Aidan  

Hey everyone. Welcome back to The Growth Booth. We are here today at episode 109. Great to have you with us here. Today we are talking about AI, artificial intelligence, and in particular, how we're using AI to systematize and streamline content creation for the content marketing strategies that we use in lots of different websites and businesses that are sort of underneath my umbrella, if you like. So, specifically, today, we can give examples about how we use it for niche websites, but also how we use it for content on SaaS tools, so software tools, and a couple that come to mind that you may be familiar with are Sendpad, you can see that at Sendpad.com. You also can see it in Cartzy.com. Sendpad is probably a great example to look at because you can go in there and you can see the blog content. When you're reading that, what you may not realize up until today is that it's actually a hybrid approach to content creation where we are combining the powers of AI with the powers of human content creation and human skill as well.

I've got our own in-house AI expert with me here today. There's no one else I could think of that I could have on here that would be better to talk about this stuff because Bonnie has been working with content marketing strategies, leveraging AI, and all the different tools that you can imagine that kind of fall under this AI umbrella for probably about 13 or 14 months now, we're going right back to the beginning of 2023 and now in 2024 is we're charging forward. Every day we are sharpening the sword, so to speak, with the AI tools. So Bonnie, thank you so much for taking some time today to be with us.

Bonnie  

Oh, thank you for having me. The AI journey has been a blast. It's been fun.

Aidan  

It's definitely been a blast and it's been interesting. I don't know about you, but I sort of feel like I never know what's around the next corner because every month there's something new that we're able to do. Do you do feel that way as well?

Bonnie  

Yes, there's always something to explore. There's always new tools to onboard with them, tinker with, such a lot of fun really.

Aidan  

Also the way that the existing tools that we're using, and we'll get into more of these and stuff, but the way that they integrate with other tools. So for example, Chat GPT, obviously we've been speaking about this for a long time now, but there are some really cool integrations that you can create using the likes of a Google spreadsheet. You combine a Google spreadsheet with something like chat GPT, and I hope opens up a whole realm of possibilities for all kinds of different things. So it's not just finding new tools, it's finding new ways to use them new integrations. I think that's really been what has made the content marketing strategy that you've been really the architect of in our business so effective. So anyway, I couldn't think of anyone better to have on here today. So Bonnie, why did an AI plus human combo win out as the method that we've sort of settled on for creating content for our blogs? Because we could have gone 100% human, that's what we used to do. We could have leaned more heavily into AI, but what we've ultimately sort of settled on is that that combination of AI + human, so why is that? Why was that the one that that won out in your mind?

Bonnie  

Well, I've always thought that it was very exciting. And it was interesting that the speed of AI relative to the human effort in the beginning, we were able to generate 400 blog articles in one single month. That was astounding. 

Aidan  

That's mind boggling. Just to give our listeners a bit more context, if 400 blog articles, and these were not like 200-, 300-word articles.

Bonnie  

No, they were all 2000, 3000 and up to well, 6500-word articles. And yeah, to be able to generate those at scale was just something that we'd never done before.

Aidan  

Yeah, it's huge saving obviously in terms of dollars as well. Not just time, but dollars, because to create a 1000-, 2000-word article from a writer who's knowledgeable about a topic, I mean, you're paying a lot of money for that. So to create 400 of them in a single month is massive. It's, I think, a testament to the power there.

Bonnie  

We were all also able to use different types of content creator, rather than a sort of a copywriter, we can process- driven, process-oriented people rather than a copywriters themselves. Yeah, 

Aidan  

There's a human element that we are combining up with the AI element that you're talking about here as, instead of perhaps using a copywriter, we've been able to use someone, for example, a virtual assistant in the Philippines, to add a lot of value to that content.

Bonnie  

They added a huge amount of value, and they were able to use their process skills to really leverage AI and help us generate that content so quickly. So that was like the first major exciting thing that happened with AI. We will give the content to a copywriter to edit, to clean it up, and everything like that. But honestly, it was in the beginning, we thought it was really great content, and that was using Chat GPT. That was GPT3, GPT 3.5 in the beginning.

Aidan  

Right, yeah, we're going back over a year at this point. So obviously, the technology has evolved a lot. But what were some of the early problems that you can remember facing when we were creating these processes.

