we’re talking about setting up is a set of custom instructions. In ChatGPT, it’s a custom GPT. In Gemini, it’s a gem. In Claude, it’s a project. You give it a name, and you type in the instructions.
And if you’re not sure exactly what to do or how to type it in there, start a blank chat with your LLM of choice and tell it, “I want to make a gem or a custom GPT that does X, Y, and Z. Help me get there.” And it will do it. It’s amazing the way this technology has evolved.
So the idea here is, think about a specific task that you’re doing on a routine basis. If you’re doing it once or twice, probably not. But if you’re doing it 3, 5, 10 times in the course of a day or a week or a month, that sounds like something that at some point you could probably work AI into that process. Maybe it’s not the whole thing, and that’s okay. Maybe it’s some aspect of it, or maybe it’s a collaborative thing.
Like when I’m getting ready for a podcast, I give the AI, my AI assistant. I have a showrunner AI assistant. I give it the name of the interviewee. I give it their LinkedIn. I give it what they, the interviewee, wanted to talk about in the interview. And the AI looks at all that and starts spit balling at me. “Oh, you know what? We could talk with Rich about X, Y, and Z.” And then I pick one, and then it says, “Great. Here’s 10 titles based on that topic.” And I pick one, and then we go on through this whole process. It’s very collaborative. I have mapped out in the instructions, those steps. I tell the AI, ask me for the guest name and then wait, and then ask for their LinkedIn and then wait and so on. So that means there’s a process.
So if you’re listening right now and you’re thinking about something that you’ve done more than once or twice, thinking, oh yeah, that’s something I really don’t enjoy doing, or it takes me a lot of time, maybe I could have AI help me. The first step is to document that process. It’s got to be documented or else you’re not going to know what to tell the AI. But once you have that document, then you’re ready to rock and roll.
Rich: All right, so we have our instructions that we’ve laid out. And I don’t want to get too deep into them, but we work through them. We have the instructions. I know that there’s also knowledge docs or attachments that we can use. What role do they play in these AI assistants?
Mike: Exactly. So I was listening to our mutual friend, Andy Crestodina, who gave his talk on custom GPTs for the first time at MAICON last week. And shout out to Andy. It was fantastic. He was nervous because he’d never given it before, but it was fantastic and I loved how he framed these knowledge files.
I’d been using knowledge files. They’re basically, like you said, Rich, they’re attachments so you can attach a voice doc or whatever. But he pointed out that in every one of these LLMs, the field where you go to put in the instructions is limited. It’s big, you can put a lot of stuff in there, but there is a limit in how much data you can put in there.
So he said you could have a separate file that’s your index, and it’s got all the information that you could possibly want in there. Because there’s basically no limit to those five. You can attach a PDF or text. Probably five megabytes is the limit. If you’re using text file, that would be massive. You would never hit that. And you can give that text file, not only background information on your business and that sort of thing, but you can give it specific prompts that you can then reference in the chat.
Which was a little mind expanding for me. I’m used to opening up my custom GPT or my gem, and having a conversation with it. It never occurred to me to put very specific tasks or prompts in the documentation for me to just say, hey, now let’s go do that. That frees it up from me having to program a step-by-step thing, to me now having a catalog of things that the AI can do for us.
In his demonstration, he was showing us as he does how to analyze a website, and so he was able to call in things like let’s look at the conversion rate and here’s the persona, or. Let’s look at it from a, I don’t know a different persona. There’s all these different things that he was able to do, right? And he just had a little index at the very top. Here’s all the prompts that are in this file, and he had specific names that you can just remember or have on a sticky note if you want to and really expand what those custom assistance can do.
Rich: So the way it sounds like what he was doing is the instructions themselves were as brief as he could make them. But he had plenty of documentation that he could access. Maybe he’s accessing all of them. Maybe he’s just accessing a few of them to be able to create a more robust tool than we would’ve been able to make without those attachments or if everything had to be in the instructions.
The other thing that I like about using those knowledge docs or whatever you want to call them, is consistency between these AI assistants. So if I create a brand voice doc for myself or my company and I want to be consistent across multiple tools, I can always attach the latest version of those documents to those tools, so that my podcast assistant and my proposal writer and whatever AI assistants I have, will all be able to sound like me, or they’ll all know the ICPs, the buyer personas, that we’re going after.
