Opens in a new tab
A query typed on the screen: "What is AI overview?" - No Digital Brain article cover - Generative Engine Optimization (GEO): what we know so far

Generative Engine Optimization: what we know so far

What changes between SEO and GEO? They are closely linked: both aim to make content visible and useful. GEO looks at how AI selects and cites sources, in a landscape that is still very much evolving.

What you'll find in this article

Meanwhile, this: we still know very little about Generative Engine Optimization (or GEO) – at least in any solid, structured, measurable way.

We admit this honestly, and we believe it applies more or less to everyone who writes about it and presents themselves as an expert. After all, the very first official guide published by Google on the subject only came out in May 2026. You would need a time machine to claim total mastery of it today. But we won’t dwell on this any further; the web is a strange place.

Talking about GEO (in relation to SEO) is certainly complicated. What holds true today, in November 2026, could prove useless before long. This is a field that evolves at an almost frightening speed; Google itself constantly updates its guidelines.

For this reason, in our article, you won’t find truths set in stone. The article is rather a collection of currently confirmed facts, and of hypotheses that we ourselves test in the field every day – and that can always be disproved, or corrected along the way.

The topic, however, is not marginal. Google states that AI Overviews has surpassed 2.5 billion monthly active users and that AI Mode is over a billion. If you have a website, someone is already getting answers that draw from pages like yours. So let’s see how we can work with it.

So, what changes between GEO and SEO?

The term GEO originated – curiously, perhaps – in academia. It was in fact introduced by a scientific paper by a group of researchers, including some from Princeton. The acronym then passed to consultants, along with its sister term AEO (Answer Engine Optimization), and today it generally refers to the work done to appear in the answers of AI Overviews, ChatGPT, Perplexity and similar tools.

Google’s position, in its guide, is clear-cut: optimizing a site for generative search features means optimizing for the search experience. So from its point of view, it is still SEO. AI Overviews and AI Mode, after all, rely on the same ranking and quality systems used for the SERP. To appear as a supporting link, a page must be indexed and eligible to be shown with a snippet, as with any other result.

The only real difference, to date, is the new feature introduced by Google at the end of August directly in Search Console. Today, whoever manages a site can choose whether or not to have it appear in generative AI answers, and therefore also whether they want the site to be used as a source or not. Those who do not give their consent will receive neither traffic nor impressions from LLM systems.

Yes, but so, really, what changes?

For Google, in substance, little; measurement, however, is different. Search Console offers a dedicated reporting system for performance in generative AI features, and it is – to date – the only tool capable of doing so.

ChatGPT, Perplexity and the other assistants do not yet provide us with any data, so any GEO strategy on those platforms has to be handled with greater caution.

How semantic systems read information

Generative search AI does not respond only with what it learned during training. On this front, Google talks about two techniques:

  • retrieval-augmented generation (RAG)
  • query fan-out

RAG is a grounding technique, which means: the system retrieves relevant, up-to-date content from the search index, uses it as a factual basis (hence grounding), and generates an answer supported by links to the sources. In short, Google uses indexed sites the way it always has, then reworks and summarizes them.

Query fan-out, on the other hand, is the mechanism that expands the information retrieval phase before the answer is generated. Query fan-out means “expanding like a fan,” and it works like this: the model breaks the question down into subtopics or sub-questions (e.g. “price,” “requirements,” “comparison,” “risks,” “official sources”) and generates several related queries.

You can analyze query fan-out, or at least part of it, directly on the platform. If you look closely at how Google’s interface behaves when it generates an AI-based answer, you will notice some clear signs of this “behind the scenes” work:

  • the intermediate searches used to scour the web
  • the sections into which the answers are organized
  • the follow-up and further-exploration suggestions
  • etc.

We can’t talk about a complete audit, but it is certainly an interesting starting point.

Some tips for GEO-optimized content

Finding your way through the countless solutions that have circulated online in recent months is not easy. For this reason, the tips that follow come directly from Google’s official guide and from the first scientific studies on the subject. We have also filtered them to account for everything that – by the search engine’s own admission – we can safely stop doing.

What should you invest in, and what can you stop doing?

