#345 – Our AI Agent Made Us 6 Figures

Never lying to the audience is one of our core values.

And I don’t think you can say we really lied, but we haven’t been 100% truthful with you lately.

And today we want to come clean.

For our defence, I have to say, what we did was more of a social experiment that you get to benefit from today than a real deception.

But before I tell you what we have lied on, I have something to say.

Every time I open YouTube, I am buried under a sea of hyper videos about AI that will tell you how to make money with it.

I get excited, then I click and start watching, and all I see is a basic tutorial with half-assed prompting, prepped in five minutes, and there’s no way this kind of stuff could ever be used in a professional setting and get good results.

And for some reason, many people seem to love this kind of content, even though it’s mostly spam, and even though that kind of stuff will get you in trouble in the long run if you use it on a project that you care about.

And if I have to be perfectly honest, this kind of content doing so well has discouraged me from talking about how we use AI publicly, mostly because I did not want to be assimilated with this kind of stuff that promises a lot, delivers very little, and promotes spam, which is everything we hate.

But that does not mean we haven’t been pushing AI forward inside our company.

Quite the opposite, actually.

While many are cooling off on AI, we’ve been massively expanding its use in many areas of our business because the models are quietly getting better over time.

But I knew I didn’t want to be that guy that overpromises and underdelivers when I get to talk about it, so I had to put a proof first by using it in a real professional setting so I could show you that what we’re talking about actually works, unlike most of the AI content that you find on social media.

And that’s exactly what we’ve done.

And now comes the confession.

If you’ve taken part in our recent Product Review Profit launch, which is the course that we released recently featuring Jared and teaching you how to scale the Amazon Influencer program, which hundreds of you have bought, well, most of the content you’ve been through during this launch was AI created.

More specifically, the sales page and the sales emails to our list. 80 to 90% of the words that you saw there were written by Claude 3.5 Sonnet, and these words generated 6 figures of revenue in just over a week for us.

And that is the proof that most content creators never give you when they share their AI system on social media.

We’ve created AI content so convincing that it’s not just invisible, as nobody called us out during the launch, it also gets people not just to visit your site, but also convinces them to the point where they swipe their credit card on your site.

And we’re not claiming that maybe it can happen.

We’ve done it without telling anyone to prove our point.

And that’s why we had to lie to you, because we had to see how robust the system was in a real setting with lots of money at stake.

Some kind of marketing Turing test, if you will.

And in this episode, we’ll show you exactly how we did it along with other game-changing applications of AI inside our business.

But before we stop, let me thank today’s sponsor, Search Intelligence, the top tier digital PR agency that can cover all your high-quality links need.

More about them later.

For now, let’s get started.

All right, so let’s jump in right into the episode.

And we’re first going to talk about how we’re actually using AI in the past few months, because it’s like we haven’t talked a lot about it.

We’ve been quite critical of AI, especially when it came out, especially in the way people have been using AI.

But the reality is, while we haven’t talked a lot about it for the past few months, we really ramped up AI users.

It’s kind of funny, like the market went from super excited about AI to many people not giving you shit anymore.

To now, for us, we’re like, well, it’s not really good enough for production.

To actually, there’s a lot of use cases for AI inside our company.

And how do we use this to…

I’m on a quest to basically double the productivity of everyone in the company in various ways, and we’re going to talk about that.

And while we have been using it to create content for our blog or to help us create content for our blog, it’s some of the other ways, particularly around our sales for a recent launch, which I think are the most interesting, or certainly the most financially beneficial to us.

So do you want to maybe talk about how we use it in the launch?

Yeah, that’s the thing.

I think it’s like the blogging is literally the shittiest use case of AI in my opinion.

Like, it’s kind of useful, but it’s not the best way to make money.

I think, let’s start with a big reveal that we kind of didn’t tell anyone and nobody noticed, apparently, is that if you’ve been to our product review profit launch, which is the course that we launched with Jared a few weeks ago on Amazon influencer, which has gone really well, a lot of people are doing well, etc.

Well, the interesting part of that launch is that most of the sales page and most of the email, pretty much all the emails, were written by AI, actually.

It’s like I written edited by, like we edited the AI output, but the initial draft for pretty much everything you’ve read that was marketing material, minus the ads, the ads were not written by AI, was AI written.

But this wasn’t just your let’s put in a prompt, hey, write me an email selling this product.

