Indexed but Empty: How an AI Swarm Chose Our SEO Topics and Wrote the First Post

Our blog was indexed, technically sound and completely empty. Google knew the address existed, but had nothing to offer a reader, so there was no organic traffic to speak of. This post describes how a swarm of agents solved that on its own: it picked the topics, wrote the first article and set the rules every following one is built on.

Indexed doesn’t mean visible

Indexing only means the search engine knows your page exists. Ranking starts when the page actually contains text that answers a specific question someone is typing into a search box. A blog with no articles has no such entry point, so however clean the technical side is, the result is zero.

If your site has a blog that hasn’t received a single new post this year, you’re in the same position. The section exists, the sitemap is tidy, and nobody arrives from search because there’s no content to bring them.

The first job wasn’t writing, it was choosing topics

We gave the swarm a narrower task first: work out what is even worth writing about. The criteria were threefold. A topic has to have low competition, it has to be relevant to the Croatian market, and it has to be something we can back up with what we actually know.

In practice that means we don’t fight for broad terms like “AI automation”, where we’d be competing with everything ever written on the subject. Instead we take narrower questions that a business owner or an operations manager genuinely types into search when something specific is bothering them.

There’s a rule alongside it that keeps the system from going in circles: the agent maintains a list of already published articles and checks every new topic against it. If the topic is already covered, the agent has to find a fresh angle or propose a different one.

From topic to published article

The source of truth

Every business fact sits in a single file approved by a human — services, company details, terms of engagement, guarantees. The rule is unconditional: what isn’t in there doesn’t exist and can’t be written. No invented clients, metrics or references. When a topic calls for proof we don’t have, the article is written on principle rather than padded with made-up numbers.

Editorial rules

Next to the facts sit the rules for how we write: the brand’s editorial standard and a separate standard for the Croatian language. They define tone, article structure, how loanwords are handled, and they rule out literally translated English metaphors that leave the reader decoding a sentence. These rules aren’t a recommendation — they’re part of the system prompt of every writer in the swarm.

Validation and publishing

Finished text passes through a validator before anything goes out, and then an agent publishes it. The system’s work is visible in a public activity log, so publications can be followed in real time.

Why this is proof, not a demo

The article you’re reading is a product of that same chain. It wasn’t written to show how the system could work — it exists because the system worked. That’s the strongest form of proof we currently have, because we have no public client references and we’re not willing to invent them.

We’re honest about expectations too. The first organic clicks realistically arrive in two to three months, and they grow with the volume of published content. A system that writes on its own doesn’t speed up Google, but it removes the reason blogs stay empty: nobody has the time to sit down every week and write.

What your company can take from this

The same pattern applies anywhere a company produces text according to rules that repeat — quotes, procedures, reports, answers to standard enquiries. The value isn’t that “AI writes text”. The value is that the rules get written once and the system follows them every time, without the drift that creeps in when someone is busy or when a new colleague takes over the job.

The prerequisite is always the same: there has to be a source of truth about your business and written rules for how it’s described. That part is human work and it’s done once. Everything after it can be handed to the swarm.

If you’re thinking about where the same kind of automation would pay off in your business, start with the process you repeat most often by hand. For us that was content. For you it’s probably something else, but the logic is identical.

Frequently asked questions

Why publish one article instead of ten at once?

Because the plan is tied to topics we can back up with facts from our own source of truth. We defined 3–5 low-competition topics and publish them at the pace each topic deserves. Ten articles written in a single day wouldn’t be better proof of capability — just more volume.

How do you know the agent won’t invent a fact about the company or a client?

Every business fact — services, prices, guarantees, company details — lives in a single file approved by a human. Anything not in that file, the agent is not allowed to claim. If a topic requires a number we don’t have, the article is written on principle and educationally instead of inventing one.

When do you expect the first organic traffic from these articles?

Realistically, within two to three months. Indexing is fast, but ranking takes time and a volume of content that builds up over weeks. Anything promising clicks within a week usually means paid ads, and that’s a different conversation.

Does a human do anything in this chain?

A human sets the rules and approves the facts — the editorial constitution, the language standard and the list of verified data. Topic selection, writing, validation and publishing all run through agents.

Can the same approach work on our site?

Yes, but only once there’s a source of truth about your business and written rules for how it’s described. Without that, the system has nothing to stand on and starts inventing, which is worse than an empty blog.