[Journey] Building AI content sites for SEO ($1600 MRR) — A Software Engineer's perspective

good luck on your journey. nice to see the detailed stats!
curious about what % of your revenue comes from your most successful site?
also do you have any tips on how to manage such a large number of domains?
 
I had a site similar to this. Different niche that earns me over $3000 worth of stuff just from a single site. But I have forgotten since it's all were managed and automated with OpenRouter models. It was a one specific niche and product. But a few days ago my hosting company (it was a cheap hosting company as that was one of my testing projects) and the site went down and is gone. I still own the domain but all the content is gone. I am building it back for a better hosting company. I contacted Bluehost and no reply. Sucks! As I spent some decent credit on Claude models at OpenRouter for content.

I had similar experience and one of the guys here also had similar experience and lost around 5000$ website you can find it his journey here somewhere. He ranked with GSA /Ranker X strategies a full automated site and then he got suspended and no access to files. Even for shill websites / pillars never use cheap hosting just grab Namecheap or here BHW /hosting section provider and go for it. I would try to recover content with web archive or switch domain to 301 with similar website and start GSA blasting directly to it.
 
which LLM models are you using?

Don't many LLMs create fake data or inaccurate info?
 
Congrats! Would love to hear some more about the GSC automation you have - is it giving you a daily/weekly list of keywords to target, or pages to update/topics to write etc?
 
Congrats! Would love to hear some more about the GSC automation you have - is it giving you a daily/weekly list of keywords to target, or pages to update/topics to write etc?

It's a closed-loop system. GSC data flows in daily → a scoring engine evaluates every past change (did it help or hurt?) → those learnings feed back into what gets optimized next. The key insight from 575 experiments: most SEO "improvements" actually hurt rankings. So the system learned which changes work at which positions, and only executes the high-win-rate moves. The AI agents do the content work overnight in parallel. I just check the dashboard in the morning.

Screenshot 2026-04-28 at 21.58.14.png

Screenshot 2026-04-28 at 21.59.04.png
 
It's a closed-loop system. GSC data flows in daily → a scoring engine evaluates every past change (did it help or hurt?) → those learnings feed back into what gets optimized next. The key insight from 575 experiments: most SEO "improvements" actually hurt rankings. So the system learned which changes work at which positions, and only executes the high-win-rate moves. The AI agents do the content work overnight in parallel. I just check the dashboard in the morning.

View attachment 520156

View attachment 520157
Thanks, that's great! So I guess you/ai agent then log whatever changes have been made in a database?
 
Thanks, that's great! So I guess you/ai agent then log whatever changes have been made in a database?
yes, i'm using git commits to track changes, so it can be backfill with new score / verdict logic.
 
1778467879931.png

A little update on Site3: the progress looks quite good so far.
 

Attachments

  • 1778467858444.png
    1778467858444.png
    133.8 KB · Views: 12
Screenshot 2026-07-10 at 21.09.14.png

Update on site3: survived last few Google core update
 

Attachments

  • Screenshot 2026-07-10 at 21.07.55.png
    Screenshot 2026-07-10 at 21.07.55.png
    251.8 KB · Views: 9
when ur making the 300 articles how are u doing your keyword research? are these programmatic pages at all?
 
ok nvm that was a dumb question as i read thru the thread yes it is programmatic pages. haha. but i am very curious to hear about your keyword research process
 
ok nvm that was a dumb question as i read thru the thread yes it is programmatic pages. haha. but i am very curious to hear about your keyword research process
I have two processes for keyword research, 1 is pillar template ( pick pillar -> LLM generate spokes), and two is append existing cluster: using serp data for generating missing keywords.
 
My journey building AI content sites for SEO — a software engineer's perspective



Background: I'm a software engineer. CS degree, been working in AI/ML for a while. A few years ago I got into SEO almost by accident — started applying the same engineering mindset to content. Build, measure, iterate.

That led to building a lot of travel sites. Over 100 at this point.

Most are dead. One works really well — peaked over 1k clicks/day recently. A few are somewhere in the middle. Revenue is around 1600$ per month

View attachment 515500

View attachment 515501

I'll share GSC screenshots as this thread goes on.



Why travel

Honestly, it was a good fit for what I was trying to do technically. Travel content is highly structured — attractions, prices, opening hours, itineraries. Easy to model, easy to generate at scale, easy to validate.

It's also brutally competitive. Which made it a good stress test.



What I actually built

I didn't just publish AI articles and wait. That part I figured out pretty early doesn't scale the way people think it does.

What I ended up building is closer to a content ops system:

— daily GSC monitoring per page, automated
— a scoring function that decides which pages to touch and how
— an autopilot that commits changes (content refresh, meta rewrite, internal links) based on that score
— a recheck loop at 5–7 days to confirm or revert

The content is AI. But the decision layer on top of it is engineered, not vibes-based.

That distinction matters a lot. I'll get into the specifics as the thread goes on.



What this thread is

Just a log. I'll post what I'm testing, what the numbers say, what broke, what worked.

No course. No upsell. I just think this space moves fast enough that it's worth documenting in public.

If you're running AI content sites and thinking about how to systematize it — follow along. And if you're already doing something similar, I'd genuinely like to compare notes.



More coming soon — starting with the GSC data.
Great luck on your journey! Following!
 
Back
Top