Home / Our view on GEO
Part of the industry thinks GEO is an empty buzzword invented by people selling training courses. I understand the irritation, and I disagree. GEO shares about 80% of its practices with SEO. It is the remaining 20% that now decides who gets cited and who does not.
Their argument is solid, and we should start by acknowledging it. Well-structured content, real authority, a technically clean site: these three pillars make you rank on Google and get you cited by an AI. Nothing new. And within eighteen months, the word GEO has indeed become a convenient sales pitch for reselling audits to worried marketing teams.
The reasoning goes wrong at the conclusion. From “GEO borrows a lot from SEO”, they conclude “GEO does not exist”. That is not the same thing. Local SEO borrows most of classic SEO, and nobody concludes it does not exist. The same goes for e-commerce SEO. A specialism is not defined by what it shares, it is defined by what it adds.
My position fits in one sentence: GEO is the AI layer of SEO, with its own metrics, its own formats and its own trade-offs.
Three gaps, in the order in which they cost you visibility.
AI crawlers do not behave like Google's.
They do not execute JavaScript the same way, do not follow the same crawling rules, and get filtered by security protections that let Googlebot through without any trouble.
A common result: a site that ranks perfectly on Google, invisible to AI platforms, without anyone noticing. No SEO tool reports this.
A written answer is not read like a results page.
On Google, a hundred links coexist and a patient user eventually finds you. In an AI answer, three brands are named and only one is chosen. There is no second page.
So the notion of position changes in nature, and SEO metrics do not cover it. You need to measure whether you are cited, where in the text, and whether a link points to you.
What you produce is not in the same place.
In SEO, most of the effort goes into your own site. In GEO, the decisive part happens elsewhere: directories, comparison sites, discussions, specialist platforms.
It is the most counter-intuitive gap for an experienced consultant, because it moves the work to ground that SEO treated as secondary.
The third gap is not a hunch. It is what we have consistently observed since Cockpyt AI started measuring.
More than 60% of the sites we analyse receive citations whose source is not their own site. AI platforms talk about them based on pages they do not control.
An independent study by Stacker and Scrunch in March 2026, covering 87 pieces of content, 30 brands and 8 AI platforms, reaches a comparable order of magnitude: in 64% of cases where an AI cites a brand, it relies on third-party sites, and in one case in five those third-party sites are its only source.
Two different samples, two different methods, one converging result. That is why I consider this point established rather than a working hypothesis.
The practical consequence is uncomfortable for many brands: you can publish the best content in your market on your own site and still be missing from the answers, because AI platforms look elsewhere. What others say about you often weighs more than what you say about yourself.
AI Overviews has been live in more than 120 countries for two years. In France, the rollout only started on 22 July 2026, delayed by the dispute over press neighbouring rights.
During those two years, the French market watched US data from a distance. That is over, and the worry arrives all at once, on a subject other markets had time to learn gradually. This concentrated panic explains part of today's commercial noise, and therefore part of the irritation of the SEOs who call it bullshit. They confuse the noise with the substance.
These four points are convictions, not measurements. They are dated, and I will correct them publicly if the facts prove me wrong.
AI traffic will rise, search traffic will fall. Not a sudden switch, but a steady erosion on the search engine side and steady growth on the assistant side. The volumes will not cross for a long time, and that is not the point: what matters is the direction of both curves.
There will be clear winners and clear losers, and few people in between. This is the point I think is most underestimated. A results page spreads visibility across ten players. An AI answer cites three. The concentration is mechanical: brands that settle into these answers will gain a lot, and those that are absent will not lose a little, they will drop out of the conversation.
Visits from AI platforms arrive with a more advanced intent. The user has already got their explanation, compared options and ruled out candidates. When they click, they are no longer discovering, they are checking. Lower-volume traffic can therefore be worth more than higher-volume search traffic.
GEO will gradually move away from SEO. The shared 80% will shrink. The more conversational interfaces become autonomous, the more the mechanics diverge, and the more the skills separate. I think many of those who call it bullshit today will change their minds within two years, without necessarily saying so.
A position without a method is just an opinion. Here are the four choices that shape our measurement, including when they cost us commercially.
Every score comes from a real query to the AI platforms at the time of the scan, never from a stored database of answers. AI platforms look for sources every time a question is asked: an answer recorded last month says nothing about your visibility today.
We capture the answer as it is shown to a logged-out user, the one your customers actually see, wherever the platform allows it reliably. Every check runs with no account, no conversation history and no personalisation, so the same question is asked the same way on every scan.
AI platforms are not deterministic: the same question asked twice does not get the same answer. With one check a week, a single unlucky draw is enough to move your curve. One check a day gives thirty measurements per question per month. The headline figure remains a 7-day average, so you read the trend rather than the noise.
If our method changes, the change is recorded in your history. A series of measurements is only worth something if you know when the instrument changed. Comparing two periods measured differently does not produce a trend, it produces an illusion.
I have been an SEO consultant since June 2016, in Nice. Before that, a master's degree in digital communication, a BTS in marketing, and above all a travel blog launched in 2014 that served as my lab: trying to make it take off is how I discovered SEO, then decided to make it my job after training with Olivier Duffez.
Since then, more than 300 clients supported as an SEO consultant in Nice, mostly very small businesses, SMEs and freelancers. Highly competitive, unglamorous sectors: healthcare, real estate, CBD, sophrology, weddings. That is where you learn what really works, because there is no budget to make up for mistakes.
Cockpyt AI was born on 14 February 2026, with Laurent Séjourné, who builds the technical side. The tool comes from a need I could no longer cover by hand: my clients had been asking me for months whether they were cited by AI platforms, and I answered with screenshots taken one by one. It was neither measurable, nor reproducible, nor defensible.
Measuring is the first step, and an essential one, but a dashboard fixes nothing. That is why the GEO audit produces a prioritised action plan rather than a report.
Since October 2026, three modules have shortened the distance between the diagnosis and the fix. The Deep Scan checks whether your presence on a question really holds. AI Perception compares what AI platforms say about your brand with what is true, and points to the sites to fix. TheAI agent answers your questions about your own data. What comes next will be written here when it is delivered.
Co-founder of Cockpyt AI, SEO and GEO consultant in Nice since 2016. This page states my personal position, not an industry consensus. You can challenge me at [email protected].
The best way to settle the debate for your own market: look at what AI platforms answer today when asked about your sector.
Run a free scan 14 days, no credit card, no commitmentPRODUCTS
RESOURCES
COMPANY
