BACK TO SEO: EXPERIMENTING WITH AI AND SCALED CONTENT

Welp, if you read my about page you know I’m coming back-ish to SEO. I say back-ish because I never truly left. I just stopped deploying sites for a while.

But now I do feel like deploying more sites. It is much faster to get all the boring work done with AI, and I can focus on the good stuff I like doing, like watching the websites die a slow and shameful death.

In 2015 they told me to publish 100 posts a month

This time… I am doing things a lot different.

I remember back in 2015, tons of SEO people like Neil Patel and some other dudes were all talking about how publishing in volume was important. I remember receiving the advice of posting 100 posts in my first month, and I was like “how the hell is that even possible?”. The max I was able to do was 30.

Everything felt so time consuming, from formatting the post with the right headers to making sure the content read well and had a nice flow. Well, that can be mostly automated now. I say mostly, because there are a lot of things we really shouldn’t fully automate, like content creation.

What counts as scaled content? There’s so little clarity

I understand Google says it is fine to use LLMs and LLM content as long as it is reviewed properly. But I feel there is also a lack of transparency on what and how we are allowed to scale up content, even a little:

Am I not able to use templated pages?
What counts as a template? How much % of “likeness” counts as a template?
Can I use my own database and my own generated data to create content, like I did on indiedex.gg?
Is it a matter of how many pages I have published?

There is so much I still don’t know, and so little clarity out there. Like most of SEO, it comes down to experimentation and our own interpretation of what counts and what doesn’t.

Experiment #1: indiedex.gg and its 960K generated pages

So let’s experiment. The first one is indiedex.gg. It has over 960K pages: ~120K games x 2 pages per game x 4 languages. All of them are generated from my own data processes on enriched game data from Steam:

I generate my own pre-calculated similarities and data tagging for each game, based on my system’s understanding of what the game is and what other games are like it.
I then generate some LLM content to figure out “who would like this”, “who would not like this” and “what should be recommended”.
Then I translate that to Spanish, Portuguese and French.

Does it fall under scaled content abuse?

According to Google’s spam policies, do I fall under scaled content abuse?

My app does something nothing else out there does, it does it well, and it provides value to users. I have received many positive comments (some of which are on the indiedex homepage), and I can see in my analytics that the average session reaches 2 minutes, with users averaging 4 pages. And that is the average: some outliers spend over half an hour playing with the discovery engine, exploring all kinds of games and making lists.

Noindex for pages without enough useful data

One of the things I have done is programmatically set noindex on pages that I feel don’t have as much useful data as other pages. I don’t mean content, I mean data.

It comes down to how I enrich the data: when the data I accessed before the processing didn’t have high enough quality, the system couldn’t make the best judgement. To me, that means those pages wouldn’t provide as much value.

The results so far

For now, I will leave indiedex as it is. The SEO looks really good: I launched just a little over 2 months ago and already have over 1K organic visitors per month.

indiedex.gg in Google Search Console, last 3 months. Click the image for full size.

That spike at the end of the chart was likely the September Google update rolling out. It seems the August update had flattened my site, and then the September update fixed it. Around then I also added more content that gets generated per page, so the pages become more valuable, but I doubt Google had it crawled by then.

I believe Google will kill this site. It is unfortunate, but just based on the amount of pages, I can see their automated systems picking it up as a red flag.

Well, anyways, back to what matters. What am I doing?

15+ sites, each one a different experiment

I have deployed over 15 sites over the past couple of weeks. I won’t be sharing their identities at this time, but each site is a different experiment, ranging from:

different techniques to write content with LLMs
different ways to structure pages
different review processes
and even a workaround to Google’s scaled content policy, which I will be sharing later

My experiments are not super scientific right now. They are mostly a hobby, and this initial batch is really for me to find myself in the SEO space again, after taking a few years to focus on my software engineering career.

As I make more sites and monitor what is happening to each one, I will probably make my tests more sophisticated and repeatable. For the things that work and don’t work, I want to be reporting back on this blog.

What I’ll write about in the meantime

SEO does take a while though, so in the meantime I’ll probably write about some other things that I find:

resources and techniques
easy initial backlinks
how to get good backlinks without spending too much time on outreach
random thoughts about the industry
how to make some money with SEO as someone who is not opening an agency

My new pet project

Every engineer needs a pet project, and I feel AI took the fun out of building software, so this is my new form of pet project. What can I learn that will make me a better marketer and business owner, so I can use those skills to protect myself from the bloodbath that AI is starting to cause in the software industry?