A few years ago, things were fairly predictable: you collected the semantics, wrote decent content, fixed the technical issues, built links — and over time you earned the positions you needed. Algorithms changed, but the overall model stayed the same: the crawler scans the page, indexes it, and then the classic ranking algorithm decides who gets shown at the top.

With the rise of generative search modes like Google Search with AI Overviews (SGE) , Bing Copilot, Perplexity, and ChatGPT integrations with web access, this picture has changed significantly. Above the familiar list of results, another layer has appeared — a block with an AI-generated answer that is compiled by a large language model (LLM) from multiple sources.

In practice it looks like this: your site can still formally stay in the top 10, but a noticeable share of attention and clicks goes to the generative block . On some projects I’ve seen situations where CTR for the “classic” results dropped by half even though positions barely moved. The user reads the AI summary and moves on without clicking any of the links.

That’s why the classic formula “we optimize for Google” is no longer enough. We also have to think about how our content is “read” by the LLM: which fragments end up in the AI answer, which pages look trustworthy and deep enough for the model. This is where the idea of Generative Engine Optimization (GEO) came from.

Below is a look at GEO through the eyes of a hands-on SEO specialist who has already seen, in 2023–2025, how generative search affects real projects: where it “eats” clicks and where, on the contrary, it unexpectedly helps.

What GEO is in simple terms

Generative Engine Optimization (GEO) is not a replacement for SEO and not just a buzzword. Essentially, it’s an extra layer on top of classic SEO: we still care about site structure, technical health, links and semantics — but we add one more level — how our content looks to generative search systems .

If we simplify it, GEO revolves around three questions:

  • Can an LLM understand what this page is about without extra human explanation?
  • Are there fragments on the page that can be turned into a concise, accurate AI answer?
  • Does all this still play nicely with classic SEO and organic visibility?

It’s important to fix one thing from the start: GEO doesn’t cancel basic website optimization . If a project has problems with indexing, speed, structure or links, no GEO framework will save it.

How generative search “reads” pages: a practical view

From an SEO perspective it’s easy to dive into the theory of embeddings and multi-dimensional vectors. In day-to-day work I care more about which properties of a page make it convenient for a generative answer. On several projects I kept seeing the same pattern: AI summaries were built not from the “strongest” pages in terms of links, but from those that were logically structured and clearly explained the topic .

Roughly speaking, there are several layers that matter to an LLM on a page:

  1. Semantic completeness. The page should properly cover the intent, not just list definitions. If the content only answers “what is GEO” but ignores “how to apply it in business” and “what the risks are”, it often lacks depth.

  2. Structure and readability. Clear H2–H3 headings, logical transitions, sane paragraphing. When the text is broken down by meaning, it’s easier for the model to pull out specific pieces into an AI answer.

  3. Factual accuracy. The fewer contradictions and outdated claims, the higher the chance that the domain as a whole will be treated as a reliable source. This is especially noticeable in topics related to Google and AI documentation.

  4. E-E-A-T signals. Who stands behind the content: a real specialist or a “nameless” text with no author? Are there links to studies, official sources, implementation experience? Here GEO directly intersects with Google’s classic approach to high-quality content.

On one B2B project we ran into exactly this: two articles on the same topic were side by side in terms of ranking, but only the one with clear subheadings, examples and neat links to documentation was being pulled into the AI answer. The second, more “blurred” one was ignored, even though the domains were similar in age and link strength.

Which signals matter for GEO

I wouldn’t think of GEO in terms of “new ranking factors”. In practice it’s more convenient to look at groups of signals that increase the probability of appearing in a generative answer. Some of them overlap with classic SEO, but the priorities shift a bit.

1. Topic depth and usefulness for humans

Content at the level of “definition + three paragraphs of fluff” rarely shows up in generative blocks. Models are literally “looking” for something they can turn into an answer that feels like a real consultation, not an encyclopedia summary.

The practical takeaway: if a page covers several subtopics, each of them deserves a separate, meaningful block. You don’t have to write a 15-screen “novel”, but one-paragraph sections that try to cover everything at once are working worse and worse.

2. Internal consistency

Generative systems have clearly become more sensitive to inconsistencies within a single domain. When one page claims SGE is already launched in all countries and another says it’s still just in testing, it hurts not only UX but also overall trust in the site.

On one client project we ran a simple experiment: we aligned terminology, rewrote a few outdated articles and removed some controversial statements. After a couple of updates, that domain started to appear more often in AI snippets for niche queries — and we hadn’t touched the link profile at all.

3. Topical authority

Here GEO almost entirely mirrors the logic of topical authority. If a site consistently publishes content about SEO, AI search, analytics and keeps its materials up to date, it has a higher chance of being used in generative answers than a resource where one GEO article sits next to anything — from crypto to recipes.

It’s not about volume, it’s about connectedness: clusters, internal links, logical navigation, a sane blog structure .

4. Structural clarity

An H1 that honestly reflects the topic of the page; H2s that look like answers to separate questions; H3s that expand on the details. Plus clean lists where they really help rather than “for decoration”. This is convenient both for people and for the model that tries to assemble a concise summary from the text.

How I approach GEO in practice

A theory-only approach doesn’t get you far, so a couple of years ago I started building GEO logic directly into the content workflow. Not in every article and not always perfectly, but on a number of projects it’s already making a noticeable difference in engagement and visibility.