A little bit more complicated than that, right?

Yeah, yeah, yeah.

So it’s like, let’s talk about that.

But like, yeah, it’s like, just want to say, like, most marketing, and this launch did well, by the way.

Like, it’s not like it’s not like it was like a downgrade or anything.

And that’s something that I want to say with AI usage, like, I will only implement AI in our company if it’s actually making us do better, all the same.

Like, I’m okay with the same if it saves us time.

But like, otherwise, like, we’re not downgraded anything for AI.

So anyway, sales page, we put some bureau over here is was written by AI, everything on the sales page and all the emails.

And so it wasn’t started with AI, it was started with like first working on the launch itself.

It’s like, you know, we made a big document of like, you know, what are the problems people are trying to address when they buy this kind of product?

It’s like, you know, they want a quick side hustle.

You know, they’ve been dealing with Google issues.

They all the kind of like trouble that that people wanted to do, like, they don’t like building your website, technical stuff is difficult.

All the objections that we’ve kind of heard over the years in our support, when in our feedback, when people comment on our ads, like these types of things.

So we kind of pulled it all together into a document, essentially, and used that.

And also, like, we also in a document, we also added like how the product addresses these objections.

For example, if you’re making review videos with your phone, well, you don’t need to build a website, there’s no complicated take.

If you post videos on Amazon, you don’t need to generate traffic, things like that.

And also arguments like, you know, the RPM, for example, that is like one of the highest compared to like YouTube or something.

So we made a big document with lots of information about who it’s for, what are the problems people are having with running side hustles online, and how essentially this side hustle answers a lot of these issues so that we could ground AI when we’re actually writing stuff.

So that was step one for me.

It’s like me, myself, as the operator, understanding what actually we’re trying to solve and what we’re trying to solve.

And AI, I can brainstorm with it.

I had some chats with it where I kind of like, it’s more like a soundboard, you know, where you just kind of like give your arguments and counter arguments, etc., to try to help you voice these things.

But the thing is, like, I’m good at figuring this stuff out, but like, I’m not a non-native speaker, right?

So it’s like, it was, it’s sometimes difficult to articulate in a way that makes people excited, for example.

And that’s why AI was like extremely useful.

So I’ve actually built, I’ve actually built like a closed sonnet.

So I use sonnet, which is significantly better at writing than than chat GPT, actually, like a really, really a lot better.

And I built like what they call a custom project, but it’s similar to a custom GPT, right?

I’m going to share my screen with you so you can actually see.

You guys can see as well if you’re watching the video.

If you’re not watching the video, I’m just going to describe what you can see.

So what my essential project work like, work as is like I first wrote a system prompt.

A system prompt basically tells it, you know, what it is, your copywriter, you will provide it with input, like a test description, and then you will get like documents, company information, SOF file, etc., and how to sync and then generate, like, generate copy.

And then, yeah, you know, it has a bunch of best practices and I give it a lot of documents on how to work.

So the first thing I do is I do a swipe file.

This swipe file has a fucked on of high quality sales pages that I curated myself that I think are good.

Basically, it’s like, there’s no way to automate that.

It’s just like you need to kind of like keep an eye out and build swipes.

Those just copy, you copy the whole whole page and paste it in word for word.

Yeah.

Okay.

Yeah, you can see like the file is like it’s pretty messy actually, if you look at it.

It doesn’t matter.

The point is that it learns the language of good sales pages that I like.

There’s a big text file with that.

I wrote a text file was like about the company.

So like core values, who we are, things like that unique selling points.

And I use the AI to generate writing guidelines.

So we worked on, you know, who is our target audience and then how to write, etc.

But again, that’s a chat I did with AI that generated writing guidelines.

I’m not going to give them to you guys because they’re different for every company.

What you should know is like what you should put in there, basically.

And now what I can do is like this AI trained, I can literally take anything and I’m like, oh, let’s generate a PAS.

So that’s problem agitation solution for a product teaching you how to start a YouTube channel.

Like that’s a very generic prompt, right?

But because it has all that training data in there, it’s actually probably going to give me something that’s pretty relevant that I’m probably not going to copy paste directly into a sales page.

But because I built that project, it’s a good writer, I should get something decent.

You can see it’s pretty slow, actually, which is not the best, but let’s see how it goes.

So you can see that the copy is coming.

