From search engine to AI intermediary: how the customer path is changing

The current inflection point on the internet is easiest to describe through one simple substitution. Not long ago, users asked the network for a path to information. Now, increasingly, they ask it for an already assembled judgment. In the old logic, a person entered a query, received a list of links, opened several pages, manually compared wording, prices, and signs of reliability, and then built the picture for themselves. In the new logic, a substantial part of that work is delegated to the AI mediator. The user begins not with navigation across documents, but with a conversation: “explain,” “compare,” “what should I choose,” “who is stronger in this category,” “what is the difference,” “which option fits my case.” Only afterward, if the answer proves insufficient, do they move on to sources.

This shift does not yet mean the death of search, which is precisely why it is often underestimated. Search as infrastructure remains enormous. But the interaction surface is changing before our eyes. McKinsey writes that about 50% of Google search queries are already accompanied by AI summaries, and that this figure may exceed 75% by 2028 [1]. In 2025, Google reported that AI Overviews had more than 2 billion monthly users, while AI Mode had already surpassed 100 million monthly active users in the United States and India [2]. In February 2026, OpenAI reported more than 900 million weekly active ChatGPT users [3]. In other words, we are no longer dealing with a niche technological experiment, but with a mass shift in the interface through which people access information.

It is especially telling that users themselves already perceive this environment as a source of solutions rather than as a curious add-on. According to McKinsey, about half of consumers deliberately use AI-assisted search, and 44% of those users name it as their main and preferred source of information for decision-making — above traditional search, brand websites, and review platforms [1]. This figure matters not only in itself. It means that the point at which the first impression is formed is increasingly located not on the company website, but in an answer the company does not directly control. What this shift costs a business in monetary terms — and how to estimate the scale of losses from being excluded from machine answers — is explored in the article on the economics of invisibility; here we focus on the mechanics of the new journey itself.

The classic customer journey in the digital environment could be described as the chain “query -> list of links -> comparison of documents -> shortlist -> action.” In the new environment, an intermediate link increasingly appears between the query and the list of links — a synthesizing interlocutor. The chain now looks different: “question -> AI summary -> follow-up -> preliminary filtering -> possible site visit.” The difference seems subtle, but in practice it changes the entire attention economy. Previously, a brand could lose at the click-through stage, yet still remain present on the results page. Now it can lose earlier — at the level of whether it was included in the machine judgment at all.

The speed at which this layer is growing is also confirmed by outbound referral data. Similarweb estimates that in June 2025, AI platforms drove more than 1.13 billion visits to external websites; that was 357% more than a year earlier [4][5]. By its estimate, the average monthly number of visits to generative AI platforms grew 76% year over year, while app downloads rose 319% [4]. Figure 1 shows this one-year surge in referrals to websites. It is important, however, not to fall into statistical hypnosis: according to the same data, Google Search generated about 191 billion referrals in June 2025, still orders of magnitude more [5]. But this is precisely where the central meaning of the shift begins. AI does not have to overtake search in gross traffic volume already today in order to radically change user behavior. It only has to capture the stage of initial judgment.

The quality of this new traffic also says a great deal. According to Adobe Analytics, in retail, visitors arriving from generative AI sources viewed 12% more pages per visit and showed a 23% lower bounce rate than users from non-AI sources [6]. In travel, Adobe recorded a 45% lower bounce rate for such traffic [6]. In its documentation for site owners, Google notes a similar effect: clicks from results pages with AI Overviews appear to be higher quality, and users more often spend more time on the site [7]. This is an important detail in the new logic of the customer journey. There may be fewer clicks, but each one increasingly arrives later in the funnel and with higher intent.

Hence the first serious change for marketing. The traditional top of the funnel no longer lives only in the search results. It partially dissolves into the conversation. The user who previously read five documents and only then formed a brief opinion now receives that opinion at the outset. The AI answer may already name the main market players, note the strengths and weaknesses of solutions, filter out obviously unsuitable options, and highlight pricing and technical constraints. The brand website becomes not the first point of contact, but the place where a choice already underway is refined.

The second change is subtler still: for ordinary users, the distinction between “search” and “talking to AI” is rapidly losing meaning. What matters to them is not the technological mode, but the user experience. They see a short answer at the top, can ask a follow-up question, get links, and continue the conversation. In such an environment, the search engine itself becomes an answer system, while chat becomes a form of search. That is why the debate of “search versus AI” now describes reality poorly. What we are seeing, rather, is the dissolution of the old boundary between search, reference, comparison, and consultation.

The third change concerns the tempo of decision-making. The AI mediator can shorten the path not only through speed, but through cognitive offloading. It takes on the preliminary comparison, the translation of complex language into simpler terms, the condensation of large amounts of material into a few paragraphs, and the delivery of first judgments. For the person, that is convenient. For the brand, it means that a substantial share of the struggle for attention now takes place earlier and in a more compressed form. If a company fails to make it into that preliminary filtering stage, its chances of recovering later are lower.

To describe the next stage of this transformation, we suggest looking not at the “market share of individual platforms,” but at a broader measure: the share of informational sessions in which the first answer ultimately delivered to the user is meaningfully mediated by an AI layer. This includes AI Overviews, AI Mode, standalone chat interfaces, and other answer surfaces where the user receives a first interpretation before moving to documents. Figure 2 presents an editorial forecast of this share over the next 1, 3, and 5 years. In the base case, we estimate it at roughly 18% for 2026, 25% for 2027, 39% for 2029, and 54% for 2031. This is not an official market metric, but an analytical estimate built from a combination of public signals: the scale of AI Overviews, the size of the ChatGPT audience, the growth of AI referrals, and the behavioral-shift data published by McKinsey, Google, OpenAI, Similarweb, and Adobe [1][2][3][4][6].

This forecast is intentionally moderate. It does not assume that classical search will disappear. On the contrary, search will remain the deep infrastructure of retrieval and verification. But the user-facing surface will increasingly look not like a list of links, but like an answer with the option to go deeper. One year out, this most likely means an acceleration of a process already underway: AI will even more often become the first point of contact with a question, but a substantial share of site traffic will still come from classical search, direct visits, and other channels. Three years out, the distinction between search and AI will become much less meaningful for the mass user: they will remember not the mode, but the convenience. Five years out, the most probable world is one in which, in a substantial share of informational sessions, the market’s first filter is precisely the answer layer.

All of this also changes the role of the brand website. Previously, the site aspired to be the stage of first contact. Now that role increasingly passes to the AI mediator. The site becomes a place for checking, deepening, confirmation, comparison of details, commercial contact, and action. That does not make it less important. But it does make it part of a longer and more complex chain. In that chain, the brand has to be understandable not only to the search index and not only to the person who has already arrived on the site, but also to the answer system that decides whether it is worth sending the user there at all.

That is why the current paradigm shift looks not like an instantaneous revolution, but like the gradual disappearance of the gaps between channels. Search, reference, comparison, recommendation, and consultation are being pulled into a single point. That point is the AI mediator. And the struggle for visibility, trust, and demand begins even before the user reaches the brand page — in the structure of the first answer.