3%.
This is the share of traffic already generated by generative AIs like ChatGPT (86%) and DeepSeek (3%) compared to traditional search engines.
 
This study by OneLittleWeb (April 2025) indicates that while we are still at the beginning of the adoption curve, the use of AIs is likely to dismantle the GAFA more effectively than regulations (and all related lawsuits) themselves. This new usage reopens the game of major technological monopolies.
And for a reason, traditional search, with its endless scrolling, blue links, and associated clicks, is gradually giving way to the use of answer engines. It may seem trivial, but it ultimately questions the entire model built by the GAFA. Where one would search a search engine for buying a car, for example, navigating through dozens of results and as many site clicks was necessary, the generative assistant now provides a direct answer and tailored recommendation.
The shift? These new searches escape advertising exposure, the foundation of the economic model. All signals are green for these usages to continue developing massively, as seen in foreign investments in France last year in the EY Attractiveness Barometer, where AI is one of the sectors with the most projects or the billions announced at Choose France – mainly focusing on the capacity (data centers) to operate in France and Europe.
 
Intermediation, vice repetita?
If GAFA intermediated the media, we have never been closer to them being intermediated in turn. In a new world where boomers may continue to use only traditional search, new generations only generative, and intermediate generations a bit of both... The giant's model is disrupted. Why? They based their business on a global and unique product or service deployment where "fine-tuning" doesn’t exist or barely, except perhaps for the price variable adapted to the purchasing power of the country where they operate. Creating, operating, and maintaining new systems, demanding in infrastructure and investments, while keeping the old ones to respond to this "generationalization" of the web is rather unnatural.
 
But it is certain that the giants will fight to keep their core of usage and data from leaving. The announcement of the widespread deployment of AI Overview at the I/O developer conference is the perfect illustration: it does not herald the death of its golden goose but a first attempt to prevent the flight of usages without deeply changing its traditional search interface and continuing to demonstrate to advertisers that its advertising inventory is still the most attractive.
Isn't satisfying multiple usages and generations on one platform a too wide gap on a fragmented market? It might be the necessary step (let’s remember that we are still at the bottom of the adoption curve) before usages completely change and eventually propel Gemini, Google's generative assistant.
 
An Economic Model to Find
Beyond the question of environmental cost and the number of liters of water per query, another big question remains: what economic model behind generative AI engines? Subscriptions were the first lever, but it is clear that the model is not profitable for consumer engines. Will advertising reappear in answer results? It is a strong possibility. When I look for my new sneakers, advertisers might pay to appear in the top three responses or even to be the suggested brand?
One thing is certain, as long as the economic model is not found and the new generative AI interface is not entirely finalized, the space is wide open for new entrants... and not just for Tech giants, but also for strong brands that could launch their agentic AI. A way to arrive as a specialist, offering added value and a level of response exceeding that of generalist AIs. Obviously, a way to regain what GAFA's proprietary and non-interoperable models could no longer allow: re-intermediate their online relationship with the client, to be the unique point of contact and collect useful data again previously left to search engines and other major social networks.