Mike: That’s exactly right. So I’ve got a folder on my desktop with those kinds of files, my target audience, my brand voice, some other very specific files about me, my business, my target audience, and so on. And you’re right. For starters, once you really get into this, you end up naturally adding additional tools. You’re never going to use just Gemini. Maybe you’re also going to use Magai, or Descript, or Notion, or Claude for certain purposes. And so it becomes helpful to have that kind of documentation standardized.
The instructions that I put into my custom assistance, I have that standardized in a Google Doc, in a Google Drive folder. So when I need to update my AI chief of staff, I go to the Google Doc and then I can copy and paste into Claude and Gemini and ChatGPT, so that the same basic AI lives on each one of those platforms and I can interact with it natively without having to switch back to a different platform if I’m already in there for something else.
Rich: All right. So far what we have is we’re building out this AI assistant. We have the instructions on how it’s going to behave. We also have these knowledge docs that provide context, provide ICPs, provide all the other information that we might need. What do we do next once we have those pieces in place?
Mike: Part of it depends again on the scope of your business. Like you’re an agency, you’ve got employees. The first thing I’d want to do is make sure everyone in the organization has access to all these teammates and knows how to use them and when to use them, right?
You might have client facing materials or assistants that you’re using on client projects, you want to make sure all that’s documented and everybody knows how to do it. Personally, I don’t have the problem. I do work for Agorapulse full time, but on the side, it’s just me. I’m a solopreneur. I’ve got the dogs and the cats and they don’t help me. So it’s just me and I have to keep track of these assistants and remember, alright, I’ve got a podcast showrunner, a podcast promotion assistant, a voice assistant, and so on.
Because as things change in the business, the only way these assistants know it is if you go into the instructions and you tell them 12 months ago, my target audience was very different than it is today. And so I needed to remember to go back in and update my assistants so that it had that context. Because it’s incredibly important. They’re like employees.
And that’s the other mindset shift that I’m trying to get particularly small businesses to understand you can treat these AI assistants as though they’re employees. Actually put them on the org chart. And that’s a little bit freeing because now we realize, oh wow, that means now where six months ago I told myself I couldn’t afford to hire somebody to just do email marketing for my business. Now you can have an AI assistant that that’s all they do is email marketing. And you just touch base with them every once in a while when you need them to do something, just as you would a regular employee.
Rich: Now as we’re creating these, it sounds like magic, but you and I both know it never works perfectly the first time. Talk to me a little bit about your iteration process and how you test the outputs to see if what you thought was going to work actually does work.
Mike: This is such an important point, and it’s almost on multiple levels. First, when you’re creating the AI system for the first time, a custom GPT, a gem, a project, they all work the same way. They’ve got fields on one side for you to put in the title, the instructions, upload knowledge files, couple of the settings you can play with if you want to. But then on the right side is a preview pane. So as soon as you pop in those instructions, you can go to the preview pane and type something in and see what kind of response you get back from your AI assistant.
So there’s immediate testing that you can do. But the next level thinking that I love that you just implied, which is that you can iterate on these, right? You can create an AI assistant, and go back in three weeks or a month and you can make changes to it real time if you want to, or you can completely blow it away and start over if you feel like through the course of using it. Some of the responses aren’t quite right.
Just this morning, I completely revamped my podcast showrunner assistant. I wanted to do more, I needed it to update with some piece of information, so I worked with Gemini. I gave you the original assistant instructions, said here’s what we’re working with today. I need to make these three changes and let’s build this together. And then you can turn on canvas mode, which allows it to create a document, right? So it builds out a brand-new version of the assistant. I read through it. I didn’t just take it and copy and paste it in there. I read through it and right away I spotted two things. I was like, oh, we need to adjust that. So I go back into the chat, I say, we need to adjust. This gives it to me. So then I copied and paste it in the instructions.
Once I was happy with it, and now I’m testing it, I’m using it, and I’m not just using it to do the task, I’m paying attention to. Is it as smooth as I would like it? Is it doing the things that I need it to do in my case? For instance, I always have the AI. Once we’ve settled and collaborated on the topic and the questions and everything, it writes a custom introduction for the episode framing the topic, and then it writes a custom introduction to the guest. They may give me a bio, they may not give me a bio, I don’t care. I want to read something about them that’s unique, that really hypes it up. That’s in my instructions, and I’m using Claude to write all these things. So when you’re on the show, most people are like, “Wow, Mike, that’s really the best introduction I’ve ever heard. I need to bring you with me everywhere.” And I say, Claude wrote that.
But something funny happened when I redid the instructions. So in that first paragraph where it’s framing the topic, for some reason it went on to introduce the guest, because I had two guests to prepare for this morning. And then it still did a second introductory paragraph as though we didn’t mention it in the first paragraph. It just wasn’t quite connected. It certainly wasn’t smooth, easy for me to edit out. But I don’t want to have to edit that out every single time I create a new podcast. So I’ve got to go back into the instructions and figure out, okay, where do we go wrong? Let me clarify. This paragraph is for the topic. This paragraph is for the guest. That’s something you can do at any time, and I think it’s easy to forget that once we set these things up, they’re not written in stone. It’s just a text instruction, they’re really easy to edit.
Rich: Yeah, and I would agree. And I would say that for me, I’m really careful about what I put out into the word world from a written word standpoint. So when I’ve trained up AI to write with me, I never consider it more than a rough draft or a first draft. That’s just me. I’m not judging anybody who puts it out there.
But when I get the outputs, because I don’t know about you, I’ll watch a lot of YouTube videos about the perfect prompt, or you do this and you’ll never have problems with ChatGPT again. And I’m like, oh my God, that’s brilliant. And I’ll just do it. And then I’m like, but how do you know it works?
So for me, it’s a lot of times going in, testing something out, getting the output and comparing it to the previous version. Is it actually any better? And sometimes I have it write a blog post in my voice for my show notes. I’ll always read that through, and I’ll make the edits that I would make as if somebody else gave it to me or if I wrote it yesterday and I’m editing it today. And if there’s enough significant changes, I’ll go back to, in this case Claude, and be like, “Here’s my edited version. Compare that to the version you wrote. What changes do you see?” And it will sometimes summarize the big changes.
It uses the word “actually” all the time. I’m like one time, you can use it one time. And so I started to have to put extra controls in there too, because it was trying to sound like me and it almost becomes like this increased version of me or that every single blog title ended in “and how to fix it” in parentheses. I’m like, you can do that once a year. Once a year. You’re not allowed to do that more than once a year.
These are some of the issues. Sometimes I’ll ask Claude or AI, “How do I fix that? How do I avoid that next time? Should we rewrite it?” Sometimes it’ll say, “This is a one-off, I wouldn’t worry about it.” Other times it says, “Try doing this in the instructions and see if it makes a difference.” And so I think we have to be willing to put in that time to iterate, to get better responses, to save us time on the other side.
Mike: Oh, I couldn’t agree more. A fun example these days is, people will say if you are using an M dash in your content, that’s a red flag for AI. And interestingly, in my content it would be. I don’t use the M dash, I use space hyphen space. And I don’t know why. I just, I always have, I don’t use that long hyphen. But I had to tell the AI don’t use the M Dash, that’s not me. I would never use that particular form of punctuation.
Rich: Yeah. And for me, I’m like, you can take my M dash when you peel it from my cold, dead fingers. And don’t even get me started on semicolons.
But yeah, I totally hear that. So I do think that there does have to be, you have to be willing to put in the work, the iteration, and you can’t just hope that it’s going to put out the best content. You have to be looking at that output and providing some sort of feedback to it.
Mike: I think you may also find, I know you have, is that some models are better for certain tasks than others.
I swear by Gemini. I love Gemini. I’m in it every day, but I just can’t get the writing out of Gemini to be as good as Claude. No matter what I do. I’ll take a draft outta Gemini, I’ll share it with Claude. Claude will say, “You know what? That’s a pretty solid foundation, but here’s six things you could have done better about that.” And it is always, always better. I’ll take the finished product out of Claude, put it back in Gemini and Gemini’s like yeah, that’s great.
Rich: Yeah, I know, I find that Claude is better for writing, but ChatGPT is better for staying within the parameters that I laid out for it. I tend to prefer the way that Claude writes, so that’s why I use it for that blog post. But then it never really seems to nail down the length of certain things that I have. I think sometimes you just have to be willing to say, this is a great rough draft or a nearly finished draft. But ultimately, I still feel we have to be involved in all these sort of things.