Bonnie  

I think it was the sort of the deluge of people that started using Chat GPT and then GPT 4. Because of all the people using it, we found that there were patterns that identified any writing as 100% AI, like uniform sentences with very similar sentence lengths. AI writing was, in some cases quite fluffy and vague, it used filler words instead of getting straight to the point. So that meant some rewrites on our end as we went along. The rewrites were carried out mainly for like improved audience engagement, and within the contents, just to be able to express things a little bit more in more detail.

Aidan  

The fluffiness, if you like, and these filler words that came in there, do you think that was in part due to the prompt, or the request that we are giving the AI tool? Because I know that, you know, over the past 13, 14 months, that we've done my way, I say we, but you mainly, have taken huge leaps forward and developing the prompts and developing the engineering the way that we asked AI tools to give us the information, but do you think that that fluffiness and filler word aspect was in part due to less sophisticated prompts on our part? 

Bonnie  

For sure, yeah, I mean, it's been such a learning curve. Again, going back to the speed, the speed has balanced that out, so it allowed for us to make those mistakes as it were, and do the rewrites and make sure that we engineered the prompts so that the prompts were more effective, and they were more specific, and they could generate a better output, essentially.

Aidan  

Yeah. What about some concerns, and we've seen this about the content that you get from AI, and let's use chat GPT or GPT for as an example, about it not always being correct? It's not always factually accurate.

Bonnie  

Yeah, I can remember in the beginning, we were working on an Amazon affiliate site, and it talked about it an Amazon product, it was like a rubber trash or it was something, and it made this generalization. You said, "Well, hang on, that's not true." It was about brands, I remember. It made some sort of sweeping generalization that was not wholly accurate and was not relevant. It kind of made it up. They call those hallucinations. So you need to check your facts. 

Aidan  

Well, is there something that you use to check factual accuracy? Or do you or one of the team pull out those claims, if you like that, that we may want to sort of verify? How do you approach that?

Bonnie  

I think that's where the human element comes in this there's not a, you know, Well, nowadays, you could also run a to like Claude 2.1, or something else, some other just to check the relevancy and accuracy. So you can use multiple tools together. But in the early days, I think it was just scan and make sure making sure that we're on the right page, you know? 

Aidan  

I think overall, the efficiency of the AI, the processes that you've put in place, and sort of combining that with the human element has, has made incredible content. And regarding the human element, I think one of the ways that we are able to make sure that the content is accurate is because the humans that work in the different niches that we're part of become experts than their own right in those niches. So let's use Sendpad as an example. The humans behind the behind the robots are people who are now very well-versed in email best practices. So if there's some kind of a claim, let's say that we're creating some content around marketing strategies for email marketers to use during the vacation season, holiday season, something like this, then the team that we've got, because they've been involved in email marketing and creating content around email marketing for some time, would be able to identify and sort of flag "Hey, this is potentially not 100% accurate," and then dive into it more. So I guess, all of this to say that, I think, over time, the human element becomes a source of expertise, and that greatly improves the AI side of things. Would you agree with that?

Bonnie  

Yeah, so the AI, it is still learning. We know, a lot of information about our SaaS tools, like Sendpad and Cartzy, much more than this, so many nuances to them, and the AI wouldn't necessarily know that they have the generic information. Now, Bing Chat can crawl the web. So being an AI, and with the quality of the web, you're limited to the information that is on the website or out in the in the wild, as it were. 

Aidan  

Sure. So in terms of considerations, what are the key considerations in your mind for when the content is being developed?

Bonnie  

Well, this goes back to marketing 101 and thinking about the customer. To generate attractive content, you really need to think about how best to please your audience. When you look at demographics, and psychographics, that sets the tone with engaging content using the correct voice. So you're asking, is it an audience of millennials, or is it an audience of baby boomers? And that will dictate the tone. The culture too comes into play in the creation of the prompt.

Aidan  

So before asking the AI tool to create content, you take into consideration the demographics, psychographics, and you include that information in the request that you pass on to the AI tool. Is that accurate? 

Bonnie  

That's correct. The more specific you can be about that, the more targeted you can be to your niche. With that sort of backbone of targeting to your niche, it's best to basically do a study on your ideal client, first of all, and then pour that into the prompt, so that we can engineer the right output that really speaks to the customer in their own language.

Aidan  

What about content length? We spoke a moment ago about fluff and filler words. Have you found that there's an optimal sort of length for the type of blog content style that we're creating?

