1 · When you say SEO, you think Google. That is exactly the problem.
Ask around you: "what is SEO?". People will answer "ranking well on Google". It has become the common meaning of the word, to the point where we forgot it had another one.
The shift has a simple reason. For fifteen years, Google held more than 90% of web search in Europe. Optimizing for Google was optimizing for search, full stop. The shortcut was almost true, so we took it, and we ended up mistaking the tool for the craft.
The shortcut no longer holds. A share of search has moved elsewhere, and a whole discovery layer is being built inside AI answers. Keeping "SEO equals Google" in mind in 2026 is working with a map that stops halfway across the territory. The rest of this article is the half you are missing.
2 · The real definition of SEO: being found in an engine, any of them
SEO means search engine optimization. Optimization for a search engine, that software mechanism that answers an intent by ranking content. The word "engine" is singular and indefinite, not "Google".
A search engine is any place where someone types an intent and gets ranked results. Google is one. YouTube is another. Pinterest, Amazon, the App Store, the TikTok search bar, each ranks content by its own rules. An AI assistant that answers while citing its sources does the same job too, in another form.
Doing SEO, in its pure definition, is producing content that these engines understand, retain and surface. The engine changes, the craft stays: being found where your customer searches. GEO does not invent that craft, it adds a field.
3 · YouTube is one of the biggest search engines, and you do no SEO there
YouTube is the biggest video search engine in the world, and one of the biggest engines full stop. People search it for a tutorial, a review, a demo, exactly like on Google. It is a search engine in its own right, and most B2B brands publish nothing on it.
This is not a comfort detail. Across the 75,000 brands Ahrefs studied, YouTube mentions show the strongest correlation with a brand's presence in AI, at 0.737. Ahead of backlinks. Ahead of page count. Ahead of Domain Rating, the reference metric of classic SEO. Brand mentions in general follow just behind, between 0.66 and 0.71.
A word of caution: correlation is not causation. Big brands have both a YouTube presence and an AI presence because they are big, not only thanks to their videos. But the signal is too clear to ignore. A model that answers feeds on what talks about you everywhere, not only on your well-marked-up page. A video presence is material on an engine your competitor neglects too. SEO, on a field left empty.
4 · GEO is SEO on a new engine: the generative one
An assistant like ChatGPT, or a Google AI Overview block, does two things: it retrieves pages, then it cites some of them in its answer. Being cited is the equivalent of ranking. Not being cited is the second page of Google, 2026 edition: nobody sees you.
Retrieved does not mean cited. ChatGPT retrieves dozens of URLs per query, but cites only about half, 49.98% to be exact across 1.4 million prompts analyzed. The rest serves as background context, with no attribution. Appearing in the model's computation and being named in its answer are two different things, just like being indexed and ranking first are two different things on Google.
Calling it "GEO" makes it sound like a new craft you have to relearn from scratch. That is convenient for selling a service. In practice, you start with 90% of the kit: you already know how to produce content for an engine. What remains is learning the rules of this one, and that is where the numbers become useful.
5 · The format AI cites most: the "Best X" listicle
This engine has its own rules, and the first one touches the format. The "Best X" listicles are, by far, the page type ChatGPT cites most. Across a study of 26,283 source URLs, they make up 43.8% of pages cited by ChatGPT and 48.9% of AI Overviews. No other format comes close.
The reason is mechanical. A query asked to an AI is often a choice intent: "the best tool for X", "which software for Y". A page that compares and ranks answers exactly that shape of question. The model just has to reuse a structure that is already built. On Google, a comparison is one format among many. On the generative engine, it is prime material.
What it changes for you: an honest, up-to-date comparison on your field is often worth more for your AI presence than yet another deep article. Honest is the word that matters. A "Best X" where you place yourself first without an argument is spotted, by the reader as well as by the model. The goal is to be the reference source on the comparison, not to hand yourself a medal. This is exactly the logic we apply to our clients' content, see our SEO and content approach.
6 · Where the room actually is, and where you will never get in
A large part of the field is already taken, and out of reach. 67% of the 1,000 pages ChatGPT cites most come from sources a marketer cannot influence. Wikipedia alone weighs 29.7%. Brand homepages, 23.8%. App stores, 6.6%. You are not going to rank in Wikipedia's place, nor in Apple's homepage's place.
That leaves 32.3%. This is your playing field, and it is precise: educational pages (19.4%), reviews (5.8%), press articles (5.2%), blog posts (1.9%). Those are the only formats where content work moves the needle. The rest is authority you do not manufacture with articles.
The read is freeing rather than discouraging. You now know where not to waste your energy, and where to concentrate it: solid educational content, reviews, a press presence. A third of the pie is playable, two thirds are not. Better to play the third well than to dream the two thirds.
7 · Being cited by AI has nothing to do with your Google ranking
The most counter-intuitive point now: AI search and Google do not point to the same pages. 28.3% of the pages ChatGPT cites most have zero organic visibility on Google, that is 283 of the 1,000 pages studied, with zero ranking keywords. They are cited regularly by the model while never appearing in Google results.
It is not an isolated case. When Ahrefs compares the URLs ChatGPT cites to Google's top 10 on the same query, the overlap drops to 10%. ChatGPT does scrape Google data, yet it does not follow its choices. It is an autonomous discovery layer, with its own favorites, not a reflection of the Google ranking.