And you can see that because it learned from my sales pages, it’s actually like it’s doing a job of like a pretty good copywriter, like better than maybe someone I could hire an upwork, like attention YouTube creator.

If you think hard work, a nicer camera or hacking the YouTube algorithm will get you more views and subscribers.

Here’s why you will never walk after helping over a thousand YouTube creators care, look at their channel, read below and learn a promise strategy to highly engaged YouTube channel with more views and subscribers in just 30 days.

It’s actually good.

It’s a little bit cheesy, but I’m pretty impressed overall.

And the reason this works versus just like if you had just put that prompt in with without any of the context that you gave it before, you would have got garbage basically, 90% of the time.

So the reason this works is because you’re giving all that background information.

And to be honest, if you’re working with a high level copywriter, they’re probably going to make you do this anyway.

So it’s not like the workflow is dramatically different when you’re working with AI in this context versus working with a professional.

Exactly.

It’s like anyway, they would ask you examples of sales pages you like anyway, they would ask you like what tone of voice you want to use anyway, they would ask you all that stuff, except I get my results right away.

So you see, I put the same prompt into cloud, but without the training and I’m getting something that’s much more generic here, something I would not necessarily be very happy with, but like using my sales page copywriter, if I go check this, it’s like this result is way way more usable.

And we probably generate way more sales.

So essentially, that’s one of the first uses we have.

We use this as a copywriter and then just basically tweak it, tweak what we get into sales pages.

But we use that for email marketing the exact same way.

I built a swipe file of high quality emails.

And after that, we actually made a plan, actually made a plan of like a schedule for emails.

I wrote some bullet lists of like what I want to say in each email more or less.

But like I give it five bullets, I write me an email that’s super well copyrighted.

If you want to see samples, go check in your inbox for the emails we sent it during the preview profit launch because pretty much all of it is AI written.

And I actually AB tested some AI written emails against my emails that I write.

And most of the time, the AI actually did a better job.

I actually won against against my copywriting, which was pretty impressive.

And this is one of the big benefits of being able to rapidly create these types of emails is you can AB test till your heart’s content.

We didn’t always do that in the past and previous launches.

And I remember, especially the one like just after new year, you were like working furiously between Christmas and new year trying to get all the emails.

And it’s like you’re in that mode, like nobody should disturb Gael right now.

He’s focused on the emails.

But now it’s like, oh yeah, I just generated them.

No problem.

But yeah, it goes really fast.

So it gives you a lot of agility, I guess, in your business.

Even Jared, right?

It’s like at some point, Jared needs some help with feedback on his emails.

I was able to just take his stuff, throw it in there and just have it fixed a lot, actually.

I don’t know if he used it, but like it was it’s so easy now to just implement this.

And then I just need to edit the files about the company or something.

And I can implement this somewhere else.

But yeah, the AB testing so powerful, like, for example, let me show you another example of how we use this AI.

So this is OptinMonster, right?

OptinMonster is like the optin popups we run on our website.

That’s how we collect a lot of emails.

And it’s like a fresh batch of popups that we’ve launched yesterday.

Like if I refresh it, there’s probably going to be more optins.

But the idea is that we should we know we all should do AB testing stuff, but it takes time.

And it’s like, I’m not necessarily very creative.

Like, I don’t know what headline to test or whatever.

Like, it’s difficult, right?

With AI, with this same copywriter, I’ve been able to generate lots of variations of our popups.

So you can see, for example, here we’re testing three different ones.

And you can see the conversion rate is a big difference.

So you can see this one is at 2.79% optin rate.

While this one is at 1.9.

While this one is at 1.89.

1.89 to 2.79 is actually a big difference.

Yeah, like 1.89.

So like that’s a 47% increase.

So like just from like quickly running this, and I literally throw a screenshot of my pop up to Claude, to this copywriter that knows how we write, I’m like making variations.

And then I pick the ones I like the best.

And same with this one, for example, look at the mobile one.

4.96% optin rate, which is pretty good.

3.28, but look at this one 5.94.

For pop up, that’s really impressive.

Yeah, versus 3.28.

And it’s like, you know, it’s not a ton of numbers yet.

But there’s still like 38 options already.

Like, it’s gonna move, but it’s like, it’s gonna win, I think this one, you know, but like, you’re looking at almost double the optin rate by using this stuff.

And it’s like, yeah, it figured out that this headline works best on mobile, for example, and it’s super interesting.

So like, that’s one super strong use case for AI.