Bonnie  

Yeah, well, we found that the length of like 1500 words is really a sweet spot for article blog creation. What's important, as well, in the structure of the article is that you make sure that the content length, the length of each sentence is variable, not fixed. AI can come out with a lot of very patterned sentences, that sort of a familiar AI pattern. We need to make sure that within that 1500 words, we have the right tone, we have the writing engagement to the audience, we have the right sentence structure, so it's not AI and, again, engagement is everything.

Aidan  

What about the relationship between the content that is written and the requirement for it to be sort of optimized for search engine optimization? In most cases, you know, we are creating content for our existing email lists, our existing audiences, but we're also using this content to tap into new audiences. One of the ways that we're doing this is through search engine optimization. So you know, we are creating content that gets ranked in Google. What have you learned about sort of balancing the SEO component there?

Bonnie  

Yeah, well, there's a fine balance between the integrity of the article and making sure that it's keyword optimized. When we originally start an article that basically, what we're focused on is really what people are searching for in Google, so we look at Google search, and we look for the titles that are ranked highly. Then we go to Semrush, and we look for keyword optimization. We can also look to Google for that information as well. There's a lot of behind-the-scenes work that we do, technically, if you like, before any content is written so that we're on the right track to start with and we're navigating properly.

Aidan  

I hope that one of the takeaways that our listeners are getting here today is if you want to leverage AI to help you with content creation for a blog for a niche website or anything like that. It's not enough to simply go in there and say, "Hey, ChatGBT give me a 1500-word article about setting up a hydroponic system." That's not going to get you the kind of content that is going to have any chance of providing much value to user, let alone ranking in the search engines. So hopefully, that's a bit of a takeaway of this, there's so much more that we can do to do better than that. 

I think if you really want to leverage a strategy, like what we are doing, then you have to think these things through to make sure that you are going at it in the most comprehensive way possible. That sort of leads on nicely, I think to the process. I'm keen to delve in a little bit more into the process behind the scenes for creating a blog content. So where do we start here? What's the first thing that you look at when we are creating the content?

Bonnie  

So just going back to your last point, Aidan, I would say that one of the things that we do is exactly not making just a single Line request to the AI. What we are testing really in the background is preparatory type of testing. So we do a lot of work on refining the prompts, we do a lot of work on keyword optimization, we do a lot of work on generating viral titles. To do that, we use a spreadsheet, and we do a lot of keyword research to find optimal blog post titles and keywords. Blogs written in the early days contained affiliate links, so there was also quite a bit of work to do on searching for the products. Bing can now help with this Bing AI, but it wasn't able to back then, and then gathering all the affiliate links as a second step using Amazon or whatever affiliate you're working with.

Aidan  

So once you've done that initial research there, what comes next? I guess, at this point, you've identified "Okay, we're finally going to find a keyword we want to create some content around." What comes next?

Bonnie  

Well, going back to the beginning, we then marshalled the team who was going to work on the AI side and who was going to work on the human side. So of course, the AI side is less time consuming. However, we found that already having a structure is really helpful in creating new content as research is done for you. So that makes the job of the human side less intensive, but equally rewarding, as they're effectively cutting down and reshaping what's already there and of course, checking accuracy. So to answer your question about the process, first of all, we organize the titles and the keywords. Next, the AI writer generates the first draft of the blog post. They already have the prompt in hand, they already have the titles, they already have the keywords. 

Aidan  

Just jumping there for one second, when you say the AI writer, are you referring to the person on our team that uses the AI tool to generate the content? Is that what you mean when you say AI writer? 

Bonnie  

Yeah, it's the AI generator. So in our case, we have a wonderful team of VAs, who helped us write the first draft. Then that goes to another VA, who is a specialist copywriter.

Aidan  

Right. Okay, cool. So that AI writer generates the first draft, and I believe that you do that on like a section-by-section basis, is it? 

Bonnie  

Yes. So what happens is we create an outline first. Because we're working with tools that need a little bit of help, sometimes to output a lot of content at once, GPT4 provides the outline, and then from that outline, we sort of copy and paste the sections back into GPT4 and say elaborate on this point, then we have another prompt that will help us elaborate on the point that has to be written. We have an outline, and then we augment each of those specific parts of the piece that we're writing.

