Quick GPT-3 journey(case study)

Nice, doing this with a site right now actually.
 
Good luck! Following this journey.
 
I really thought I would have more time for this.
Squeezed in an hour today.
Here are some updates.

Published 13 more articles.
Total published: 20
Total indexed: 20
Comment: Apparently, Google is back to indexing posts really fast. Whenever I post something new, and submit it in GSC, it gets crawled and indexed in about an hour. Personal opinion is because of Cloudflare IP, so they don't downplay crawl rate.

As suspected, the website started getting out there because of the low comp keywords.
Here are some stats from GSC:

1660985952020.png


I've manually checked every post on google to see if it really shows up, or sandboxed.
Everything is on google. Nothing dissapeared.

Can't post a screenshot with the keywords, but here's the total number of keywords the domain is showing up for according to GSC:
1660986023999.png
64, but half are probably variations of a keyword.
Got some clicks from Google, and to my surprise, from Bing.

I was hoping to have at least 100 articles by tomorrow. Job got in the way.
Will do an effort and post at least 30 more this weekend.

Will invest 30 minutes today to do an about page and try to link the author's name with some random profiles on social media that match the name, E-A-T wise. Not sure how best to do this.

Will hunt for a cheap grammarly account and try to test it on a few articles. I've heard it can do magic with the words.

Semrush picked the website up, bit it's way off.
1660986248114.png


Semrush report 3 backlinks.
I've checked and all of them are auto aggregator blogs.
Still a DF link, but they're crap.

Will do my best to post a higher number this weekend.
The posting is still manual.

I've played a bit with WP ALL Import thanks to @blogmen but for now it messes things up.
Will have to invest another hour or two to tweak my code into doing it automatically.
I know how, but there are these small edits I do which not sure how to best accomodate them from the software.

I remember someone said AI is not good at technical writing.
For the stuff I ask it to write, it does a good job.
I mean, if I were to look this info up and my AI blog would show up, I'd be happy with what I read.

Again, sorry it's taking longer than I expected.
I want this to be quick.

Either it blasts off or I'll try to see how much I can grab for it on flippa, passing it as a legit blog.

Will keep this updated.
Happy weekend!
 
I've played a bit with WP ALL Import thanks to @blogmen but for now it messes things up.
Will have to invest another hour or two to tweak my code into doing it automatically.
I know how, but there are these small edits I do which not sure how to best accomodate them from the software.
I put one article per line and use a different symbol as csv separator because scraped data is dirty and scrambles the lines/columns order when you use comma or ordinary symbols.
 
link the author's name with some random profiles on social media
I thought about doing it but got scared isn't it dangerous?
Why arnt you using rest api to upload? It works perfectly
 
I thought about doing it but got scared isn't it dangerous?
Why arnt you using rest api to upload? It works perfectly
I don't know HOW dangerous it is. I see it as pretty harmless. Worst case, you just say you copy/paste the wrong profile, and nobody cares.
Found a bunch of profiles that look abandoned but have decent content posted.

Will move to either the wp rest API or wp all import once I figure out how to automatically pick a relevant picture from unsplash
 
I wish I had the time to update this more frequently.

Here are some small updates:

Posted 30 more articles.
Total articles: 50
Total indexed: 49 (one refuses to index no matter what, will let it sit a few more days then delete the slug and republish it, or just ignore it)

Too early to tell if the last 30 articles are creating any impressions yet as I've posted them all in one go today.

I do get some traffic from other search engines, like bing, yahoo, ecosia.

This is from GSC:
1661934466495.png

1661934514593.png
Semrush started picking up keywords as well.

Applied to adsense before I post more, hopefully they will approve it.

Spent 2 hours editing the design a little bit.
Changed font, headings heighy, weight, line weight, height, etc.
Made it look like healthline, clean and simple.

Got my hands on a grammarly account.
Was hoping that grammarly will try and rephrase some of the questions

Seems it's not the case, as it sees the grammas in order and doesn't want to do any edits.

I think I've spent around 10 hours on this so far.
Since I earn $30/hour on my regular job, I expect to sell it for $300+ on flippa.
My hope is it will fetch around $800 but that's just wishful thinking, will see how it goes.

As soon as I have 150 published posts, will make a decision to either sell it, or keep it.
It depends if Adsense approves it, and then Ezoic.

So far, it seems the google update didn't touch it, as there are 5 keywords on page 1 that I was tracking and they didn't move yet.

I've noticed a few patterns openAI makes when creating text.
Trying to do some rules to auto edit them, like creating <ul> lists out of 1.2.3. paragraphs openAI does.
Also trying to auto edit those First, Second, Finally paragraphs, as most articles have them and they stick out if you really look at it.

Found a relatively cheap VA, will try and test him with 40 articles if he can polish them and add an image.
My goal would be to pay him for 20 hours and have him edit 50-70 articles. Will see how it goes, he asks for $2/hour.

Don't want to invest too much as I'm not sure how much I'm going to get for it on Flippa if I decide to sell it.
It my calculations are corect and I can fetch anywhere around $500, I might scale this to doing 3-4/week with the help of a VA.

Side note, last night I was toying around with openAI.
Looked up an article on Google regarding to a topic I know.

Then I started creating questions for each paragraphs I read from the article and asked openAI to write 200 words about it.
Took me about 1 hour to create a nice article which is totally different in quality and would surely pass any review.

Sadly, it took an hour, and about 4000 tokens to have 1,700 words that I chosed out of all the words generated.
It's not yet profitable to scale.

Ok, will keep this updated soon.
I hope that in 2 weeks it can reach 30 clicks/day which would bring it to the 1,000 visitors/month mark.
 
Great, keep us updated, hope you succeed, wishing you luck.
 
That looks cool, but where did u get those questions if u have so little time?
 
Following, looking forward to see how your project pans out
 
Yes, they have special instruct models and it rocks.

I'm glad you found the reason behind your deindexation.

Try some tricks, maybe use a spinner so it won't look like openAi. You can also translate it and translate back. Many ways to make it look differently than OpenAi.


You need to learn how to do prompt design. And you need to use an instruct model like davinci-text-002 (not davinci). If you already did it and it doesn't work, try to guide it more by giving longer prompts, fine tune for your use case / niche or switch niches. Of course, it won't outperform human writers, but will serve for massive websites with 10k articles. Generally, openAi is general in terms of output, so take the generalistic approach and easy niches.
Can you give us some prompt example?
 
this is very inspiring. I'm learning how to code and working on creating my own website builder at the moment. following!
 
I spent a lot of time trying to do this in pyrhon any tips?
I haven't been able to do it yet on a functional level. Playing with some regex formats that resemble an article.
That looks cool, but where did u get those questions if u have so little time?
I talked about it in the first post.
I have a few PAA blogs, I'm doing A.I. articles on the best performing keywords from those PAA blogs.
 
following
 
Back
Top