What would you say, sorry, what would you say if you have to quantify it was the financial impact of doing this in this launch?

How much extra revenue did this generate?

It’s hard to tell because it’s like, you know, the way people buy, it’s like, it’s not always the last touch.

It’s like they get convinced over time, etc.

What I know is that over time, like, for example, these pop ups, like they will have a big financial impact on the business, because we’re gonna collect like 20 30% more email, like I just need to set myself a task every month to just check the A/B test.

And then just generate them in it’s a it’s a 20 minute task to do that.

It’s like, eventually, we’ll need to test designs, and it’s a bit more work and so on.

Like, if you just test like this is just testing the headline, basically, like it’s so fast, with AI and then you give you the screenshot, and it does it.

But again, if you just choose a generic AI where you don’t train your agents, you’ll get generic results.

This is the same copywriter that wrote the sales page that did the the variations of this headline.

And that’s why it did it so well, for example.

And again, like the way I actually got it to generate relevant headline is I gave it the transcript of the lead magnet video people opted for.

So it knows what comes in the next step completely.

And it extracts interesting beats in there.

And it generates that A/B test for you and generates like very relevant, very relevant titles.

So eventually, when I stop getting gains on the titles, I’ll jump onto the bullets and then I’ll jump onto the CTA text.

And after that will A/B test like completely different designs.

That’s when it will take a bit more time.

But like I can see, like we could easily double our art in rate for something that used to take a lot of mental energy and now it’s just one prompt away, basically.

So that’s a really, really, really strong result.

I feel like there’s just a business here, you know, you could do like a CRO for opt-ins like at scale using by building these kind of AI systems.

It’s quite interesting.

Yeah, it’s so powerful.

I’m gonna stop sharing it, but you get the idea.

Like this is one of the main things.

It’s like for sales funnel.

It’s insane.

Like nobody cares.

Nobody tries to detect your AI content when you’re doing a sales funnel.

So it’s completely fine if you are actually just using AI provided the content is quality.

And in this case, like, you know, tens of thousands of people have gone through this launch and this funnel and see these pop-ups, et cetera.

Nobody is asking if it’s AI generated or not.

And nobody ever mentioned that throughout this launch.

It’s not an issue provided you train quality agents on Europe.

So that’s the first use of AI.

So the thing is like, that’s how we use this for product launches, which is cool.

But we also use the same process kind of for the news channel, for Autoiac and news, actually.

So if you’re watching Autoiac and news, which is our other YouTube channel, it’s doing okay.

The first version of every single script that we read and end up making a video of is actually AI generated.

Now, how does the process work?

Again, let’s talk about the process.

Again, it’s manual first.

Like you cannot just go to the AI and expect it to make a great video for you.

You need to first, we have a news collection process.

We use a tool called Readwise Reader that is basically curating all the best sources that we have decided to put in there.

We can flag them.

We can take notes.

These notes go directly into Notion and that goes into building an outline.

Outline process that, by the way, is the same one described in TAS and Autoiac Pro.

So if you have these courses, this is still useful, but except you can use AI to work with you right now.

Once we’re done with that really big outline of bullets, we give it to AI, again, pre-prompted agent.

It’s prompted on our previous scripts, prompted on everything.

It gives us 80% of the words we say on camera, I would say.

It’s about that.

The last 20%, we just edit it, we shoot it, and that’s it.

We have a YouTube channel that is partially powered by AI.

We still do work, but it saves a lot of time.

It’s a huge time saver, this one.

Yeah.

Yeah, it’s great.

Some of the editing is done by AI too, actually, like all the retakes, et cetera, we actually use Glink on Premiere to actually do that as well.

AI powers a lot of the news channel, actually.

I want to stress though, because I know there are people out there that will be like, “Oh, I can just use AI one click, get everything I need, and then just speak it out, read the script without editing it.”

It doesn’t work.

Every time there’s several layers, two or three layers of editing before it goes out.

So you are ultimately that last line of defence of quality, and it’s really important to keep that.

I think that’s kind of how it works.

AI is kind of in the middle of the process, taking some of the time taking stuff.

It’s a good writer, but it has nothing to say, basically.

It has nothing interesting to say unless you feed it to them.

So it’s like, you need to do the first step where you do the outlining and you come up with all the cool ideas and the cool stuff you wanted to say, then it’s going to use really good words to put it together.