Aidan  

So that sounds like you start out with a basic outline, could be like a table of contents, something like that, and then you say to the content, that AI writers, "Maybe we need a paragraph or a few paragraphs or a few 100 words around this section here," and that's their starting section, and then there's another section below that in the same article, and that is talking about something a little bit different. Is that more or less the way that it's done?

Bonnie  

Yeah, so piece by piece, we generate the content. It used to generate about 3000 words and then we cut that down to 1500 words to a sort of a multi-layer process that's focused on clarity and SEO. We're not keyword stuffing as a verb, but we're making sure that the keywords that we have are utilized optimally, basically.

Aidan  

What is the multi-layered approach? Can you delve into that a little bit more about some of the things that you sort of look at there?

Bonnie  

Yeah, it's about factual accuracy and adding personal insights, featured images, charts, and tables. So what we're trying to do is basically please the Google algorithm at this stage. We then go off and we then do originality checks to make sure that the percentage of AI is as low as possible and that there is no plagiarism so we have unique content as such.

Aidan  

Right. Okay. And what about, like, rewriting? For example, do you find that the content requires a lot of rewriting when it first comes out in that first draft? And if so, do you leverage tools for doing that as well? Or is that more of the human element again?

Bonnie  

It is the human element. Sometimes we use Quillbot as well, it's an AI tool that helps with rewrites. If we have a particularly complex paragraph to rewrite or something just doesn't make sense, the way it's written, it's easier to pop it into Quillbot and then rewrite it using the human side. Because again, Quillbot is AI, so we're still going to have to de-AI the piece after Quillbot.

Aidan  

I think it's also worth stressing here that every single piece of content that we create, first and foremost, we are thinking about the human experience. I know that you and I often sound like a broken record, because we're saying, "Look, we need this to be better, we need this to be even better. And we need to make sure that this is written for the human." We're not trying to impress the Google robots. That's like a secondary level of importance. Because if the human is not taken care of first, then you've got no chance at all of getting that content ranked anywhere, and anyone that comes to read it is going to probably think it's horrible content anyway. So it's really important to always, first and foremost, make sure that you're writing for the actual reader who's going to come to the website. 

Once the content has been created, how does it get published? And what about search engine optimization?

Bonnie  

We have a detailed focus on optimizing titles, descriptions, and keywords. That is our basic SEO, sort of effort, if you like. Just going back to your previous point, I think, you know when that you see a good piece of original writing, what a really good, engaging piece of original writing looks like. Rather than just importing information, it sort of takes you on a journey. That's really what I like to see the output being in from all our blog posts. 

To publish the content, we simply uploaded to the website. After that, we do more SEO work on backlinking.

Aidan  

Backlinking is, I think, a really important part of the recipe here because you can have the best content in the world, but if no one's out there talking about it, then it's not going to get as much exposure. 

Just to start sort of wrapping things up here, I'm keen to get your take on whether or not you think ChatGPT, I'm talking about Open AI, really GPT 4, for example, do you think that this is going to be dethroned by some of the up-and-coming competitors? We've got the likes of Bard from Google, we've got Claude, which is another one we speak about a lot, and others out there. What's your take on that?

Bonnie  

Well, what I really like is how Claude can consume an entire textbook, Claude 2.1, and then output sensible, sensible commentary or a sensible critique on it. That's phenomenal, and then that has huge implications for anything scholarly, or anything academic. Imagine uploading, take all the Shakespeare's works and being able to ask questions on that. Imagine being able to upload Mark Twain's works even, to be able to easily ask questions on that content, obviously, as well, it has huge implications for legal. Again, you have to be careful as to the interpretation because their interpretation is subjective at the end of the day, but it's still a very exciting time of possibilities and potentials. How quick will it be to write a book in a year's time? You can write a book at a click of a button almost now by outlining it and then taking various chunks of the outline and asking AI to expand on that. There's a huge amount of possibilities. 

Aidan  

I think the tools also, Bonnie, are only as good as, right now anyway, as the humans that are operating them. So right now, I think our main constraint in the world of AI is what the humans are imagining. For example, there are definitely ways that you can create a book from start to finish at the click of a button that's got images, they could even have dynamic content in there, and much, much more. The only thing stopping that is the human sort of building the process behind the scenes. I think there's huge improvements we're going to see there. 