  1. Organize your pages for the people who read them

Among the most widespread tips is to break content into small, self-contained blocks, designed to be extracted by AI. This is the so-called chunking: each section of an article thus becomes a self-sufficient unit of content, clearly titled and understandable even outside the context of the page.

Here, it may surprise you, but Google puts it among the things you can ignore. Fragmenting the text serves no purpose, because its systems are able to understand multiple topics on the same page, and still show the user the most relevant part.

Nor does it make sense anymore to talk about an ideal length. Short pages and long pages work the same way; it all depends on the audience, the type of content and the topic.

Of course, though, organizing a text is still useful. But it serves more as good editorial practice. Paragraphs, sections and headings that make the page easy to follow for those who read it, and that remains – when all is said and done – the only thing that truly matters.

  1. Include real data and field experience

On this point, Google’s guide insists more than on any other. The advice is to offer a point of view that other sources don’t have. An example? Talk – on your blog – about what the professionals on your team think; talk about what one of your clients said, or about a very personal insight concerning your business.

In this sense, Google distinguishes between commodity content, that is, based on common knowledge that anyone could write (e.g. “What is GEO?”), and non-commodity content, such as a dissertation on why calling yourself a GEO expert today doesn’t make much sense.

The former, Google observes, are also the ones a generative model can produce on its own. Non-commodity content is yours alone. Or belongs to whoever writes it. In this sense, company blogs should be structured as real editorial projects, or research projects. And we know it is a significant investment. But it is the only one that truly works right now.

On the scientific side, the GEO study cited earlier found that adding source citations, quotations and statistics increased content visibility by more than 40% across various queries. A figure to be read with caution, of course: these are still experiments. Yet it tells us a lot about how content works right now.

  1. Focus everything on authority (E-E-A-T)

Here we pick up on what was said earlier. Creating non-commodity content, in fact, is the easiest way to feed the E-E-A-T model (Experience, Expertise, Authoritativeness, Trust). A generative system that has to select its sources looks for texts written by people who know the subject they are talking about well.

According to Google’s page on helpful, reliable content, the E-E-A-T model is not a ranking factor in itself. Also because Google has no tools to verify expertise on a given topic with certainty. It can, however, estimate it based on certain parameters:

  • the presence of a byline and an author page
  • the citation of sources and fact-checking
  • the presence of evidence of field experience
  • consistency with the site’s topics
  • etc.

The weight of E-E-A-T nonetheless increases for topics that can affect people’s health, financial stability or safety. These are the so-called YMYL topics, “Your Money or Your Life”. If your site operates in that area, you will necessarily have to keep these parameters in mind.

  1. Finally, do SEO the way you always have

In short, keep investing in SEO. Because GEO, however much it sounds like a revolution in search engines, rests entirely on classic optimization. A generative model cannot cite what the crawler is unable to find, crawl, understand or index.

If a site has structural technical problems, prohibitive loading times, broken navigation paths or a confusing SEO architecture, it will remain invisible anyway – both to traditional list-based results and to the mechanisms that retrieve AI-generated answers.

So keep taking care of the usual aspects. Work on a clear information architecture, create consistent internal links, make sure loading is fast, and so on. The common temptation is to file classic SEO away. In reality, GEO is an extra layer, almost an additional piece of work. If you do the basic SEO well, the citation will surely come.

The web changes too fast. Rely on someone who keeps up!

Until recently, Google said that no additional requirements or special optimizations were needed to appear in AI Overviews and AI Mode. Today there is a dedicated guide, and even a specific report in Search Console. A few months from now something might have changed again and, somewhere, someone (like… us?) will have to keep an eye on the sources and tell effective practices apart from baseless hypotheses.

The web moves fast, but moving in the right direction takes method and experience.

If you want to understand how to adapt your site to this new scenario without chasing every single rumor, let’s have a chat. We’ll grab a coffee (even a virtual one!), look together at where you stand, and think through how to move forward.

No Digital Brain promises neither magic formulas nor algorithmic shortcuts. But it works alongside you to take you where you want to go. Write to us today and find out what we can do for your project.

Daria Costanzo Avatar

Would you like a standalone website?

Contact us to design a customized plan, with no middlemen or sales filters.

Contact us