[Journey] 1 million UVs/month in 12 months using AI generated content. Let's do it!

[Sartre]

I Though $500/month

I'm in europe and my business generates significant income anyway so it's not an issue.

 

[Shaybib]

Have you encountered a problem where Pegasus returns numbers instead of text for some sentences?
Any tip how to avoid it?

 

[Sartre]

Have you encountered a problem where Pegasus returns numbers instead of text for some sentences?
Any tip how to avoid it?

I haven't seen this one. Switched to a custom model now though, no longer using Pegasus. How often does it happen?

 

[desperado596]

I haven't seen this one. Switched to a custom model now though, no longer using Pegasus. How often does it happen?

impressive. you are improved much. are you using any kind of gpt2-3 bert etc?

 

[spectrejoe]

Not sure if mentioned in the thread but how are you indexing all the new posts you upload? Do you just let it index naturally? Can you check how many are indexed in GSC and how many posts you've made so far?

 

[spectrejoe]

Not sure if mentioned in the thread but how are you indexing all the new posts you upload? Do you just let it index naturally? Can you check how many are indexed in GSC and how many posts you've made so far?

 

[Shaybib]

I haven't seen this one. Switched to a custom model now though, no longer using Pegasus. How often does it happen?

This happened many times...
I succeeded to avoid it by checking for how many times the char - appears as the result as the text is lwase somthing like: 667-678-678...
I think it returns the tokens number represent instead of the words.
What is the advantage of custom made model?

 

[Sartre]

impressive. you are improved much. are you using any kind of gpt2-3 bert etc?

we're using a NLP model right now. Hard to believe but, yeah. Just put in a lot of work. I'm going to show you an example down below.

Not sure if mentioned in the thread but how are you indexing all the new posts you upload? Do you just let it index naturally? Can you check how many are indexed in GSC and how many posts you've made so far?

90%+ indexed. Just like my whitehat sites.

Not sure if mentioned in the thread but how are you indexing all the new posts you upload? Do you just let it index naturally? Can you check how many are indexed in GSC and how many posts you've made so far?

i leave it to index organically. Created a custom theme so that every post is always max 2 clicks away.

This happened many times...
I succeeded to avoid it by checking for how many times the char - appears as the result as the text is lwase somthing like: 667-678-678...
I think it returns the tokens number represent instead of the words.
What is the advantage of custom made model?

it doesn't make major mistakes and can be fine tuned for a specific domain.

Here's an example:

Original:

SEO In a Day
SEO is simply not as hard as some people pretend like it is; you can get 95% of the effort with 5% of the work, and you really absolutely do not need to hire a professional SEO to do it, nor will it be hard to start ranking for well picked key search terms.

Of all the channels we’ll be covering, SEO is the one that there is the most lies about. Some of it is subtle, but some of it is widely spread and believed by so-called SEO employees who actually don’t know what they’re doing.

SEO is very, very simple, and unless you’re a huge company it’s probably not worth hiring somebody else to do. It’s also something that has a lot of bullshit around it. Consultants want to make it seem incredibly difficult so that they can charge you a lot, but I'll show you exactly how to do it, step-by-step, and you'll win your rankings.

How Google Search Engine Works
In order to really understand what we need to do for SEO let’s look at how Google started, how it’s evolving, and develop a framework from which we can comprehend how to get ranked on Google.

First, we're going to reverse engineer what Google is doing, and then just follow their rules, picking the right topics, and get your site ranked.

Paraphrase:

Search Engine Optimization in a Single Day
SEO isn't as difficult as people make it out to be; you can get 95% of the results with just 5% of the effort, and you don't even need to employ a professional SEO to get started ranking for important phrases that you carefully choose.

The most erroneous information exists on SEO, out of all the topics we'll cover. Many SEO consultants who don't know what they're doing share and accept this misinformation, some of it is subtle.

If you're not a really big firm, it's probably not worth employing someone else to conduct SEO for you since it's so basic. It also has a lot of snakeoil around it. In order to charge you a lot of money, consultants like to make it appear impossible, but I'll show you precisely how to accomplish it and you'll win.

Google's Methods
Let's take a look back at how Google originated, how it's changing now, and build a foundation from which we can grasp how to get ranked on Google.

To begin, we'll analyze Google's algorithms to determine what works and what doesn't, and then we'll use that information to optimize your website.

this paraphrase took 8 seconds to complete.

 

[prvncqr]

we're using a NLP model right now. Hard to believe but, yeah. Just put in a lot of work. I'm going to show you an example down below.

Can you share more info about the model? You don't need to share any methods. Just curious to know what model you use.

 

[Sartre]

Can you share more info about the model? You don't need to share any methods. Just curious to know what model you use.

me and my partner built a NLP model technically more similar to old-style NLP like this: https://ai.googleblog.com/2021/12/a-fast-wordpiece-tokenization-system.html than to ML. It's more of a synonimizer than a full rewriter. For white hat sites I'm using a human editor because I don't want to take any risks for websites that are already on Mediavine, but it speeds up the writing process 5-fold. For spam sites I'm just posting whatever it produces. We will see how this works out in the long run.

This is a new spam site started a month ago, so far the most successful launch of an automated site:


US-centric niche.

Taking 2-4 weeks to improve the code and then going full-scale with these. I just hired another programmer temporarily to help out.

 

[prvncqr]

me and my partner built a NLP model technically more similar to old-style NLP like this: https://ai.googleblog.com/2021/12/a-fast-wordpiece-tokenization-system.html than to ML. It's more of a synonimizer than a full rewriter. For white hat sites I'm using a human editor because I don't want to take any risks for websites that are already on Mediavine, but it speeds up the writing process 5-fold. For spam sites I'm just posting whatever it produces. We will see how this works out in the long run.

This is a new spam site started a month ago, so far the most successful launch of an automated site:
View attachment 201969
View attachment 201970
US-centric niche.

Taking 2-4 weeks to improve the code and then going full-scale with these. I just hired another programmer temporarily to help out.

Great. How are you monetizing the automated sites?

 

[Shaybib]

This is a new spam site started a month ago, so far the most successful launch of an automated site:

This is remarkable!
How many pages on this site?

 

[Daemon]

This is a new spam site started a month ago, so far the most successful launch of an automated site:
View attachment 201969
US-centric niche.

Taking 2-4 weeks to improve the code and then going full-scale with these. I just hired another programmer temporarily to help out.

Impressive. How are you launching these automated spam sites? Are they on expired domains, or new domains? Do you build links to them from your existing sites?

Are they micro niche focused sites built after a lot of keyword research, or they more general sites?

 

[pcsup]

me and my partner built a NLP model technically more similar to old-style NLP like this: https://ai.googleblog.com/2021/12/a-fast-wordpiece-tokenization-system.html than to ML. It's more of a synonimizer than a full rewriter. For white hat sites I'm using a human editor because I don't want to take any risks for websites that are already on Mediavine, but it speeds up the writing process 5-fold. For spam sites I'm just posting whatever it produces. We will see how this works out in the long run.

This is a new spam site started a month ago, so far the most successful launch of an automated site:
View attachment 201969
View attachment 201970
US-centric niche.

Taking 2-4 weeks to improve the code and then going full-scale with these. I just hired another programmer temporarily to help out.

This is awesome, congratulations... quick question what's an NLP model ?

 

[callmemaximillian]

@Sartre astonishing journey!
Could you, please, share how do you handle headings creation(h1,h2,h3) with NLP?

 

[Sartre]

Great. How are you monetizing the automated sites?

not yet. My plan is to improve this so much that I can get these into Mediavine with a little human editing. I'm going to have more help from Monday, hired a Python developer. I have a few white-hat sites on Mediavine already.

This is remarkable!
How many pages on this site?

364k words

Impressive. How are you launching these automated spam sites? Are they on expired domains, or new domains? Do you build links to them from your existing sites?

Are they micro niche focused sites built after a lot of keyword research, or they more general sites?

I'm buying aged domains on auctions, I was a domain trader for years. It speeds things up. But it's not a necessity. I was using expired domains before, but this works much better.

definitely specialized niche-sites with an author picture, bio, about, privacy pages, etc. I want to pass manual reviews.

This is awesome, congratulations... quick question what's an NLP model ?

https://en.wikipedia.org/wiki/Natural_language_processingBERT is an example of an NLP model.

@Sartre astonishing journey!
Could you, please, share how do you handle headings creation(h1,h2,h3) with NLP?

headings are always super long-tail keywords. Article titles are always broader keywords like let's say the keyword is "vegan foods for beagles", it would be "Vegan Foods for Beagles [A Definitive Guide]"

 

[Shaybib]

words

Can you share how many pages or average page word count?

 

[Sartre]

Can you share how many pages or average page word count?

there is no one answer, it depends how many long-tails the app finds for that article. anything from 400 to 1300 words in general.

 

[absolvo]

me and my partner built a NLP model technically more similar to old-style NLP like this: https://ai.googleblog.com/2021/12/a-fast-wordpiece-tokenization-system.html than to ML. It's more of a synonimizer than a full rewriter. For white hat sites I'm using a human editor because I don't want to take any risks for websites that are already on Mediavine, but it speeds up the writing process 5-fold. For spam sites I'm just posting whatever it produces. We will see how this works out in the long run.

This is a new spam site started a month ago, so far the most successful launch of an automated site:
View attachment 201969
View attachment 201970
US-centric niche.

Taking 2-4 weeks to improve the code and then going full-scale with these. I just hired another programmer temporarily to help out.

Interesting. How are you monetizing this?

 

[Simon5314]

Wow i was just trying dame thing with python but not using any machine learning just abusing commercial tools i just published about 80 postes last 2 days .thanks for valuable info you just saved me lot of time btw from where you get the keywords to search for the unsplash api using ml for them ?