Inside Google, the number became a talking point this week: one billion users. The company said its AI agent feature in Google Search has passed the billion-user mark, a figure that turned a product experiment into one of the largest deployments of generative AI in the world, and it has become the yardstick the company uses to measure the distance between itself and ChatGPT.
The feature works by letting users conduct multi-step tasks in natural language. Instead of issuing a query and clicking through results, a user can ask for a product comparison across specifications and prices, a trip plan that weighs flights, weather and hotel reviews, or an analysis of a spreadsheet of data, and the system assembles the answer itself, searching, reasoning and presenting in a single pass. The capability has been rolling out through 2026, and the company said usage has accelerated sharply in the first half of the year as Google embedded AI agents deeper into the core search product.
The acceleration is a competitive necessity. OpenAI’s ChatGPT has become the default destination for a generation of users who want answers rather than links, and Google has treated the agent feature as its principal response: keep the query on Google, and give it an answer good enough that the user never leaves. The billion-user figure is the evidence that the strategy is reaching people, at least in scale terms.
The harder question is whether it reaches people profitably. Every AI-generated answer costs Google more to produce than a traditional set of links, because each one requires live model inference, and the gap between the cost of a conventional search and the cost of an agentic one has been the central concern of the company’s financial analysts. Search has been Google’s most profitable business because its marginal cost per query was close to zero; the agent feature rewrites that equation.
The figure itself deserves a closer read. A billion users of an AI agent feature is a scale no rival has matched, but the definition spans casual experimentation and regular use, and analysts said the number’s meaning depends on how much of that base comes back. Google has not broken out daily or weekly engagement for the feature, and the company’s internal conversations, according to people familiar with them, are focused on converting the base from novelty to habit. The comparison with ChatGPT is instructive: OpenAI’s product has a smaller user base but deeper engagement per user, and the contest between the two is partly a contest between reach and depth.
The company has been careful about how the economics play out. AI answers are presented alongside traditional results and advertising, and Google has been testing ad placements inside the agent experience, including sponsored products in comparison queries and promoted suggestions in planning tasks. The model, in outline, is the same as the one that made search profitable: attach commercial placements to whatever users are looking for, even when the answer is generated rather than clicked.
Whether the same margins are attainable is unsettled. Advertising in an answer-oriented interface depends on users trusting the placement, and the sponsored slots have to feel like part of the answer rather than an interruption. Analysts who follow the company said the revenue per AI session is still below the revenue per conventional search session, and that closing the gap is the operational task of the next several quarters.
The monetization experiments are also visible. In product-comparison queries, Google has begun showing sponsored options inside the agent’s answer; in planning tasks, it has tested promoted recommendations; and in data-analysis requests, it has offered premium features that sit behind the assistant’s free tier. Each test is a small probe of the same question that has defined search’s economics since the beginning: how much commercial interruption will users tolerate inside a product that feels like an answer rather than a list of links? The answers so far, the company says, are encouraging, but the revenue per session remains below the conventional-search baseline.
The scale itself changes the competitive picture. A billion users is a base that no rival agent product has matched, and it gives Google an advantage in the data that makes agents better: every comparison, every trip plan, every data analysis is a training signal for the next version. The company’s position is that the flywheel of usage, improvement and more usage will compound faster than its competitors’ feature advantages.
The comparison with ChatGPT is no longer a battle of demos. OpenAI’s product leads on the quality of its reasoning model and on the depth of its agentic capabilities, while Google’s leads on distribution, on the habit of search and now on the sheer number of people who use an AI agent without ever calling it that. The competition has settled into a contest between the best answer and the most convenient answer, and the billion-user figure is Google’s argument that convenience wins.
The next test is commercial, not technical. Google has shown it can get a billion people to use AI agents in search; it now has to show it can make as much money from them as it did from links. The company’s share price will follow the evidence, and the evidence will arrive in the form of ad revenue per query over the coming quarters. The users are there. The ledger is next.