It’s a really, really good writer, much better than pretty much any finance writer I’ve ever worked with, and you get the results instantly.

Then you need to still edit because it’s going to have a lot of fluff and stuff that bloats your content in the final output.

So you’ll want to re-edit that, but provided you put that amount of work, it’s an amazing work partner because there’s no delay as well.

Then you can scale it up.

You can make many variations like the A/B test, et cetera.

It’s awesome.

Now it’s time to experiment with that same process with the blog, actually.

I know it sounds a little bit scary.

Again, we put a lot of manual work into these, the outlines, et cetera.

We put a lot of work, but you can see an example.

We’re still experimenting now.

For example, we have the latest blog post on the blog now.

It’s hard to get paid for Amazon reviews.

Then this is actually written by AI, but it’s full of examples.

It’s full of real-life case studies, et cetera.

The reason why is because we’ve put that into the outline.

We’ve actually fed that to the AI.

It didn’t come up with it on its own.

As a result, the blog post is decent.

It’s starting to rank now.

It’s mid-ranking.

It can do better.

We might need to optimize it, but overall, the process is sound.

We don’t think the content is lower quality than when we were Swiss finance writers, to be honest.

Question for you then.

What’s changed?

Because when we tried to do high-quality AI content in the early days of chat GPT, it just wasn’t really quite there yet.

Is it just the new models that have gotten better, or is it our understanding of how we use it?

One is context size.

Context size is how many tokens the model can have.

That basically says how much instruction you can load in for it to output.

If you output a swipe file, the swipe file is 30, 40,000 tokens, 50,000 tokens, something like this, which is all the examples.

Then the writing guidelines, then the value compounding, et cetera.

We’re using 60,000, 70,000 tokens of just files for input.

Early chat GPT, the first chat GPT was 8,000 tokens.

You could literally not input that information in there.

Now, GPT 4.0 is 128,000 tokens, close 3.5, and all cloud models are 200,000.

Gemini 1.5 Pro from Google is 2 million tokens, actually.

You could literally input books in it.

I would still be able to retrieve all of it.

It’s like that’s the first thing.

The second thing is, yeah, the models are smarter and better.

3.5 Sonnet is a big, big upgrade.

Most people don’t really …

I mean, AI people talk about it, but most people that don’t really follow this too closely haven’t realized, but it’s a really good writer, it’s a really good coder.

It’s like that’s pretty much it.

Then obviously, as time passes and these technologies are settling in, we’re learning how to use them better, how to instruct better.

For example, I use the Cloth has actually a prompting tool, which I can share.

I’ll show you this quickly, and I’ll show you an example.

The thing is like this tool, it’s like you basically go on console.ontropic.com, you can make a free account and it’ll give you five bucks of API credit that you can use.

You have this screen basically, and you click on “Generator Prompt.”

It’s basically an AI that prompts AI very well basically.

For example, when I write a prompt, I have this thing.

I give the role, “Where are you?”

I give the goal, I create high quality copy that’s engaging and boost sales basically.

I give it how to work first with my prompt and analyze documents in a knowledge base, then the system prompt, then brainstorm how to execute the task, then execute the task.

I tell it the output format, right?

I put the copy, then the editor knows that on how you tackled all the instructions with your output.

That’s an okay prompt in terms of structure, but it’s not very well written.

If I click “Generate Prompt,” for example, actually this tool now uses all the best practices to prompt, and it’s generating a prompt that actually is going to generate much higher quality responses.

Now we have AI prompting itself basically, and that’s one of the tools that I use to generate system prompts for agents, for example.

That’s the stuff that has made things better, I think.

And now for a quick word from the episode sponsor, Search Intelligence.

How much do you think this digital PR campaign is worth?

Wait just one minute before you answer.

Watch as we hijack the news with Taylor Swift and Barbie to get you the links.

Now we can take this dataset from ONS and paste it right into your emails to get you the links.

Now how much do you think this campaign is worth?

Don’t answer.

Wait until you see Bloomberg, BBC, Forbes, Top Tier National, original news outlets, React PR, Data PR, Expert Commentary, over 70 team members at your disposal.

This product is so hot I need a fan!

And then listen to what campaigns we do at no extra charge.

Gain character wages, bananas and sleep, best days for dogs, luckiest UK areas, and many more.

All for just how much did you guess?

10,000?

20,000?

Even more?

No!

It’s just 5,500 pounds.

