SEO optimization

I asked an AI assistant how to optimise my website. It gave me 304 perfect pieces of advice. And completely blind ones.

Preliminary note: The AI initially generated a list of 304 technical tips. I decided not to publish them. What follows is the explanation for that decision — and, perhaps, an argument more useful than all 304 points combined.

Let me tell you a small story. A true one.

I opened ChatGPT and asked: "What are the most common SEO mistakes made by websites that publish AI-generated content?"

It answered impeccably. Structured. With bullet points. With examples. It explained that AI-generated text tends to be generic, that it lacks genuine topical authority, that Google is getting better and better at detecting it, that E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness — cannot be simulated but must be demonstrated through authentic human experience.

An excellent answer.

Given by a tool that doesn't know it's talking about itself.

The tool that cannot see itself in the mirror.

This is the first level of irony. The AI assistant explains to you the SEO vulnerabilities of AI-generated content with no trace of awareness that it itself is the exact source of the problem it is describing. It's rather like asking a drunk man how to avoid drink-driving and he gives you a theoretically impeccable answer, complete with statistics, recommendations, and an authoritative tone.

Technically correct. Existentially blind.

Researchers at Aalto University documented precisely this mechanism: users who consider themselves more familiar with AI display the highest degree of overestimation of their own performance — a complete reversal of the classic Dunning-Kruger effect.

But the same effect works in the opposite direction, at the level of the tool itself. The AI assistant describes with clinical precision the vulnerabilities of a type of content — and has no mechanism that would cause it to recognise the fact that it produces exactly that type of content. Not out of malice. Out of architecture.

And if you haven't heard of AEO, you're already losing ground. I noticed the first sign on a Tuesday morning, when I opened Analytics as usual. The traffic was there — impressions were rising, positions remained stable, older articles were holding their places in the top 5. On paper, everything looked fine. But something didn't add up: actual visits had been declining slightly, steadily, for several months. Not dramatically. Not catastrophically. Just a thin thread slowly unravelling from a ball I thought I was holding firmly in my hand.

I began looking for technical explanations. Loading speed, Core Web Vitals, indexing issues. Everything was in order. And yet, something had changed. Not in the site. In the way people search.

One evening, I ran a simple experiment. I took 10 questions from my field — questions for which I knew I had well-written, well-ranked articles. I put them one by one into ChatGPT. Then into Perplexity. Then I asked Google Assistant. The answers came immediately, complete, with sources. My site didn't appear in any of them. Others did — smaller sites, with less authority, with less traffic. But structured differently. Written differently.

That was when I understood I wasn't losing an SEO battle. I was losing a battle I didn't even know I was playing.

It's called Answer Engine Optimisation — AEO. And it is, at least for me, the most significant change in online visibility over the last 10 years.

It is not a term invented by a marketing agency to sell new services. It is a measurable reality: according to Sparktoro data, nearly 60% of Google searches today end without a single click. The user receives the answer directly — on screen, by voice, through AI — and leaves. Without visiting anyone. Without leaving a trace in Analytics.

I spent several months studying how this new world of search functions. What follows is everything I learned — structured, with real examples, with the mistakes I made, and with what ultimately worked.