You mentioned a moment ago about being able to upload an entire work from Shakespeare or Mark Twain, I haven't tried doing that. Just recently, I uploaded about 10,000 words from Aidan Booth. slightly, slightly different, different contexts there, but I did upload 10,000 words, which was a transcript of some training that I provided. I asked AI to give me a high-level breakdown, essentially, take these 10,000 words, and give me a two- or three-page summary. I was blown away by what it was able to do. That was with a transcript, but if you can do the exact same thing with the video, what if you had a TED talk, for example, and it was a 60-minute TED talk, and it was jam-packed with information? What if you just gave that link to a tool and said, "Analyze this and give me a summary." Well, that's completely doable already. I mean, it's just incredible. It's mind blowing. I think right now, one of the biggest constraints that that the average user has, and we're included in that as well, you just don't know what you don't know. You don't know that something's possible until you hear that someone else is doing it, or until you really sit down and think, "How could I approach this using the AI tools that that we've got there?" 

So anyway, off on a bit of a tangent, but you mentioned about uploading entire works from Shakespeare, and I just thought that folks would get a kick out of knowing that I've loaded entire works from Aidan Booth and it worked.

Bonnie  

I mean, it's worth that all that content. How wonderful is it to be able to repurpose that so easily in all sorts of other forms. So we have blog forms, we could have video content. Speaking of video content, Bard now has the capability to watch videos for you and to handle complex questions about specific video content. I guess that's through YouTube.

Aidan  

Yeah, it's not just Bard either, we've got and we use tools as well that for example, if we've got a training course, we can upload 10 videos or however many videos from that training course, the AI tools that we use can learn and understand the content of those, say, 10 videos. And then if you are a customer and you're going through the course and you say "Oh, what is the information around search engine optimization and keyword density when I'm creating a blog post?" if that there was contained in the 10 videos that I had uploaded, then the AI tool could say, "Oh, yeah, that's covered in video number three, 14:31. The best practice is to do such and such…" These tools can learn and then feed back this information. It's just incredible. So I'm really excited about that. 

Bonnie  

Yeah, that's like finding a needle in the haystack. It's so exciting for any course you would take, to be able to go back and say, "Oh, what was that part about?" and instead of having to watch hours and hours and hours of video, you can go back immediately, almost instantly, to that particular piece of content that you need. 

Aidan  

Yeah. What about over the year ahead, Bonnie? I think this is the last question I really have for you here. Is there anything that you are most looking forward to diving into in the AI world?

Bonnie  

The AI arts space is a very interesting one. Prompts have become more sophisticated. I see artists combining output from one platform with another. An example of this is say you can you can get a prompt from chat GPT 4 for Pinterest about your demographics, psychographics to actually create your whole brand from one single prompt from GPT 4 and ask Pinterest to generate that. The arts space also, you've got DALL-E and you've got Midjourney. You've also got other tools coming up that are even better. Obviously, DALL-E, it's still DALL-E 2, Midjourney is evolving every single day. You can you can input one element from DALL-E and ask Midjourney to improve upon it, for example. So it's a very, very creative space. I'm just excited to see what people will create from it.

Aidan  

Yeah, we don't know what's coming next, but we do know that it's exciting. So Bonnie, thank you so much for your time. Couple of things I wanted to emphasize, I think the balance of tech and expertise is really, really important. It's that combining of AI tools with human skills that in my mind right now, as we are in 2024 and looking for the months ahead, is something that is vitally important. The people that can get that combo right I think are the people that are going to win. 

It's not just about creating massive amounts of content. It's about creating the best content, using all the tools that we have at our disposal. And then also, just making sure that these tools are used in an ethical way. I think there are lots of ways that you can use AI tools for good or for evil right now. I think it's just making sure that you're really clear in your mind about what you want, what your objective is, and making sure that the output that you're generating ticks all the boxes there. 

Bonnie, this has been fantastic. I'd love to have you back again in the near future to drill down into some of the tools that we've mentioned even more. Thank you so much for being here, Bonnie.

Bonnie  

Thank you for having me. It's been fun.

Aidan  

Alright guys, that's a wrap. This is episode number 109. As always, you can find it over at thegrowthbooth.com, navigate to episode number 109. You'll get show notes, you'll see the transcription links to different things that we've spoken about. And you can also see the video content on thegrowthbooth.com as well as along with finding it at YouTube and the audio content as well wherever you like to listen to your podcast. That's a wrap. We'll see on the next episode of The Growth Booth. Bye for now.