That’s right.

It’s five and a half grand.

It’s an incredible value but it’s true!

It’s digital PR from Search Intelligence.

Order today!

P.O. Box search-intelligence.co.uk.

Accept B2B clients.

Thanks again to the sponsor of this episode, Search Intelligence.

If you’re looking to boost your link building with high quality digital PR campaigns, head over to search-intelligence.co.uk.

And now back to the episode.

Are there any resources or YouTube channels that you go to kind of like learn about this stuff or how do you know it?

I play a lot with it.

Like I have had inspiration.

So for example, like Marketing Against The Grain is like a podcast that like talked a lot about AI around marketing.

That was pretty good.

But I’ve heavily deviated the stuff I’ve learned from there to what we do now.

It’s not what we talk about.

Like yeah, it’s like it will take practice.

Eventually pick a project, learn how to prompt and it’s like there’s no technical skill needed here.

Just clarity and new explanations.

Right.

Let’s talk about some of the problems with with AI content in general because there are certain stigmas out there and I think the way people approach or think about AI content or prompting is at times a little outdated.

Have you heard of this saying the internet is a mile wide and an inch deep?

I mean I’ve heard the mile wide, inch deep thing.

But yeah, it’s like when I remember when we first started in like 2010, that saying was going around and it kind of feels like since AI content came out, the internet has gotten 100 miles wide but still only an inch deep because people are just churning out lots and lots of content without too much care or editing or anything like that.

And there’s this kind of weird dynamic where as a creator, a tool creator or an AI tool creator or someone who has a tool already, the ones that incorporate AI into it, if you go for like the lowest common denominator type of content or use cases where you’re teaching people how to make $100 a day with chat GPT in six seconds or something like that, then you get more notice and people watch that, people buy your tool, like they want the quick hacks but it’s ultimately not doing them any good in the long run.

It’s not doing the internet any good in the long run.

So I feel like we kind of need to break out of this cycle somehow because everyone’s convinced that this is what everybody else is doing successfully but I’m not sure that it is.

It’s the same for like you were asking me like what kind of content are you consuming to learn this stuff?

I mean it’s the same like most quote unquote AI videos on YouTube etc. are their whole shit like it’s always the same stuff.

There’s nothing new, it’s boring, they just show you the very very basic stuff, they never get into the detail of prompting properly or really like they tell they teach you how to build a giant social media thing that’s going to do all your posts but they won’t show you an output because the output is so terrible that it’s like there’s no way someone’s going to engage with that content and that’s my problem.

It’s like I think the problem is just the kind of people who engage with this.

There’s basically two types of people right now.

There’s the people who see AI as a quick shortcut to do things without putting any work in and these people are very much into AI, that’s most people.

And then there’s the people who kind of like work for high quality publications like New York time whatever big editor etc. who are like fundamentally anti-AI because quote unquote it’s low quality and it’s kind of like threatening their way of life.

And I like to believe that we’re not in either of these sites.

We like high quality but I want to adopt this stuff for high quality and I can see how like the fact that we did this launch blind and not tell anyone for me was like a real Turing test.

It was like how do you know you’re talking to a machine or not talking to a machine.

It’s like yeah you’ve literally been interacted with machine content during this whole launch and we got many people to swipe their credit card doing that and to me that’s like the ultimate proof that this can actually generate quality enough provided it’s done properly.

So yeah and that’s the problem.

It’s like most of this content that people create like most AI tools are shit.

Most one clicker article creators, most article optimizers, most SEO AI tools etc. they’re completely useless.

That’s why you’re seeing me not recommending anything like I could be making lots of commissions on this podcast recommending some tool you know.

My recommendation is don’t buy any tool.

Build your own agents learn how to prompt and then build something that’s so custom to your company it feels like your voice.

And I think that’s what the big hit was with this launch for example.

How you visit the blog post you guys get Google check it is that it sounds like this and because of that people don’t notice.

So yeah I think that’s a problem.

Do you think there’s still a big stigma around work that’s done by AI?

I mean if you’re a writer for example you have to pay a premium to get human level content.

Society isn’t really at that stage where it’s just focused on the output it still cares about how it’s done and it values it differently.

They care about the effort.

I mean think about it right.

Let’s imagine I launch a service I’m like you know what let me do your funnel with AI.

I’ll generate your sales page, I’ll generate your sales email, I’ll generate your ads.

