ChatGPT Search Hits 20% of Google's Volume, OpenAI Says

For years, the default answer to almost any question has been to search for it on Google. This week OpenAI offered a data point suggesting that habit may be loosening. According to Adam Fry, OpenAI's consumer product lead for ChatGPT, the volume of search-like queries going through ChatGPT is now roughly one-fifth of the volume Google handles.

It is a big claim, and it comes with caveats. Let's look at what was actually said, how "search" was defined, what else OpenAI revealed the same night, and what it might mean for those of us who type questions into a box all day.

What OpenAI actually announced

Fry gave the figure on Monday night at an OpenAI-hosted dinner in New York City. In a conversation with Big Technology's Alex Kantrowitz, he said ChatGPT's volume of search-like queries is "roughly one-fifth" of Google's. He pointed out that ChatGPT started doing this only about a year and a half ago, and reminded everyone that early versions of the chatbot had no up-to-date information and would simply say a question fell past its training date.

That context is worth pausing on. A product that could not tell you today's news now claims a fifth of Google's query volume, at least by one measure. Whatever you think of the number, the speed of the shift is the real story.

The headline figure came with a broader growth announcement. According to Big Technology, OpenAI says ChatGPT went from 1 billion to 1.2 billion weekly active users in 60 days. Fry put it this way: "The growth is just going big." Those are company-reported numbers, delivered by an executive at a company event, so they deserve a healthy dose of skepticism until someone outside OpenAI can check them.

How "search" was defined

The most important part of any statistic is the definition behind it. Asked whether each conversation with an embedded search counts once or whether every question requiring a web search counts, Fry answered that every time you ask a question and ChatGPT goes to search the web, that is a search. A long chat with 20 back-and-forth prompts is therefore not one search; only the turns that trigger a web lookup are counted.

A separate report on KuCoin's news feed, which attributes it to Huo Xing Cai Jing and Phoenix Tech, adds that Fry excluded tasks like defining terms or giving simple instructions. The same report notes that he did not disclose the source of the Google volume figure used for the comparison. That is a significant gap: a ratio is only as solid as both of its halves, and we only know how one side was counted.

So the fair reading is: OpenAI says that, by its own definition of a search-like query, ChatGPT's volume is about 20% of Google's. That is not the same as saying one in five searchers has abandoned Google. Many people use both, and a single person's habits can span several tools in one afternoon.

Why people are using a chatbot to search

Fry offered an explanation for the shift, and it rings true to me. About two years ago, he said, people came to ChatGPT mainly for writing help and coding help, things you wouldn't normally search for. Over the past year, usage broadened to topics that need fresh information: looking up a person, a place or a thing, plus finance and health questions where people want up-to-date, factual answers and where ChatGPT does a web search. In his words, it is basically anything a user would once have taken to a traditional search engine.

The second argument is personalization and memory. Fry, who used to work at Google, contrasted the old Q&A box with a conversation that keeps context. His example: you ask how tall the Eiffel Tower is, but the assistant knows you are planning a trip to Paris, so it can answer with that in mind. A search engine treats each query as a fresh start. A chatbot can treat your questions as a journey.

I find that genuinely appealing, and also a little double-edged. The same memory that makes answers feel tailored means the assistant is building a detailed picture of your life. Convenience and privacy are rarely free of tension, and it's worth deciding for yourself where your comfort line sits.

The rest of the news: voice, images, and merging modes

The search number was only one of several stats. According to Big Technology, OpenAI says voice input, which covers dictation and spoken conversations, has 250 million combined weekly active users, and that people are generating 1.5 billion images per week. Fry credited the voice growth to newer voice models, which he said are much more natural to speak with, including the ability to be interrupted and to give small verbal acknowledgments mid-conversation.

He also confirmed that ChatGPT's "Chat" and "Work" modes will be merged, so users no longer have to choose between a quick reply and a long agentic task. His answer on timing was just "coming soon." Anyone who has stared at that toggle wondering which to choose will probably cheer.

On interactive answers, Fry said he is excited about what OpenAI calls Intelligent UI, where responses include visuals and interactive diagrams instead of plain text. If you want to understand where that idea comes from, we covered it in our piece on GPT-6 and Intelligent UI. His own example was asking how the heart works and getting an interactive diagram of blood flow. He is especially bullish about it for learning.

Ads, money and an IPO in the background

Search is valuable largely because of advertising, so it is no surprise that ads came up. Fry said OpenAI's ad business passed a billion dollars in annual recurring revenue within 200 days, and that it is growing extremely quickly. He also said OpenAI had passed 50 million subscriptions several months ago and that the number has grown considerably since. These are, again, company statements.

According to the KuCoin-hosted report, Fry said ads have not hurt the product or its usage and that most users do not seem to mind them. It adds that last month OpenAI announced it would start showing images of sponsored products and services in ChatGPT, rather than just company names, logos and product links. The same report observes that the company may be preparing an IPO early next year. For more on the financial side, see our analysis of what OpenAI's valuation means.

Here is where my critical hat goes on. When a chatbot answers a question and also earns money from sponsored products, the line between help and persuasion needs to be very clear. Classic search has lived with that tension for decades, with labeled ads sitting above organic links. A single synthesized answer is harder to label. Users deserve to know, plainly, what is paid and what is not.

What this means for Google and for publishers

Does a 20% ratio threaten Google? Not necessarily in the short run. Google's own volume is enormous, and the comparison is built on OpenAI's definition. The KuCoin report also says OpenAI's search likely relies heavily on Microsoft's Bing alongside its own index. That is the report's characterization, not something Fry is quoted as confirming in the Big Technology conversation. In other words, the ecosystem is more tangled than a simple two-horse race.

Still, the direction matters. If more questions are answered inside a chat, fewer people click through to websites. We have already explored what that does to the open web in how Google's AI search is reshaping online publishing, and the same pressure applies when the answer engine is a chatbot. Publishers who relied on search traffic are watching two giants change the rules at once.

If you are curious about how the alternatives stack up in practice, our comparison of Perplexity and Google AI Mode is a useful starting point, and the ChatGPT vs Gemini guide covers the two assistants most people are weighing.

How to think about it as a user

A few practical habits come to mind. First, treat chatbot answers to health and finance questions as a starting point, even when they cite sources. Fry himself noted these are the areas where up-to-date, factual information matters most, which is exactly why you should double-check them. Second, notice when you use each tool: a chatbot is great for exploring and summarizing, while a classic search is often better when you want to see many sources side by side. Third, pay attention to what the assistant remembers about you.

Finally, ask who benefits. Asked why public polling on AI is so negative, Fry argued that the more people use AI, the less scary it will feel. That may be true for some, but trust is earned through transparency about data, ads and errors, not only through usage numbers.

A habit in the middle of changing

The honest takeaway is this: nobody outside OpenAI can yet verify that 20% figure, but the trend line is believable. People increasingly begin their questions in a conversation rather than a list of links. Search is not dying, but it is becoming something more personal, more conversational and, probably, more commercial.

So here is a question for you: the last time you wanted to know something, where did you go first, and why? Your answer, multiplied by a billion people, is quietly shaping how the internet will work.