One in five searches typed into Google would, by OpenAI’s reckoning, have gone into ChatGPT instead.
Adam Fry, who leads the consumer product at ChatGPT, confirmed that search-type queries on the service now equal about 20% of Google’s search query volume. The figure, reported on November 4, is the first specific measure OpenAI has offered of a contest it has been running against the company that has dominated search for two decades.
The same set of numbers described growth of a different kind. OpenAI added 200 million users in the past two months, pushing weekly active users to 1.2 billion. People create 1.5 billion images a week on the service, and voice mode has 250 million weekly active users.
The 20% deserves careful reading. It is a share of query volume, not of revenue, time spent or advertiser spending. Google’s search business earns money by auctioning placements against queries, and the queries that carry the most commercial intent, shopping, travel, local services, are the ones a chat interface handles least naturally today. A chatbot can answer a question about a product; it cannot yet show a page of results with four paid listings above them.
That asymmetry is the reason the number is interesting without being decisive. If ChatGPT handles a fifth of the questions people ask, it has achieved distribution that no rival to Google has managed in twenty years. If it handles a fifth of the questions and a much smaller share of the questions advertisers pay for, the revenue consequence is smaller than the volume suggests.
The growth in weekly users is the stronger claim. Twelve hundred million people using a product every week is a scale that puts ChatGPT in the same category as the largest consumer internet services, and adding 200 million in two months implies that adoption is still being driven by new users rather than by existing ones using it more. Companies usually report the metric that flatters them; weekly actives flatter a service whose monthly figure is larger still.
Voice is the part of the mix that receives less attention than it should. A quarter of a billion weekly users of a voice mode is a large number for an interface that was, until recently, a demonstration. It suggests that a meaningful share of queries are being spoken rather than typed, which changes what the search market looks like: a spoken question is harder to attach advertising to and easier to answer with a single reply.
Images tell you what people do when they have a tool that produces them. Fifteen hundred million a week is not a niche behavior. It is a consumer product in its own right, and it explains why OpenAI has been willing to spend on inference capacity at a scale that puzzles analysts who model the business as a text service.
None of these numbers came with a methodology. OpenAI did not say how it counted a search-type query, whether it included queries inside voice sessions, or how it estimated Google’s volume for comparison. A comparison built on an external estimate of a rival’s traffic is an approximation, and the company that makes it has an interest in the result.
Google’s position is not static either. The company has been putting AI answers at the top of its results pages, which keeps the query inside its own product and changes the economics of the page underneath. It has distribution agreements that place its search box on phones and browsers, and it sells advertising against a query stream that its rival is only now approaching.
The strategic question for OpenAI is what it does with the volume. Search queries are expensive to serve and, in the current product, free to the user beyond a subscription. Advertising would change that, and the company has been building the pieces: shopping results, citations, a format that looks more like a page of links than an answer. Each step toward advertising makes the comparison with Google more direct and the 20% more meaningful.
The strategic question for Google is whether a fifth is the ceiling. Rivals have reached double-digit shares of search before and stalled, usually because distribution was tied to a browser or a phone and the default was hard to dislodge. ChatGPT’s route is different: it does not need to be the default anywhere, because people open it on purpose. That is a slower way to gain share and a harder one to reverse.
Fry’s disclosure also lands in a market that is being repriced around AI spending. Investors have grown impatient with the gap between the capital committed to data centers and the revenue those centers produce, and any evidence that consumer usage is compounding is welcome inside the companies doing the spending.
The number to watch next is not the 20%. It is whether the company begins to publish the figure alongside revenue from search advertising, because that would mean the query volume had been turned into a business rather than a statistic. Until then, one in five is a claim about attention, and attention is the input to a market whose output is measured in dollars.