Pretty easy at this point like duplicating what we have already.

It would probably take me like one or two days per business but like could generate lots of money right.

If I sell that as a service pretending I’m doing it myself with my reputation who I am etc.

I can probably charge lots of money potentially percentage of sales or whatever.

If I tell people I’m just going to use my AI but to generate most of that content how much are they willing to pay probably a lot less than if I tell them I do it manually.

I guess it’s the same as you know pretending you’re the one doing it but actually farming it out outsourcing it to somewhere offshore.

It’s always been the case.

Yeah it’s just now you can you can do it with AI and the power of AI for me even more than the cost is the speed.

Like when I give a task to a freelance writer I usually wait like a few days before I get something back.

So the feedback loop is so slow it’s very difficult to progress and your head is somewhere else and something has happened etc.

With AI it’s like you do it instantly you can just deep walk with it for 12 hours if you want and this is always here.

That’s actually a good point because we’ve experienced some kind of like irregularities and randomness I guess with even with using these models that you’ve built but you get that as well with humans right even more.

Yeah they get sick they’re just like their motivation deeps all of that like all that stuff just is just not happening and but like the speed the fact that I can keep my brain in the task get the output then just be like okay no it should be like that no it should be like that like it’s so difficult for me to jump back on cell space copywriting you know two days later three days later after I’ve worked with someone and the context switching the cost is so high whereas I can do everything in one go in one session I could generate an entire sales page which would never happen before whether that’s me writing or whether that’s someone else doing it.

So yeah I think there’s a stigma around AI but the stigma is because of all these lazy people who want to one-click generate stuff that generate 99% of AI content because they can do large volume because of the low level of care and so like the one percent of people who do a good job they’re almost invisible and they kind of want to be invisible because they can preserve the value of their work that way whereas people would be willing to not pay as much money for the output if they knew it was a generator.

It’s true like even in this podcast when you’re like yeah and we wrote this article with AI on our site I’m like oh you know how are people gonna feel about that and yeah but people want to see us all right for sure for sure but like the fact that I’m worried about it is yeah I don’t know it’s a problem but like it’s yeah well we put a lot of time into it like the outline of the article it was edited later like it’s not like like it’s not like we generated this article and wanted it still took like four hours to do this article like it’s not like but like it took four hours of a non-native speaker who knows the topic but doesn’t write like very well it wasn’t me and and we got like you know if she added all these human hours it would have taken if there was no AI it would be way more than four hours like usually for one article on the site it’s like you can quadruple that amount of time you know.

You know what’s really interesting about AI content creation is that like the type of people who pre-AI were just giving their writers a list of 100 keywords and saying go write these hundred articles they’re the people who are you know having one sentence two sentence prompts and just not editing it publishing it and spamming basically but when we were creating content or you know still do create content with human writers we spent loads and loads of time like figuring out the exact structure and like how we can differentiate ourselves and what media we can add to break up the wall of text and all these things there’s a lot of thought goes went into it and I feel like now with when you’re working with AI you’re kind of it kind of forces you to better internalize how you want to do things because the best way to learn something is to teach it we’ve learned that by creating all these courses over the years like it’s really helped us to improve our systems and I feel like with AI you’re teaching it you’re essentially teaching someone how to do something but you need to kind of internalize it yourself better at the same time.