The consequence is direct. You can be invisible on Google and cited by AI. You can be first on Google and ignored by AI. Steering your AI presence while looking only at your Google ranking is reading the weather of another city.
8 · JSON-LD and AI: the urban legend
You have read it everywhere: for GEO, you must stuff your pages with structured data in JSON-LD. It is the trick you find on 9 out of 10 GEO vendor sites. It is false, and the data is clear.
Ahrefs tracked 1,885 pages that added schema markup between August 2025 and March 2026. AI citations barely moved. The variations are statistical noise: minus 4.6% for AI Overviews, plus 2.4% for AI Mode, plus 2.2% for ChatGPT. Nothing that looks like an effect.
The explanation comes from a technical test: when you look at what the systems actually read at the moment they retrieve a page, none of them use the schema. Across five systems tested (ChatGPT, Claude, Perplexity, Gemini, AI Mode), all extract only the visible HTML. JSON-LD, hidden Microdata and RDFa are ignored. The reason is simple: a language model treats the page as a continuous flow of text. An isolated JSON block, with no sentence around it, is noise in that flow, not a priority signal.
The test everyone can run: put your birth date in a Person block in JSON-LD, without ever writing it in the text, and ask a model your age. It will not find it, even after crawling the page. Write it in a sentence, it answers within a minute. Which does not mean JSON-LD is useless. It keeps all its value for Google and rich results, those stars, FAQs and breadcrumbs that decorate your search results (that is the job of technical SEO). Expecting it to feed an LLM, though, is aiming at the wrong recipient.
9 · AI Overviews: the organic click drops by 58%
The stake is not only theoretical, it is measured in lost traffic. Google's AI Overviews take 58% of clicks away from the top organic result. A year ago, the same study measured 34.5%. The trend is accelerating, and it goes one way.
One nuance changes everything, though. 99.2% of the keywords that trigger an AI Overview are informational. Commercial and transactional queries together weigh less than 10% of cases. The AI block eats the traffic of "how to", "what is", "why" content, not that of "buy", "price" or "near me".
The business read is simple. Your purely explanatory content will lose clicks, that is settled, and the erosion is speeding up. Your transactional and local content stays protected for now. The calculation of where you put your editorial effort changes with that number in mind: a "what is X" guide gets nibbled, an "X in Liège" page much less.
10 · Each engine cites its own sources
Even more disorienting, and it kills the magic recipes: there is not one generative engine, there are several, and they do not look alike. 86% of the most cited sources are not shared between ChatGPT, Perplexity and the AI Overviews. Of the 50 most mentioned sources, only 7 are common to all three. For a near-identical question, each engine picks from its own pool.
This kills the idea of a single recipe "for GEO". Being visible in ChatGPT does not make you visible in Perplexity nor in the AI Overviews. Each has its biases, its trusted sources, its way of retrieving. It is the same problem as believing that a good Google ranking automatically makes you visible on YouTube.
The practical consequence runs with the whole article: you do not target "AI" as a block, you target engines, plural, each with its own field. A brand present everywhere (educational content, press, video, mentions) has chances in every pool. A brand betting everything on one perfect page only has them in one.
11 · The surface shifts every two days, the meaning does not move
This last rule avoids a costly mistake. An AI Overview changes on average every 2.15 days. From one observation to the next, it has a 70% chance of being different, and 45.5% of the cited sources are entirely new. If you track your AI citations day by day, you will see your brand appear, disappear, reappear with no apparent logic, and you will go mad.
Except there is a constant. While the exact words, the URLs and the cited entities change all the time, the meaning of the answer stays stable. The model keeps rephrasing the same idea with different sources. The surface stirs, the substance does not move.
The lesson is sharp: do not optimize for one phrasing, nor to appear in an answer at one moment. Optimize for the meaning, the entity, the topic. Be the solid reference on a question, and you will reappear in the rotation, whatever the phrasing of the day. Chasing every variation is chasing the wind.
12 · You no longer produce for Google, you produce for engines
Put all these rules end to end and the method takes shape, with nothing that looks like a new discipline. It is SEO, read correctly.
First, get out of the dependence on a single engine. A presence on YouTube works both video search and your AI presence, the most correlated signal. An honest comparison on your field targets the most cited format. A transactional or local page stays sheltered from the AI Overviews erosion. You spread, you do not bet everything on Google.
Then, write to be read by a model: information in the text, not only in the structure. Clear answers, sourced numbers, meaning before phrasing. That is what gets cited, and it is also what reads well. You do not chase the citation of the day, you build authority on the topic.
This is exactly the logic of our study on Liège e-commerce: open data, in clear text, designed to be picked up and cited, not to decorate a page. On the technical side, we publish an `llms.txt` file that tells AI engines what we want cited, and we take care of the content itself above all. For the specific case of product feeds in this shift, see the article on the product feed as an SEA, SEO and LLM asset.
At bottom, GEO is not a reason to throw away your SEO. It is a reason to remember what it meant before we reduced it to Google: being found in an engine, any of them. You already have the craft, you just have to stop practicing it on a single field. People bury SEO with every new engine, and each time it rises from its ashes like a phoenix, because the craft does not die. It will rise again on the next engine.
This is not everyone's urgency, though. If 90% of your revenue comes from local transactional queries, the AI Overviews still spare you and GEO stays secondary for you this year. Better to know it than to pay for a "GEO" service you do not need.
