On November 2, 2026, Adam Fry, who leads OpenAI's consumer ChatGPT products, said that queries involving web search on ChatGPT have reached about one-fifth of Google Search's volume. Using Google's published figure of more than 5 trillion searches a year, that works out to at least about 2.74 billion queries a day. That is on par with the total number of messages on ChatGPT's consumer plans that OpenAI disclosed for July 2025: more than 2.5 billion a day.
However, Fry did not disclose the source of the Google search figure used in the comparison, nor ChatGPT's actual query count. ChatGPT's usage is clearly growing fast, but this "20%" does not represent share of the search market, and for now it remains OpenAI's own claim. Comparing it with Google's official data and outside research shows that the estimates vary widely depending on how "search" is defined and when it is measured.
OpenAI says ChatGPT is at one-fifth of Google, but the data behind the comparison is undisclosed
At a dinner hosted by OpenAI in New York on November 2, Fry spoke in a conversation with Alex Kantrowitz of Big Technology. There he explained that the volume of search-like queries on ChatGPT has reached about one-fifth of Google's. According to The Verge, Fry did not cite a source for the Google search volume he used as the benchmark.
According to Fry, what OpenAI means by "search" here is cases where a web search is actually performed in response to a question to ChatGPT. It excludes simple definitions or explanations of basic procedures, and covers questions that people would previously have researched with a search engine. Each time a web search is executed in response to a question, it counts as one, but the specific numbers and measurement methodology have not been published.
Asked to compare ChatGPT's scale with Bing, Fry declined to answer, but said ChatGPT's weekly active users have risen from 1 billion to 1.2 billion over the past two months. That figure, too, is OpenAI's own claim, and a rise in users does not by itself mean a rise in search volume.
Calculating from Google's official data gives at least 2.7 billion a day
To test the statement, we first need to sort out what "search" refers to. Google's search count, the number of times ChatGPT actually runs a web search, and the number of information-seeking questions put to ChatGPT are all different metrics.
In an official blog post in March 2025, Google announced that its search service handles more than 5 trillion searches a year. The figure is based on internal data from January 2025, and converts to about 13.7 billion per day.
For ChatGPT's search usage, there are at least two kinds of metrics:
- Web search execution rate: Semrush analyzed US clickstream data and found that in February 2026, ChatGPT ran a web search in 34.5% of cases, down from 46% in late 2024.
- Share of messages for information seeking: In an OpenAI research paper, the share of messages classified as Seeking Information rose from 14% in July 2024 to 24% in July 2025. The paper positions this use as a close substitute for web search.
The former measures whether a web search was actually performed; the latter classifies the purpose of the question. Even when a question asks for information, ChatGPT does not necessarily run a web search, so the two cannot be treated as the same number.
With this difference in mind, calculating Fry's statement against Google's published 5 trillion annual searches gives the following:
\[ \frac{5 trillion}{365 days}\times20\% \approx 2.74 billion/day \]
Because Google's published figure is "more than 5 trillion," applying 20% strictly would require ChatGPT to handle at least about 2.74 billion search-equivalent queries per day.
The point of comparison is the usage OpenAI itself has published for ChatGPT. According to the NBER Working Paper (w34255) released in September 2025, as of July 2025 ChatGPT had more than 700 million weekly users and 18 billion messages a week, or more than 2.5 billion a day. However, the figures cover only the Free, Plus and Pro consumer plans, not Business, Enterprise or Education.
In other words, the roughly 2.74 billion queries calculated from Google's volume is on par with ChatGPT's total message count as of July 2025. Applying the 24% information-seeking share at that time gives about 600 million a day, a gap of roughly 4.6 times from 2.74 billion.
That said, Fry said "about one-fifth," which does not necessarily mean exactly 20%. At 18%, the figure would be about 2.47 billion a day, roughly equal to the July 2025 total message count. The data are also from different times: Google's figure is from January 2025, ChatGPT's usage from July 2025, and Fry's remark from November 2026. ChatGPT's growth and the treatment of business plans could make the numbers consistent, but the latest actual figures needed to judge that are not public.
Outside estimates range widely, from 0.27% to 20%
Several research firms have published their own estimates of how ChatGPT's search volume compares with Google's. The results differ greatly, from about 0.27% for 2024 to the roughly 20% OpenAI has now put forward.
| Source | Period covered | What was counted on the ChatGPT side | Google search volume | Ratio to Google |
|---|---|---|---|---|
| SparkToro / Datos | 2024 (published March 2025) | Search-like usage, up to 37.5 million/day | 14 billion+/day | About 0.27% |
| Peec AI | July 2025 (published October 2025) | About 600 million/day classified as information seeking | About 14 billion/day | About 4.3% |
| Ahrefs | ChatGPT data from July 2025 (published February 2026) | Estimated 65% of all prompts as search-equivalent | About 13.7 billion/day | 12% (18% counting all prompts) |
| OpenAI (Adam Fry) | November 2026 | Queries involving web search | Source undisclosed | About 20% |
Comparing the lowest figure, about 0.27%, with the current 20% gives a gap of roughly 74 times. But because the periods and methods differ, this does not show that ChatGPT's search volume actually grew 74-fold.
