For fifteen years, I have ranked websites on Google with the same obsession: understanding what works, testing it, measuring it, then starting again.
I took a very “reverse engineering” approach to SEO, separating Google’s recommendations from what actually worked.
Then ChatGPT arrived and people started talking about GEO. The prevailing view suited me quite well:
“GEO is just another layer of marketing on top of SEO.”
It was the easy view, so I went along with it.
That conviction held until this channel became large enough to affect my own mountain guiding business, with effects comparable to those of SEO.
At the same time, my web clients began challenging me more and more often about their visibility in ChatGPT, Gemini, and similar tools.
Finally, the arrival of AI Overviews in French search results changed my mind for good.
From there, a question began to take shape:
How do you become the go-to source that generative engines recommend to your target audience?
I started tackling that question with my SEO reflexes: observe, collect, compare, test.

Which metrics should we track for GEO?
Traditional SEO reasoning is straightforward: a query expresses an intent and leads to a results page where links are ranked by relevance. You compare positions with monthly keyword search volumes to estimate SEO traffic, or use the number of clicks and CTR directly.
In answers from a generative engine, whether in an AI Overview or an answer from ChatGPT, clicks appear to be almost nonexistent.
Position alone does not indicate relevance. It must also be read alongside what the AI says about the cited source.
Finally, we have no data on how often any given question is asked. This is hardly surprising, because questions vary far more than keywords.
But can those different questions still reflect the same underlying intents?
Since July 2026, Cadrank, my in-house tool, has detected AI Overviews in its search result screenshots. This gives us a practical way to track the visibility of a brand or website in those answers.
For ChatGPT, I chose to generate questions that represent a website’s content, based on its main keywords.
In both cases, I use an evaluation framework that is likely to become standard across GEO tracking tools: brand mentions, site citations, position, sentiment, and sources used. From this data, I generate a metric that tracks both the overall trend and the movement of each criterion.
From SEO to GEO: making the shift
For the past two years, I have added patterns to some content that seem to increase a brand’s website presence in generated answers. Until now, I was still working by feel: I could see differences, but I did not have enough data to attribute an effect to a specific change.
I now have the cognitive and technical building block I need to iterate and improve. Spoiler: I still have not found the magic formula. The foundation remains the same as in SEO, though: be present in the index and rank well in the results.
Here is the process:
Check access before rewriting. A rewrite cannot fix content that is missing from the HTML before JavaScript runs, or a firewall that blocks generative engine crawlers. The same problem arises if the page is not included in the right index. Source code and server logs become editorial tools again.
Measure before optimizing. A flattering screenshot is not a metric. You need to define the questions that represent a real decision journey, observe the answers, keep a history of them, and accept their variability.
Compare competitors and sources. An AI can learn about a brand from its own website, a media outlet, a forum, a comparison page, or an institutional source. Off-site work goes beyond counting backlinks. You need to understand where the brand is described, by whom, and in what context.
Connect measurement to an action. A score without a backlog changes nothing. A decline might lead you to fix a page, clarify an offer, publish evidence, earn a third-party mention, or do nothing if the signal is too weak. Much of this work remains experimental. Unlike SEO, we do not yet have enough accumulated experience to make every trade-off and set priorities confidently.
Cadrank tracks presence in AI answers
I therefore extended Cadrank to answer one very practical question: when a prospect asks ChatGPT about a company’s business, does the AI mention it, and what does it say?
Every week, Cadrank submits representative questions through ChatGPT’s public interface, evaluating the generative engine in an environment as close as possible to real usage. It records whether the brand and its exact domain name appear, where they appear, the associated sentiment, the competitors mentioned, and the sources selected. It stores the answers and combines these signals into a score out of 100.
The protocol makes uncertainty visible over time. It separates an isolated answer from a trend and shows whether the actions taken genuinely move the score.
I am embracing GEO because my clients need to be present while a prospect understands their problem, compares solutions, and forms an opinion. It is also a natural extension of my SEO expertise. Mastering this topic helps me maintain a clear, coherent roadmap for organic acquisition. That decision journey now spans Google Search and its AI Overviews, as well as other answer engines whose usage is rising every month.
This work also requires more than a one-off audit. Technical work, content, external sources, and measurement need to move forward together, at a pace the teams involved can actually ship.
In a spirit of openness and shared learning, I will continue to publish what I observe, including when the results contradict my initial intuition. Subscribe if you would like to follow the journey.
AI, SEO and Web strategy session
Measure your visibility in AI answers
Start with the questions that matter and the answers we can observe, then decide whether GEO deserves a place in your acquisition roadmap.
€350 excl. VAT, up to 90 minutes
Also read: