Using A.I To Create Content With Przemek Chojecki Of Contentyze

Przemek Chojecki, Founder at Contentyze joins Hammad Akbar in this episode of Launch Legends Podcast

What’s a Rich Text element?

  • Hello
  • how are you?

How to customize formatting for each rich text

Headings, paragraphs, blockquotes, figures, images, and figure captions can all be styled after a class is added to the rich text element using the "When inside of" nested selector system.

Key Stats on Contentyze

  • Generating $5000 in revenue
  • 100,000 articles created through Contentyze

Key Takeaways on Contentyze

  • AI has progressed a lot since 2014
  • Monetizing a product is the most difficult part
  • Consulting business is difficult to scale
  • Make a product that addresses your potential customer’s pain points
  • AI helps in generating unique content
  • Use what’s best when it comes to open source
  • AI can sometimes give you buried content
  • AI cannot replicate content of actual writer
  • One needs to focus only on one thing at a time
  • Make a product easier to use.



Hey Przemek, thank you very much for coming to the show. So I hope I've got your name, right?


Yeah, it's correct. Thanks for the invitation. Certainly it's a pleasure to be here. 


Great. So, you are building an impressive platform. I know you use the platform to create like a thousand articles just in the last month, you know, AI content creator.

I'd love to talk about that. You just launched not long ago. And you've got about $5,000 in revenue and let's talk about how you got there. And, but before everything, I would love to talk about who you are. I know you have a very impressive background, so let's talk about who you are and how you actually got here.

And then let's talk about the platform. 


Sure. So, I had a very academic background. I mean, for most of my life, I was a mathematician. I did my PhD in pure mathematics in Paris. And then after five years in Paris, I actually went to become a research fellow at the University of Oxford. 

And then I, at some point decided that, okay, I want to make the switch from academia to business and for, from pure mathematics to machine learning to AI. Well, in order to have a little bit more impact on the world than people in academia usually have, and also because I'm much more fast-paced than, than the usual academic.

And I love to have this much faster feedback loop. So I love to do something and, you know, like to see the outcome of what I'm doing, or like have the people giving me their opinion on what I'm doing much faster down than awaiting for a year or more in order for someone to read your research paper and then comment on that.

So I had this idea of going to AI and that was pretty natural for me because my domain of mathematics was really very technical and there are already people thinking about, okay, can we make it. 

Can we make it a whole process of actually reading the proofs, understanding mathematics better for mathematicians? So it was natural to think about automation of this cognitive process of doing research. And I went this path. Then at some point, I said, okay, this is great.

I'm going that direction all the way through. And basically I landed on the other side with machine learning. So this is how it started for me with machine learning. And on the other hand, I was always very keen on literature and writing. So that was always my hobby. I like to read a lot. And then I started writing at some point, publishing my first books.

So in the meantime, I was looking for something where I could connect the two and the Contentyze is, well, the fruits of that. Basically Contentyze is like both worlds coming together.  


So when you were the research fellow, when you actually first got into AI, what year was that?


Let me think, that was 2014 and I was at Oxford from 2014 to 2016. 


Machine learning has progressed from that time. And it's not that long ago, but I know AI is progressing at a very fast pace. Has that progressed from that time to now? What kind of changes have you seen? 


Sure. A lot, so actually, if you look back historically at this first wave of AI, because we already have like two AI revolutions in that, I guess the first one was in the sixties.

The second one was  somewhere around in the eighties, but it just didn't deliver on its promises like the whole AI industry, because there was not enough computing power to do it. So actually the first revolution like that they want to know it and everyone is, you know, like, what AI can do that started in 2012.

So that was very recent, so I was already at Oxford when I heard from different friends of mine who are doing computer science, there's something happening. And probably you should look up that because they're like, there's so many interesting problems and there's so much potential in doing that.

But it's still 2014 when I started being in Oxford. No, those are still very early days for a disparate wave of AI and machine learning and there was a huge progress from 2014, 16 to where we are right now. Everyone understands that AI is the future. It's already everywhere, but it's becoming even more  with time.


Great. Right. So I know before content, you built something else. What was that? The age of your platform? 