Ahrefs' estimate is especially notable. It treated 65% of all ChatGPT prompts in July 2025 as search-equivalent, which it calculated as 12% of Google's volume. Counting all prompts gives 18%, which is ChatGPT's 2.5 billion a day divided by Google's roughly 13.7 billion.
The 20% Fry gave exceeds even that 18% figure covering all prompts. And under OpenAI's definition, only questions that actually involve a web search are counted, so the scope should be narrower than all prompts.
Still, the ChatGPT data used in Ahrefs' estimate is from July 2025, about 16 months before Fry's remark. Usage growth in the interim, and differences in the Google baseline, could well change the numbers. Outside studies are estimates made with each firm's own methods, and OpenAI has not disclosed its measurement approach, so it is impossible to say at this point which figure is closest to reality.
Users are up about 1.7 times: what would it take for 20% to hold?
What could support Fry's statement is ChatGPT's rapid growth. OpenAI's paper put weekly users at more than 700 million in July 2025, while Fry said they reached 1.2 billion as of November 2026. A simple calculation gives an increase of about 1.7 times.
Usage per user may also be rising. According to Semrush's US data, the number of queries per ChatGPT session rose from 1.16–1.21 for most of 2025 to 1.75 in February 2026. If the growth in users and rising frequency of use applied across the board, ChatGPT's total messages would likely have grown substantially since July 2025.
But Google's search usage is growing too. In its third-quarter results on October 28, 2026, Alphabet said revenue from Google Search and other rose 12%, from $56.567 billion a year earlier to $63.350 billion. It also said year-over-year growth in search queries was the highest in the past five years, but did not disclose actual query counts.
On the same day, Google also announced it had expanded ads in AI Mode to 10 countries, including Australia, Canada and India. It plans further rollouts in the fourth quarter of 2026 to Brazil, Mexico, the UK and others, but Japan is not among the announced countries. Google is expanding its search services, but has not published recent search volumes directly comparable with OpenAI's.
So how much total usage would ChatGPT need for OpenAI's 20% claim to hold? If the 34.5% web search execution rate Semrush observed in the US in February 2026 were applied to messages worldwide, the required total would be:
\[ \frac{2.74 billion}{0.345} \approx 7.94 billion/day \]
Under this assumption, about 7.9 billion or more messages a day would be needed, roughly 3.2 times the more than 2.5 billion published for July 2025. However, Semrush's execution rate is based on US usage data, and applying a session-level observation to worldwide message counts is only a rough approximation. The figure of about 7.9 billion is therefore not a universal minimum condition for 20% to hold.
Simply multiplying the roughly 1.7-fold rise in weekly users by the roughly 1.5-fold rise in queries per session gives about 2.5 times. That approaches the required 3.2 times, but it combines data from different regions and periods and does not show actual total messages.
Semrush's US data also show that traffic to ChatGPT has plateaued at about 1 billion visits a month since around November 2025, and that the number of sessions involving web search declined every month from November 2025 to February 2026. The results show that growth in users and growth in search use do not necessarily proceed at the same pace.
How far can OpenAI's "20% of Google" be trusted?
OpenAI presumably knows its actual search counts internally, so its data may be more accurate than outside estimates. Given that both users and usage frequency are rising, it cannot be ruled out that search-like queries have reached a scale equal to 20% of Google's. At present, however, the underlying figures and measurement methods have not been published, so independent verification is not possible.
One thing to watch is whether OpenAI publishes its latest total daily messages and number of web searches executed. If total messages had reached about 7.9 billion, there would be room for consistency with the estimate using the US search execution rate of 34.5%. But that alone would not prove 20%; the web search execution rate and the scope of covered plans would also need to be confirmed.
Another is whether third parties such as Semrush and Datos use 2026 data to estimate the number of queries that actually involve web search. A result of 15% or more of Google's would support the statement. This 15% is a benchmark set by our editorial staff, and should be distinguished from estimates that include all prompts.
What matters now is not to interpret this as ChatGPT having captured 20% of Google's search market. What Fry presented is a comparison of query counts under a specific definition, not search market share or a rate of user migration. Until OpenAI and Google publish actual figures in comparable units, it is reasonable to treat this 20% as a number that hints at ChatGPT's growth but lacks independent verification.