Someone lands on your site with a precise question in mind: how much it costs, whether you do it too, how long it takes. They are in a hurry. If in the first few seconds they do not realize they are in the right place, they close the tab and go back to the results. It is not rudeness, it is how all of us browse, you included.
The Nielsen Norman Group, in a 2011 synthesis by Jakob Nielsen still cited as a reference principle, describes a window of about 10 seconds in which a visitor decides whether to stay or leave, with an average visit that often lasts less than a minute. The exact number matters little: the point is that the decision to stay is almost instant, and it hinges on how quickly the person finds what they are looking for.
What changed: the answer no longer comes from the site
For years the path was one: I search on Google, I click a result, I read on the site. Today a piece of that path closes before reaching you. The answer engine (Google with its AI summaries, but also ChatGPT, Perplexity, Gemini) gives the answer on the results page, and often that is enough for the person.
The 2026 numbers tell the shift. SparkToro's zero-click study (clickstream data, January to April 2026, United States) measured that 68% of Google searches end without a single click to any site, up from 60% two years earlier. And the first randomized experiment on the topic, run by two researchers from the Indian School of Business and Carnegie Mellon University on over a thousand real users between January and February 2026, found that where the AI summary appears, clicks to sites drop by 38%, with zero-click searches rising from 54% to 72%.
Read together, they say one thing: a share of the people who used to look you up now get an answer about you without ever opening your site. The AI gives the answer, reading something. The question becomes: what does it read, and whose is it.
Before AI comes clarity
Here some honesty helps, because it is easy to jump to the technology part. Often people do not leave out of laziness: they would like to find things on their own and cannot, because the information is poorly organized. The answer is there, but it is buried at the bottom of a page, written in jargon, or split across three different tabs.
The uncomfortable counterpoint confirms it. Gartner (survey of 5,728 customers, December 2023) found that only 14% of issues are fully resolved in self-service, and that in 43% of cases where self-service fails the customer “could not find relevant content”. The answer was not missing: the way to reach it was. That is why the first job is not AI, it is the logic with which content is organized: titles that say what they contain, answers placed next to the questions, a fast page that does not make people leave before the content. It is the same reason a genuinely fast website starts from the architecture and not from a last-minute patch.
That said: clarity is the condition, not the complete solution. A clear, fast site solves the problem of those who arrived and read. It does not solve the problem of those who do not even arrive, because their answer is taken elsewhere. Two different things, addressed in different ways.
Path one: getting found inside the answers (GEO)
GEO (Generative Engine Optimization) is the work of making your content citable inside AI-generated answers. It is not classic SEO, even though it uses it as a base: SEO aims to rank a page in the results, GEO aims to make sure that, when an answer engine composes a reply, it includes you and cites you as a source.
Concretely it means writing so that a model can extract the answer and attribute it to you: a direct answer up top (before the context), questions posed the way a real person poses them, data with source and year, structured data that tells the machine what it is reading. It is not a magic formula and no one really controls what a model cites: it is an investment in the probability of appearing, not a guarantee.
The key point: GEO works outside your site. It makes you present where the answer is given in your place. It is the right path when the problem is that people do not arrive, because they stop at the summary. But it does not help someone already on your page who cannot find what they are looking for.
Path two: answering on the spot (an assistant)
The other path looks at the problem from the opposite side. The person has already arrived, is reading, but their specific question finds no immediate answer. An assistant on the site is for this: it answers there, the moment the question arises, on the content the site already holds.
Worth saying right away, again Gartner, same survey: self-service is liked as long as it works, but it often fails, and it fails precisely when the customer cannot find the right content. An assistant is not a magic wand and does not convert on its own: it makes sense if there is real, well-made content underneath, because an assistant can only say what the site already knows. On an empty or confusing site, an assistant is just a more expensive way of not answering.
The key point, mirroring the first: the assistant works inside your site. It helps those who already arrived. It does not make you more citable by answer engines. It solves the other half of the problem.
GEO or assistant: how to choose
They are not equivalent alternatives, and anyone selling them to you as the same thing is conflating two problems. GEO makes you citable outside; the assistant answers inside. The choice depends on where you lose people:
- If they look you up but do not arrive, because they are satisfied with the AI summary, the work is GEO: making you appear where the answer is given.
- If they arrive but leave without finding, the work is clarity, and then possibly an assistant that answers on the spot.
- If you are not sure where you lose them, start from the data: look at where traffic comes from, how long it stays, which pages it abandons. Diagnosis comes before the tool.
And before both, always, clarity: neither GEO nor an assistant saves poorly organized content. Logic comes before code, here as in custom software, where you start from the problem and not from the tool.
A concrete example: this site's assistant
To understand the second path, we built it for ourselves. This site has an assistant, based on Claude by Anthropic, that answers on the site's real content: a knowledge base drawn from the pages you are reading, in the voice we write everything else in. It does not make things up: it says what the site already knows.
It is aware of the page you are reading. On an article like this you can ask it to summarize it, and it answers on this text, not in the abstract. And if a concrete need emerges from the conversation, it turns the question into a contact request that reaches the team by email: not yet another form to fill in, but the thread of the conversation continuing on the other side.
It has usage and budget limits, by choice. It is not unlimited and does not want to be: it is a sober tool, meant to answer well to someone with a real question, not a gadget for the window. And it does not replace a well-made site: if the content underneath were not clear, the assistant would have nothing good to say. It works because it rests on pages written with care, the same order with which we line up planning, execution and reporting in Verso Flow.
What to do, in practice
There is no obligation here, and be wary of anyone presenting it as one. Google remains dominant in volume, and many sites convert perfectly well by simply staying clear and fast. But the opportunity is real, and it is worth addressing with method instead of by fashion:
- Clarity first: the answer next to the question, honest titles, a fast page. Without this, the rest does not hold.
- Then decide where you lose people: before the click (GEO work) or after (clarity and assistant). They are two distinct problems.
- An assistant makes sense if you have real content to draw on, not to fill a void.
- Measure, do not imagine: traffic and behavior data tell you where to act better than any trend.
If you want to understand where people get lost on your site, and whether it makes more sense for you to work on GEO, on content or on an assistant, tell us how they find you today: we start from the real behavior of those who arrive, not from the tool to sell.