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

Alexion

Banned - Shitlisted and did not do the right thing
Joined
Sep 9, 2021
Messages
1,455
Reaction score
3,707
Age
34
Website
join.skype.com
I'm sorry of I missed it, but are you also gathering entities / topical keywords on those articles? Are you in any way trying to integrate them as well?
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
I'm sorry of I missed it, but are you also gathering entities / topical keywords on those articles? Are you in any way trying to integrate them as well?
To create categories/silos yeah? I'm working on it, that's why the site is not up yet. I've been a bit busy with real-life stuff to put in a lot of coding this week ;)
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
I have posted ~30 high-quality generated articles on a website that I own to check Google's reaction. Submitted them all to GSC and they are already indexed.

It's a white-hat website getting around 100k views/month and these are the ahrefs stats. All links are organic, I never built anything manually.

nOuWmrO.png
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
The paraphrased articles are ranking. Page 1 within 5-6 days for some zero competition keywords. I will post screenshots when I get the chance. I refactored the code, sped up parafrasing by 10x - takes me about 6 minutes to generate a high quaLity article now.

Now polishing the last bits of my app and off to launch 3 sites.

Wow great journey, I dont understand properly but can you please tell me, for only Paraphrasing do you using spacy? or pegasus_paraphrase? or both?

I mean for your top quality Paraphrasing.

I can't reveal it because I'm breaking the TOS and really trying to fly under the radar with automating this. It's one of the commercially available tools. Non-gpt3.
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
haha, no. I've taken a different approach. Non-abstract method. Pure NLP/grammar-based. Instead of trying to understand the context, as GPT does, I'm only working on a per-sentence basis. This leads to mistakes, but paradoxically fewer than GPT-3.

I think you could use something like QuillBot Premium(has an API!) and the results will be better than GPT-3 if you put some clever code around it. Spacy is great for tokenizing. Or maybe use some low-level ML way to tokenize it (Google has some good free stuff - https://github.com/google/sentencepiece)

This is the latest version of the paraphraser, I think I'm done here and feeling proud AF:

7PvTYDq.png

Getting clicks after 6 days of posting the articles! Pretty happy about that too.

photo_2021-10-21_10-02-58.jpg

Now just gotta move fast and scale before this loophole gets patched somehow by big G.
 

lastbencher

Junior Member
Joined
May 19, 2013
Messages
169
Reaction score
59
haha, no. I've taken a different approach. Non-abstract method. Pure NLP/grammar-based. Instead of trying to understand the context, as GPT does, I'm only working on a per-sentence basis. This leads to mistakes, but paradoxically fewer than GPT-3.

I think you could use something like QuillBot Premium(has an API!) and the results will be better than GPT-3 if you put some clever code around it. Spacy is great for tokenizing. Or maybe use some low-level ML way to tokenize it (Google has some good free stuff - https://github.com/google/sentencepiece)

This is the latest version of the paraphraser, I think I'm done here and feeling proud AF:

View attachment 189442

Getting clicks after 6 days of posting the articles! Pretty happy about that too.

View attachment 189443

Now just gotta move fast and scale before this loophole gets patched somehow by big G.
I have tried couple of approaches including Quillbot and Pegasus, Quillbot no longer offers API and for some reason result is not as readable [feels spun] as pegasus . In both cases 10 to 30% plagiarism was observed easily. Quillbot approach is fast and pegasus takes 10 to 15 minutes for 1000 words article.

What are you doing for images and text formatting like bold,italic?
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
I have tried couple of approaches including Quillbot and Pegasus, Quillbot no longer offers API and for some reason result is not as readable [feels spun] as pegasus . In both cases 10 to 30% plagiarism was observed easily. Quillbot approach is fast and pegasus takes 10 to 15 minutes for 1000 words article.

What are you doing for images and text formatting like bold,italic?
for images I'm using Unsplash API(free) + Pillow (Python image manipulation) + Nider (Python library for writing on images)

bold, italic, that's just html/markdown, just gotta choose what you want to emphasize. Spacy is enough for that + there are many other ways you could establish which parts of the paragraph are significant, like words around the subject of the sentence.

If Quillbot doesn't offer API anymore then use Selenium or reverse engineer their private API :p If this would be easy everyone would be doing it.
 

ANewBeginning

Banned - Scamming
Joined
May 13, 2021
Messages
1,040
Reaction score
718
Now just gotta move fast and scale before this loophole gets patched somehow by big G.
hehe not going to get patched any time soon :p

Google keeps saying autoblogs are dead just to scare people, but the reality is that they still work and WILL continue to work UNLESS there was manual reviews of EVERY SINGLE site out there, which is impossible :p
 

Sartre

Jr. Executive VIP
Jr. VIP
Premium Member
Joined
Apr 1, 2010
Messages
1,572
Reaction score
3,895
Website
SwiftGrowth.com
hehe not going to get patched any time soon :p

Google keeps saying autoblogs are dead just to scare people, but the reality is that they still work and WILL continue to work UNLESS there was manual reviews of EVERY SINGLE site out there, which is impossible :p
this I don't understand. You could easily train a model which predicts with 99% if a site is an autoblog/spam site.
 
Top