Yeah it’s a very smart intern that knows nothing has no experience no world experience or anything but honestly the outcome is like the shit content creators they still shit content creators with AI it’s just like they never made good content they never thought about it and then they are unable to explain it to a model and oh they just don’t have the taste to tell this is good content and it’s like shit in shit out you know and the good people who actually take the time to do this stuff they will create stuff that they will create garbage in garbage out is the is the phrase in English well I like not for this podcast like and and that’s the truth it’s like and the problem is like these these bad content creators they are like it’s kind of like an SEO right there they’re hooked on the tactics like they’re like oh my god like this tactic so good I can build so many backlinks or whatever it’s like it’s not about the tactic is how you execute it it’s like how good is your AI regime or how relevant is it to the person who receives it etc it’s the same problem is that you need to work your standards if you want to actually make money with this stuff because if you don’t like the models are going to generate that content without your help we don’t need you for your shitty prompts if you don’t add value to that process you are going to be erased from the internet three to five years from now and and your capacity of curation your capacity of actually improving the response is your added value to the web and the reason why you still deserve to exist if you don’t do that I’m sorry but you will not last as a content creator and we’re so close like you can see this prompt generator on cloud like imagine if I run that as a loop against queries people make or something and it just it generates a high quality prompt based on a simple query that generates the content for them etc like why do I need these low quality people I’m Google I can do that for them you know so that’s that’s that’s the problem in my opinion AI content can be excellent it can be better than most high quality content creators do it you can edit your content it can write well if you give it the right information but 99% of people will not do that work and I think we’ve proven it by doing a blind launch not showing anyone that we were using AI and nobody noticed the sales numbers are arguably better than some stuff that we’ve done before and and and it just went through and and that was kind of like the social experiment you know so we’ve gone through this process I think where AI first came out there’s all this hype and then maybe it we felt a bit let down actually it’s not so great and then the hype went away a little bit and in the background it’s slowly gotten better and better and better I think a bit faster than a lot of people even realize now what does the future of this look like and how are our processes and systems using AI gonna change when new models come out so there’s one thing that’s still frustrating is like when you put a lot of information into an agent for example quite often is going to ignore some of it it’s just like kind of like cherry peak from your instructions and it’s going to miss like 10 20 percent I think the future models will be better retrieval and just kind of like taking everything into consideration you said you said earlier this week there we might have an issue with kind of contradictory information yeah so so much detail in some of the files yeah yeah I mean that happens too right it’s like it’s but it’s a human problem it’s not an AI problem it’s like as you build up a lot of files with instructions you know writing guidelines brand guidelines swipe files etc some things might contradict each other you might tell people you might tell the AI to be concise and to the point but at the same time to have vivid details that elicit emotions for example like let’s imagine that you could have this too it’s like imagine giving these instructions to a right human writer that they probably won’t give you what you want like because you’re contradicting or they’ll just make a decision on on their own but in this case let’s AI is doing exactly the same yeah so it’s like that’s the thing like I’m actually building a review process I’m building an assistant to help me work on assistant files that reviews all the files and highlights contradictions so we can edit the files together basically so it’s like that’s the kind of stuff that we’re working in on internally there’s no youtube video about that nobody talks about that because it’s boring but that’s actually the stuff that that gets high quality output compared to you know one click make automation to generate a thousand blog posts you know but yeah so I think in the future like probably you won’t need to prompt as heavily you will have better retrieval of information and in general it’s like it’s already a bit like as I said the sounding board like quite often like so for example also when I build an agent I have a chatbot to discuss about the output of that agent and it has the files and everything I’m like oh like for example the AI article writer for the blog it was two verbose you know it was kind of like going not telling a lot with a lot of words and it’s like I literally went to that chat agent I’m like okay well our agent right now we have a problem here’s the answer it’s way too many words for way too little information we just want to get to the point people don’t have time etc it helped me re-prompt the the whole thing basically and so I think like future model will will essentially do a better job at being this partner in building the tools with you basically that’s that’s that’s kind of the future another thing that’s quite important with the AI as well is like you don’t just build content with it you can build software with it now so Claude has these artifacts where you can build simple calculator tools etc and I think like the idea of like just creating worlds is outdated at this point and we’re going to be creating interactive content the same way we’ve been creating blog posts almost and so it’s going to take content creators to kind of like wrap their head around the idea that it’s not just text and images anymore it’s it’s it’s all of it it’s the interactive web at scale for everything generated on the fly almost

So yeah I’m pretty excited like new models coming really soon they’re coming in probably November December we don’t know exactly but Anthropic said Claude 3.5 Opus which is the high-end model is coming out before the end of the year so what impact is that likely to have if we just switch to that it’s slower but smarter basically it’s just like yeah it’s like helping me debug these things etc is going to be faster building the tools is going to be faster the coding is probably going to be better for example so building interactive content is going to be easier like interactive graphs etc dashboards explaining stats and so on like not just just towards describing things anymore like we’re going past that at this point

Alright so that’s how we use AI content and how you should consider using it but I’m curious because we have a lot of small people listening to this podcast how do you guys use it like go on YouTube drop us a comment and give me some ideas like I want to experiment more ,I want to do more stuff.

I hope you like the the crazy social experiment of the launch. It was a bit of a gamble for us to do a launch based on on AI content, but so far so good, I’ll do it again 100%. I’m excited for it. So that’s pretty much it for this podcast we hope you guys enjoyed it and we’ll see you for the next one you.