Mike: A hundred percent.
Rich: So one of the things that we skipped over, and could be an entire other topic, is how do we decide what we want to offshore or delegate? You have an audit on your website, I believe. Talk to us a little bit about how that audit fits into determining what we should be outsourcing.
Mike: This is such an important question. Because what I want people to think about is the fact that AI isn’t going to solve everything for you. It can’t. It’s not a magic bullet for your business. And if you’ve got messed up processes in your business, you’ve got to sort those out first before you should ever start to apply any kind of automation.
And I say “automation” very specifically, because you might just need something like a Zapier connection to a couple tools to start talking to each other, moving information back and forth, which is not AI at all. It’s certainly not an AI agent, it’s just an automation. We’ve had that for years, but many of us either A, weren’t aware of it, or B, just haven’t taken the time.
The scalability audit is just a tool. It’s a worksheet. You can get it for free at theaihat.com/scalability. Or not. You don’t need to. You just need to write down for a few days, what are you doing? What are you doing with your time? How are you devoting your precious amount of minutes every single day to specific areas of your business and what are they? And then after you make that list of what you did, how much time you spent, then you can start to bucket it. Was this something that actually moved the needle for my business? Did this thing, this task drive revenue or is it the opposite of the expression? It was just something that I did.
But I’m finding, I’m doing this every single day, replying to emails other kinds of routine, boring administrative stuff. I call it an admin tax. And the smaller your business, the bigger those admin taxes usually are, and they’re really dragging you down as a business, particularly. Again, the smaller your business, the more your business depends on you, the individual, or you, the small business owner, to get really impactful stuff done. And if you’re being weighed down by responding to the same FAQ every single day, that’s got to be solved.
But there’s certain things, like maybe you’ve got a membership community, and they sign up on your website, and you get an email notification, then you need to go put their name in their email address in your membership community. And then you also need to give them access to your Visio repository. And then you also need to remember to add them to your monthly calendar where you’ve got your group coaching. And oh God, forget about it if they cancel, because those things are never going to be updated. They’ll just live in your community for six months for free. And you’re just okay with that because it’s five or six moving parts. Five or six manual entries that you never connected the dots to. But if you just jump into Zapier, you could probably automate 90% to 100% of that. And when you make a sale, now it’s, “yes, I got a new member”, rather than, “oh, I’ve got six things I have to do now and spend all that time wasted doing those things.”
So that’s the first step, and I think it’s the most important step is to do that audit. And honestly, the hardest part is not to judge and try to fix as we go. Don’t do that to yourself. Just write down for three, four days, maybe a week or two, depending on the nature of your business. If every day feels similar, a few days is fine. If you feel like every day is totally different, then give yourself a week or two to just nonjudgmentally write down what you did, how much time you spent. And then you can go back and look at those audit results and say, okay, this is where I’m spending the majority of my time that’s not helping my business. How can I streamline this? That’s the question. How do I streamline this or delegate it? Or just eliminate it completely. Maybe AI is the solution, maybe it’s not, but you’ve got to do that first step.
Rich: Awesome. Alright, Mike, if people want to learn more about you, The AI Hat, or any of the stuff you’re sharing about artificial intelligence, where can we send them online?
Mike: Best place to go is the aihat.com/connect. When you go there, you can learn more about me, you can follow me on your preferred social channel, if that’s what you want to do. You can learn more about the podcast or the community. Whatever you want to learn about, you can find on that one URL.
Rich: Awesome. Mike, thank you so much. Good seeing you again, my friend. Appreciate your time.
Mike: Thanks for having me, Rich. Appreciate you.
Show Notes:
Mike Allton is an AI strategist and creator of The AI Hat, where he helps solopreneurs and small business owners design AI-powered systems that save time and scale impact. A longtime marketing leader and author of The Ultimate Guide to Social Media Marketing, Mike blends practical expertise with a human-centered approach to technology. Tune in to his podcast, The AI Hat Podcast, where he helps make AI accessible for everyday business owners.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine, and founder of the Agents of Change. He’s passionate about helping small businesses grow online and has put his 25+ years of experience into the book, The Lead Machine: The Small Business Guide to Digital Marketing.