OpenAI launches "search" directly in ChatGPT. OpenAI launches the "shopping" mode directly in ChatGPT. OpenAI launches the use of external connectors, such as Booking, Airbnb or Canva, directly in ChatGPT. Each time, the aim is the same: to fan the flames of user desire and make them want to go and see if the grass is greener elsewhere... than on Google! A love story... sorry, advertisers and money. The heart of the Google success story business model.

But the results published by Alphabet for the third quarter of 2025 confirm the remarkable stability of its business model. Over the period, the "Google Search & other" line generated $56.6 billion in revenues, up 15% year-on-year, driven mainly by the retail and financial services sectors. At the group level, this activity remains the main financial pillar, in line with previous years when it accounted for around 57% of total revenue.

Management is leaving no ambiguity on this point. On the analyst call, Alphabet CEO Sundar Pichai said that "AI is creating a moment of expansion for search" and that, as users discover the new experiences on offer, "they’re coming back to use search more". He insisted that "search and its AI experiences are designed to showcase the web, sending billions of clicks to sites every day". The stated aim is therefore not to turn users away from the engine, but to intensify their use of it.

7% more for paid clicks

This strategy is taking shape through the massive deployment of AI Overviews and AI Mode. According to Alphabet, AI Mode now has more than 75 million daily active users, with a volume of requests that has doubled over the quarter, while AI Overviews is deployed to more than two billion users. These products are not presented as marginal experiments, but as central building blocks in the evolution of the engine. Philipp Schindler, Google’s chief commercial officer, stressed that "our new AI products, like AI Overviews and AI Mode, continue to drive growth in total query volume, including commercial queries, creating more monetisation opportunities".

The key issue for investors is the monetisation of these new formats. Here again, the message is explicit. Philipp Schindler pointed out that, "even with our current level of advertising integrated into AI Overviews, we are seeing an equivalent overall monetisation rate". The figures given on the call support this: paid clicks are up 7% year-on-year and cost per click also is up 7%, suggesting that the introduction of AI has not, at this stage, degraded the economics of search.

The defence of the advertising model is particularly visible in the field of shopping. Google has enhanced AI Mode with features that allow "conversational shopping in search", while maintaining a close link with merchants and sponsored formats. The company has also introduced so-called "agentic" payment capabilities, designed to reduce the friction between search and transaction. In the same spirit, Google is developing agentic experiences in key sectors such as retail and travel, which Philipp Schindler described as "complementary to the way people search for information". He said the group was paying "great attention to integrating partner ecosystems to create value for them", underlining the lack of desire for disintermediation.

Monetising the 20% of non-commercial requests

This approach is part of an assumed financial rationale. During discussions with analysts, it was pointed out that around 20% of search queries have historically been commercial. Google’s stated ambition is to broaden this scope by monetising queries that have hitherto been of little or no commercial value, thanks to a better understanding of intent made possible by AI. Philipp Schindler summed up this ambition by explaining that the most advanced models now make it possible to "understand and predict intent like never before", opening up "new commercial paths".

In the face of this defensive yet expansionist strategy, OpenAI’s trajectory appears structurally different. OpenAI does not have an advertising engine comparable to Google’s, and its revenues are based mainly on subscriptions to ChatGPT, offerings for businesses and use of its APIs. This model, although growing strongly, does not benefit from the mass effect or economic depth of search advertising.

The usage and monetisation figures clearly illustrate this gap. At the end of 2025, ChatGPT claimed to have around 800 million weekly active users, but only around 35 million paying subscribers, or around 4 to 5% of its user base. OpenAI, which is not listed, does not publish detailed financial data, but these orders of magnitude are regularly cited by the financial press and industry analyses.

In contrast, Gemini now exceeds 650 million monthly active users, according to official statements by Alphabet when it presented its Q3 2025 results. However, Google is not releasing any public figures for Gemini-specific paying subscribers. The assistant is integrated into a wider set of products and packages - Google One, Workspace, Android and, above all, Search - and its monetisation remains largely indirect.

Becoming the "reflex"

This contrast highlights two opposing economic logics. ChatGPT directly converts a limited fraction of its audience into subscription revenues, which mechanically limits its short-term monetisation capacity. Google, on the other hand, does not need Gemini to be massively profitable. The increase in usage feeds, above all, search, shopping and travel, in other words the segments where advertising captures attention at the moment of decision and continues to represent the core of the group’s revenue.

Behind the technological rivalry between Gemini and ChatGPT lies a much more traditional battle. Google is seeking to preserve and expand its advertising revenues based on search and transactional uses, while ensuring that AI strengthens the engine without undermining its economy. OpenAI, for its part, is trying to shift some of the attention to a single conversational interface, capable of capturing uses comparable to those of search. This shift, if it were to take place on a large scale, would not be based on model benchmarks or on the quality of responses alone, but on the ability to become the reflex at the precise moment when the user is searching, comparing and deciding.

At this stage, the relationship between users and their assistants resembles an old Christmas movie script. Gemini is the beloved husband. The one we don’t really notice any more, but with whom we share our daily lives, our habits and our automatisms. He’s already in the house, in the kitchen, in the car, on the phone. It doesn’t ask us to love it any more: it’s simply there when we need to look, compare, book or buy. And every day-to-day gesture keeps the household economy going.

ChatGPT is still the lover. Brilliant, attentive, often more inspiring. The one with whom you talk more freely, to whom you confide your doubts, your plans, your ideas. You consult him, you admire him, sometimes you come back to him several times a day. But at the decisive moment - when we book a ticket, choose a product or click to buy - we often return to the official spouse.

Avoiding divorce

This is the whole point for OpenAI. As long as ChatGPT remains the preferred but occasional lover, its monetisation will remain limited to a few subscriptions and a minority of users prepared to pay for the relationship. To become economically comparable to Google, it will have to change its role: no longer just seduce, but establish itself. Locking in uses, capturing reflexes, being there not just when people think, but when they decide.

Google, for its part, doesn’t need to rekindle the flame. All it needs to do is avoid divorce. Gemini is not there to be loved for its own sake: it is there to preserve an old but extraordinarily profitable marriage, based on daily attention and transactional advertising. In this story, technology is less important than the place it occupies in users’ lives. And like all Christmas movies, the ending rarely depends on who's the most attractive, but almost always on who's already home.

(To our readers: one last fun fact. Questioned with the same prompt and the same article, the two tools, ChatGPT and Gemini delivered two completely different images. Here's the one from Gemini. :-) )