Yeah, so that was the first step to Contentyze. I also had two other startups that failed before Contentyze. One was like a global or Postmates. If you're familiar with those, that was basically, something like Uber, but for everything from people to food.

So we want it to deliver, like you could shop somewhere and deliver your shopping or you forgot your key and we could drive that for you. So that was the best part. The second startup and we underestimated our costs for the whole thing. The second startup was super technical.

We used quantum computers to optimize logistics and transportation, and that was really cool. We did that with my co-founder in Toronto, we went through a creative destruction lab. That's an accelerator in Toronto. And that was really exciting and that actually failed for a different reason.

We're extremely early in the market. That's the first thing. So we had really a lot of trouble getting our first traction or clients. And on the other hand, we had very different views on where the company should go. Should we be more like product based or service based? So in the end we split over that.

So that was before Contentyze. And then we came to PetahCrunch, which is this interview platform you mentioned. And the reason I started that was after my second startup failed, I was curious about, you know, how people build a successful company.

So I wanted to ask a couple of founders how they did it. Like, and what are the tips they might have for me and instead of just like, so I started to email those people, but then I felt like, you know, I was always thinking about how can I automate this process?

How can I make it better, faster? and I decided, okay, so maybe I can try to do this platform where there would be like AI journalists and they can ask people to actually ask those kinds of questions that I want to know. So, those were like typical questions. I asked two different companies who actually raised at least 1 million euros or much more.

And the questions were like, how did it all start? What's your plan for next year, for years, whether we're going to do with the funding and so on. There were like five major questions. So actually automate the whole thing. And I managed to interview 1000 founders, in like three months.


That sounds really interesting. And I actually can think of a use case from itself. So for example, as a founder, I'm always interviewing potential customers and my customers. So straight away, there's a use case I can see. So could you break down what you actually did? How did you, when you say you created an AI journalist, what was that journalist?


So there are two components to that, I guess first of all, you need data and in this case, the data is what kind of founders you want to interview. and that's it. So, I managed to get that because that's more or less public. Like it's probably like everyone is happy when they race around.

So that was easy to get, The data from Crunchbase, it's perfect for that kind of thing. And they have an open API, and want to pay the subscription fee. So that's super simple. So I just, you know, like have that already exported through CSV for months. So that's perfect. And on the other hand, you need to automate the process of asking for the interview and know what you offer.

Okay. And that's in this case was super simple because that was like a two step process, basically where you write an email saying something like, hi, congratulations on your recent funding ground. I'd love to interview your CEO and usually that goes to the general email that's public.

And then if someone responds then based on whether that's yes or no, you either send the questions or that's it, you do nothing. 


Okay. So here's a question. What was your criteria for yes or no? Because that could come back in so many different ways. It was your bot actually determining whether that's a yes or a no.


Yeah. So there were actually three categories. Yes, no, maybe and I mean, I did the most simple thing. So when it's no, it's no, I did nothing. When it's yes, we send the questions right away. And when it's maybe, it's undecided, or maybe there's like additional questions from the founder or from the team.

Then I jumped in and answered that question manually. The whole process was almost like semiautomatic in the sense that I still needed to put in a little bit of work. but it was much faster and I was really able to interview 1,000 founders in three months, which is like impossible if you were to do it manually in any way.

But the most painful part was actually the format of it, because all those teams were sending the interviews in different formats somewhere, like just answers in the email somewhere, like sending me PDF files, doc files, or whatever it is that they had. And that was really a pain because I had to go for everything and put that to the WordPress.

So there was like, it would be complicated to actually do an AI, which could write, read everything and do, you know, like put it on the website. So I decided I will do it manually. 


Right. Right. So what did you do after that? Why did you not pursue that as an idea? 


So, it's still working to some extent, but the problem from my perspective was that how do I monetize it?

What can I do actually and that's still like an open question for me. So the solution that I came up with is that PetaCrunch is still working and you can still see those interviews. And, if you want to probably have an interview with better guests, then actually you can play for that and that the price goes to $99 to do that and it's working. 

I mean, it's earning money but it's like a couple of hundred dollars per month. So, the question is how to do it in a way to make it a meaningful business. And I didn't have an answer to that. 