Step 1. Start not with keywords, but with questions

The classic reflex is to open a suggestion tool and collect the semantic core. For GEO I first write down the questions people ask after reading the initial AI answer:

  • “Okay, I get what GEO is, but what should my business do specifically?”
  • “How do I measure this in numbers?”
  • “What happens if I change nothing?”

Those follow-up questions then shape the structure: some of them become H2s, some H3s, some go into a “common mistakes / risks” block.

Step 2. Design a structure that’s convenient for LLMs

Here it’s quite down to earth:

  • H1 — a clear title, without clickbait games;
  • H2 — large conceptual blocks (what it is, how it works, how to apply it, how to measure the impact);
  • H3 — steps, details, examples, FAQs.

On one project, after this kind of restructuring — without fully rewriting the text — we saw higher engagement (Engaged sessions) and a modest strengthening of positions for several queries related to AI search features.

Step 3. Work with facts and references

Generative search increasingly “highlights” the sources it relies on. If an article contains numbers, dates or statements about how algorithms work, I try to tie them either to official documentation or at least to recognizable, trusted resources.

That doesn’t mean turning every text into an academic paper. It’s more about honestly separating: where I’m retelling someone else’s report and where I say “based on our own experience on projects X and Y, the situation looks like this”.

Step 4. Building in E-E-A-T

GEO without trust in the author and the domain doesn’t really work. That’s why I’m in favor of:

  • signing articles with real authors;
  • briefly explaining what this person actually does and why they’re qualified to talk about the topic;
  • referring to real cases and tasks instead of abstract “clients from different niches”.

Yes, this is a bit more effort than running an “anonymous blog”. But these signals are exactly what help get out of the “suspicious AI content” zone that Google talks about in its quality guidelines.

GEO audit of existing content: where to start

Most sites already have a solid “layer” of content written before generative search went mainstream. Throwing everything away is a bad idea, but revisiting key pages through the GEO lens is definitely worth it.

What I usually check

  • Freshness. How up to date are the examples, dates, mentioned products and interfaces?
  • Depth. Did we answer the next steps for the user, not just the basic question?
  • Structure. Can you understand what the article is about and what the reader will take away just by scanning the headings?
  • Connectedness. Does the piece fit into the overall cluster, or is it living on its own?
  • Expertise. Is it obvious from the text that it was written by a practitioner, not a generic copywriter?

After this kind of audit, some articles only need updating and repackaging, some should be expanded, and a few are honestly better moved to the archive or closed from indexing so they don’t drag down the domain’s overall quality score.

GEO for different types of pages

It’s important to remember that not all content on a site is equally “interesting” to generative systems. In some areas GEO barely matters; in others, it makes a very noticeable difference.

Long-form expert articles

These are the primary candidates for appearing in AI answers. Here I focus on:

  • thorough coverage of the topic;
  • clear, real-world examples from practice;
  • a logical structure that lets you “cut out” a fragment as an answer;
  • strong internal links to other materials in the cluster.

How-to guides and instructions

Guides work very well alongside generative search if they include:

  • clear, ordered steps;
  • concrete phrasing (“click here”, “open this report”);
  • examples of “what to do if something goes wrong”.

Commercial pages

On service pages GEO tends to play a supporting role. But even here it matters to have:

  • a properly structured description of the service rather than a generic text “about everything and nothing”;
  • clearly explained benefits and processes (what exactly you do as part of the service);
  • links to expert content (cases, articles, examples).

For example, the SEO audit page performs much better when it’s backed by several in-depth materials on technical SEO, analytics and error handling.

How to measure GEO results

Right now we don’t have a perfect tool that says “click here to see how many times your site appeared in AI answers”. So we have to combine several approaches:

  • tracking CTR and impressions for queries where generative blocks already appear;
  • monitoring engagement for key materials in analytics;
  • doing manual checks to note which articles are more often picked up in AI summaries;
  • comparing page behavior before and after structural/content changes.

It’s not the easiest part of the job, but without this observation GEO quickly turns into just another buzzword that doesn’t change anything.

Risks of the GEO approach

Like any new practice, GEO has its risks — and it’s better to call them out explicitly.

  • Chasing “perfect” text. Over-polished, emotionless content often looks like AI generation and can trigger distrust both from users and algorithms.
  • Over-optimizing the structure. Trying too hard to “please the model” can break basic SEO: keywords, internal linking, logical block priority.
  • Ignoring real-world experience. If you stick only to theory and documentation summaries, your content will be weaker than articles backed by real cases and numbers.

Conclusion: GEO as a natural next step, not a replacement for SEO

For me, GEO is not a separate discipline, but a gradually growing layer in the everyday work of an SEO specialist. Generative search is already affecting traffic and user behavior, and ignoring it means consciously leaving part of the organic potential “on the table”.

The fundamentals haven’t gone anywhere: technical health, sensible structure, quality links and clear content are still mandatory. GEO adds an understanding of how models “think” and which pages they see as trustworthy, useful and convenient to quote.

In short: we still create content for people — but now we regularly ask ourselves one more question: will an LLM be able to retell what we wrote honestly and without distortion?

Andrii, SEO specialist