GEO: the one thing we know is that we know nothing
In this article, Carine Groz, Marketing Manager at Coexya, shares her perspective on GEO — not as a mastered discipline, but as a territory still under construction, where method is worth more than rushing.
What is changing — and can no longer be ignored
Socrates said he knew only that he knew nothing. That is roughly where GEO stands today. Everyone is talking about it, conferences are multiplying, SaaS measurement and monitoring tools are proliferating — and yet nobody can explain precisely how a LLM decides to cite one brand rather than another. Black boxes remain black boxes.
But uncertainty about the mechanisms does not erase the reality of the signals. And the signals are clear.
Web traffic from AI engines already accounts for 1 to 2% of visits on well-ranked SEO websites — a modest share, but one that is growing fast and converting better than average. The click-through rate on classic SEO results fell by 34%* in 2025: users read the AI’s answer, they no longer scroll down to find the link. And for companies that have started working on their visibility in AI engines, the first inbound leads citing an AI response as the starting point of their journey are already arriving.
Google remains the main gateway to the web — 78%* of traffic. But projections are converging: by 2028, AI engines could surpass classic search engines in terms of usage. The challenge is not to panic. It is not to wait.
*Source : Ahrefs · 2025
Tools, yes — but to measure what exactly?
Faced with this reality, the natural reflex is to look for a tool. And the market responds — fast. AYM, Gumshoe, Semrush… : platforms that promise to measure your visibility in AI engine responses, with clean dashboards and reassuring scores.
The problem is not that they do not work. It is that they perhaps answer the wrong question.
These tools send a list of generic prompts to LLMs and check whether your brand appears in the response. Result: you know whether ChatGPT cites your company when asked “what are the best French IT companies”. That is a piece of information. But it is not the one that matters.
Because your target does not necessarily ask that question. A CIO will ask: how do I challenge an IT consultancy’s workload plan? A business manager wants to know: what is the most agile solution for my business, and who can help me implement it? The procurement manager queries the AI about cost optimisation and process streamlining. Three people, three contexts, three conversations with the AI — and three radically different responses in which your brand may or may not appear.
A generic tool does not necessarily make that distinction. It measures average visibility, on average questions, for an average buyer who does not exist. You get a score. But you do not know whether you exist for the people who genuinely have the power to choose you.
That is the fundamental limitation of GEO tools on the market today: they are built to be sold to everyone, so they are not truly useful for anyone in particular.
The only reliable compass: the persona
If nobody truly masters the mechanisms of GEO, how do you move forward without navigating blind? The answer is not in a miracle tool. It lies in a discipline that marketers know well, and that sales teams have always practised: a thorough understanding of your targets.
The persona approach is nothing new. What is new is applying it systematically to the way we query AI engines — and therefore to the way we measure and produce content.
The principle is straightforward. Before sending a single prompt to a LLM, you step into the persona’s shoes: their role, their priorities, their frustrations, their vocabulary. You generate the questions they would actually ask an AI — not keywords, but real questions, worded as they would word them. Then you query the engines from that point of view, and you observe: does the brand appear? In what context? As a credible reference or merely a passing mention?
This work reveals things that no generic dashboard shows. A company may be highly visible on its mainstream topics, yet completely absent from the responses given to a technical profile asking architecture questions. It may exist for a buyer looking for a solution, and not exist at all for the decision-maker who signs off the budget. These are strategic blind spots — and they remain invisible as long as you are not asking the right questions, through the right eyes.
The other virtue of the persona approach is that it makes the diagnosis actionable. When you know that a given profile cannot find the brand on a given topic, you know exactly what content to produce, for whom, and on which question. Not more content — the right content, in the right place, for the right person.
This is also what LLMs appear to value: signed, sourced content that demonstrates genuine expertise on precise questions. AI does not select generalists. It selects those who truly answer the question being asked.
Method first, tool second
Once the method is in place, the question of tooling looks entirely different. No longer “which SaaS gives me a visibility score?” but “how do I industrialise this approach in a way that is sustainable and relevant to my context?”
Because the real difficulty of GEO is not running an audit. It is running it again regularly — every month, on an evolving list of queries, across several personas, across several AI engines, while tracking what competitors appear in the responses. Manually, that quickly adds up to dozens of queries per cycle. Without structure, it simply does not hold over time.
That is where a bespoke tool makes complete sense — not as a starting point, but as the logical outcome of a proven method. A genuinely useful GEO dashboard is not universal. It is built around the personas that matter to you, your actual competitive landscape, and the KPIs that reflect your business objectives: do I appear in the responses given to my typical buyer? Am I cited before or after my direct competitors? On which topics am I a reference, and on which am I invisible?
That level of precision cannot be provided by a generic tool by design. It is built iteratively, as the method consolidates.
The advantage will not go to the first equipped
GEO is a young, fast-moving, and fundamentally uncertain discipline. Algorithms change, engines evolve, user behaviours shift faster than the best market research. Claiming to master it would be intellectually dishonest.
But uncertainty is not a reason to wait. It is a reason to build a solid method rather than equipping yourself with tools that give the illusion of control.
The competitive advantage will not go to the companies that bought the best GEO SaaS first. It will go to those that understood that the real question is not “does my name appear in AI responses?” but “do I exist, in a credible and precise way, in the conversation my client has with AI before they contact me?”
That is a question of method. And it starts with knowing who your client truly is — and how they speak.
About the expert
A graduate of the Burgundy School of Business (BSB), Carine Groz has over twenty years’ experience in marketing and communications across the media, digital and tech sectors. Following a career marked by roles in research, data and strategic consulting at M6 Publicité and subsequently within the Webedia Group, she joined Coexya in 2022 to take charge of marketing and communications. There, she oversees the group’s marketing strategy, brand development and digital performance.
FAQ on the GEO
-
/ What is GEO?
GEO (Generative Engine Optimization) refers to all the actions aimed at increasing the likelihood that an AI engine — ChatGPT, Gemini, Perplexity, Claude — cites a brand or its content in its responses. Where SEO sought to climb a ranking, GEO seeks to be selected in a response. It is not the same race.
-
/ How does GEO differ from SEO?
SEO optimises for keywords and targets a ranking. GEO optimises for questions — often phrased in natural language, often asked in several steps — and targets a selection. The KPI is no longer the click: it is being the answer, or being cited in the answer. The two disciplines are linked — a site that is invisible in SEO will be invisible in GEO too — but they are not managed in the same way.
-
/ Why is the persona approach essential in GEO?
Because an AI engine does not give the same response to everyone. A CIO’s question about the security of a cloud architecture and a sales director’s question about sales automation call for radically different responses — and therefore different sources. Measuring your GEO visibility without taking into account who is asking the question means measuring an average that corresponds to no real buyer.
-
/ Where do you start concretely?
With your personas, not with a tool. Identify the two or three profiles who have the power to choose you, build the questions they actually ask AI engines, then go and observe what those engines respond — and whether you feature. This basic diagnosis requires no SaaS subscription. It requires method and time. Everything else follows from there.