Yeah. Cause as you're telling me that I'm actually thinking, okay, that could be like a prototype service where you could come to someone like me who needs to get a lot of data from real people, potential customers of mine.

And then I can just delegate it out to you and say, look okay. I need a thousand people to answer these questions. So could you please go and get that done for me? And I will pay a good amount of money for that because that data is very valuable. I don't need the data to be published anywhere.

I just need the data so I can analyze it. 


Right. But yeah. So I was thinking about that and actually a couple of people contacted me about this kind of thing, but it would be all very customized, you know, like it would have been tailored to your needs. And It would be more like consulting from my perspective then, you know, like having an actual product that I can sell.

So, you know, I didn't want to do that because that's not scaling well. I mean, I would have to put my own time in each such project. 


Great, great, great. Well, let's talk about Contentyze. It is a great platform. I would love to hear about that. How did that come about and how did you build it?


Yeah. So it's sort of like after this whole interview survey, I knew that I wanted to go more into the content direction and start generating content at scale. I already had like, semi-successful a medium blog with like 3000 followers and it's still growing. 

So my first goal was to try to build the solution for myself that I could use in order to boost my own writing and make my writing faster, smoother and so on. So I started experimenting with those kinds of AI algorithms and started putting much more texts online. And this is how Contentyze is actually started. 

I mean, it started with a tool for myself to boost my own writing and then I remark that actually it can make it into a SaaS platform. And this is a perfect product to start selling online to anyone who's actually working with copywriting, SEO, marketing. 


Okay. Got it. So, just to break it down, how does a product work in terms of technology? because anyone who needs fresh content, then thinking about the authenticity of the content, just to make sure that it's not plagiarised, obviously that's not happening in your platform, but how did you make sure that your content is a hundred percent unique and it can be used for different purposes? 


Sure. So, the kinds of AI algorithms that I'm using on the platform, are based on statistics in the sense that each article is generated from scratch.

It's not like it's not like an AI algorithm just trying to fetch the content from different websites and, you know, like spin the content and then give you that. It's trying to generate sentence by sentence and word by word, whatever you ask of. So to give you an example, there are two main functional IDs of the platform.

The first one being that you supply a headline and it can be a question, the title of the article, like what was that? How to run a successful marketing campaign on social media, right? And then the AI algorithm itself is actually taking that and trying to predict the next sentence. 

And after that, the next sentence, but it has so much variance because, and each time you even write that you get a different unique text, because of the statistics, because the article is not trying to fetch that content from anywhere, it tries to generate it based on what it has read. And it has read like millions of articles, but it's going to generate that word by word. 


Got it. So let's go back to the AI algorithm.

How did you train the AI algorithm to actually do that? 


So I just take what's best out there when it comes to open source, try to mix and match that, you know, like basically duct tape, everything like make the MVP of that's worked pretty well. And then on top of that, I started to find AI algorithms and by finding I mean, trained the whole thing with my own tags and more and more, but actually the idea here is simple.

You're trying to make the AI algorithm predict the next word. And so you're giving the AI algorithm like a couple of sentences and asking for the next word to be predicted. And once the AI algorithm actually seems like millions and millions of facts, it can generate pretty original content. Right. But here's the problem though with this kind of approach, it gives you buried content.

So you have that the content is unique, but actually you should really check what's being generated most of the time because it can derail and go like bonkers in different directions. So yeah, most of the time it's fine. 


Well actually I had you say it somewhere else that you hire contractors like writers to write content on a particular topic. And then you get your product to create the same content distinctly, just to see how they compare, and what the authenticity is like when you do that, what kind of differences do you find? Do you find that obviously they're not similar, but the quality is similar where the content creation by your platform is very authentic and the context is right as well.


Yeah, it is similar though there are differences where especially you want to write about something more like actual data. So if you want to start citing like actual sources, then Contentyze has come to it just yet. I mean, we do that for our clients, so there's a little bit of like custom work we do for enterprise clients.

We're actually able to fetch data from like different sources. But we don't do that, like you, you won't be able to do that just from the off the shelf product. So that's one difference. Like if you hire content writers that are able to like, do this research and take data from different sources, different articles, and maybe combine that together.

On the other hand, this is for the generation from the headline. On the other hand, the other main functionality is summarizing content. And for this purpose, it's perfect. Because if you have a couple of articles that you want to spin in a sense, then you can just feed that to Contentyze, and then get the summaries from those articles.


Great. I think I know a good video creation application that does something very similar where I think it's called Lumen5 where you feed it the articles. So are they using a very similar AI algorithm? or that's something else.


Oh no, it's something totally different. I would say no. No, because actually, they're doing something different because they took the script from you and based on the screen, they tried to create the video and this is totally different because if I were to guess, then they try to go sentence by sentence, try to like, get the meaning of the sentence. 

And maybe that's like a couple of tasks, a couple of keywords. And based on those keywords, give you an image or a video, you know, and then do this, like split from like your sentences. You have videos and then match them. So it's still a little bit different, but the two compliment each other pretty well.

So actually I'm a huge fan of Lumen5. And actually I tested them together with Contentyze that I generated the script and then fed the data into the lumen5 to get the video. And that's really impressive because you have like the video creation totally automated.

You can generate the script and it can then feed it to Lumen5 to have the video. 


Great. So a minute ago you said that at the moment, you can't really replicate a copywriter where they do the research and they find content from particular articles.

They always get the inspiration from there and then mix it all together and create an article. How soon do you reckon Contentyze is going to be there? What's the process


To be honest. I don't think that that's the goal actually, so my end goal is not replacing copywriters. My end goal is to actually make them work much faster and make work easier for them.

So, in the end, like if you think about hiring five copywriters, you might need only one in the end and that's the goal, like to save the time because there will be always small tasks that it would be too complicated to teach the mashing or that will take too much time and effort from your perspective, then it's actually better to hire one person.

So my goal is not to get rid of copywriters and like not have this kind of job. Rather, cut how many copywriters you need in the end and go from five to one. 


Right. So one last question. So, I know you guys are making about $5,000 and you said the most of that revenue is coming from actually generating your own content, affiliate offers and giving content to your clients.

But you're shifting that over to a more SaaS model where you want your paid users too. You know, to increase your revenue. So let's talk about how you are generating revenue. Let's break that down. I mean, how do you make money? How do you actually break? How do you make money now? And then what's the plan for the future in terms of growth?


Sure, like right now, most of the time, most of the revenue is coming from either advertisements or affiliate links. So, because we produce so much content, you know, like when people click through and buy something from that content and we get a small percentage from that, and that's basically $5,000 per month and the good thing is that we're slowly shifting to, just having the revenue from the SaaS platform.

And that's the end goal to actually leave content to others and maybe like half a block or two, but actually be more on the side of a provider of the tool rather than the content. And the reason for that is, well, both of them take a lot of work and you can't do both very well. And I think like right now it's more important to actually give tools to people rather than try to already think about the content.

Because I think like there will be a lot of developments in AI in this particular space of copywriting, SEO that it's still too early to just, you know, like monetize all the content. But I think like in a couple of years time, I'd probably go into AI generated content only. And that would be the only thing.


Right. So, where do you see Contentyze in the next three years? What I know the end goal is to replace, not replace copywriters, reduce the number of copywriters to maybe one or two, but in terms of technology and features, where do you see that in the next couple of years?


So my end goal is to actually have Contentyze being available to anyone, just like, and that can be like a Chrome extension, Google docs extension. So whatever you are, you don't have to go to the platform, but imagine that you are writing something, whatever that is, and you just click on something like a suggestion and Contentyze can suggest you the next sentence, the next paragraph, or like, correct your grammar.

Maybe give you tips about SEO instantaneously, no matter where you are and that's the goal. That will make it a great product. Not only for copywriters and SEO marketers, but for anyone actually writing themselves, students and, and the professionals, that write a lot.

And, and that's my end goal actually, to make writing much easier for everyone. 


Great. Great. Thank you very much and a great product. So I look forward to actually using it myself for my own content. And, I hope to have you again on the podcast. Thank you very much. 



Subscribe To Launch Legends:

Key Resources:

Recommended Episodes: