OpenAI says ChatGPT now has 1.2 billion weekly active users, and that its search-like query volume is about one-fifth of Google's. The numbers come from ChatGPT consumer product lead Adam Fry, who spoke to Alex Kantrowitz at an OpenAI-hosted dinner in New York on Monday night. Big Technology published the interview on November 3, 2026. This post separates what Fry said from what anyone outside OpenAI has been able to check.
TL;DR: the claims, one by one
| Claim | Source | Independently verified? |
|---|---|---|
| ChatGPT went from 1B to 1.2B weekly active users in about 60 days | Adam Fry, via Big Technology | No. No methodology published |
| Search-like query volume is roughly one-fifth of Google's | Adam Fry, citing OpenAI internal analysis | No. Google has not confirmed or disputed it |
| 250M weekly active users of voice (dictation plus voice mode) | Adam Fry | No |
| 1.5B images created per week | Adam Fry | No |
| More than 50M subscriptions, and ads past $1B ARR within 200 days | Adam Fry | No. Earlier subscription figure was announced by OpenAI |
| Chat and Work modes will merge "soon" | Adam Fry | A plan, not a release. See our earlier report on the tabs merge |
What did Adam Fry actually say about growth?
For context, TechCrunch reported on February 27, 2026 that OpenAI had announced 900 million weekly active users and 50 million paying subscribers, up from 800 million in October 2025.
The interview opens with the headline numbers. "The company added 200 million users over just the past two months," Kantrowitz writes, "jumping from 1 billion to 1.2 billion weekly active users in 60 days." Fry put it as roughly 100 million users in 30 days and another 100 million in the next 30. His own summary: "The growth is just going big."
Our own coverage of OpenAI's launch posts has already repeated the 1.2 billion line as OpenAI's wording: the Intelligent UI explainer quotes the company saying "more than 1.2 billion people who use ChatGPT each week." In the Berlin DevDay Exchange post we noted we could not trace that number to a primary OpenAI statement.
That gap matters. The new interview gives a named OpenAI executive saying it on the record, which is a stronger attribution than a relayed quote. It is still a company statement. No audited figure, third-party panel or filing backs it.
Is ChatGPT search really 20% of Google?
A green sprout growing from a stack of pages, showing how ChatGPT weekly users and search-like queries keep growing
Fry's wording is careful: "ChatGPT's volume of search-like queries is now roughly one-fifth the volume of queries of Google." He called it "a pretty amazing statistic" given that ChatGPT started offering up-to-date information only about a year and a half ago.
Kantrowitz pressed on what counts as a search. Fry's answer: "Every time you ask a question and go search the web, that's a search." So a single conversation with 20 turns could count several times if several turns trigger a web lookup. Google counts searches differently, and Alphabet does not publish a daily query count. That makes the 20% figure a ratio of two numbers measured with different rulers.
Three questions stay open:
- Numerator. Does a query that ChatGPT answers from memory, with no web call, count? Fry says no, but OpenAI has not published the filter.
- Denominator. Which Google figure did OpenAI use? Estimates of Google's daily searches vary widely.
- Overlap. Many people use both. Weekly users of ChatGPT are not a separate population from Google users.
None of this means the claim is false. It means a reader should say "OpenAI estimates" and not "ChatGPT has reached 20% of Google" when repeating it. Fry also gave a useful hint about why the number grew: people moved from writing and coding help to lookups on people, places, finance and health, which need current facts.
What about voice, images and subscriptions?
Fry gave several more numbers. Voice, which he defined as dictation plus voice conversations, has 250 million combined weekly users, which Kantrowitz noted is about 20% of the base. Fry argued it is "incremental to the product" and credited the GPT Live 1 and 1.1 models for more natural backchanneling and interruptions. Our post on GPT Live 1.1 and its new voices covers the model side.
Users create 1.5 billion images a week, he said, with growing productivity uses such as menus and diagrams. On money, he said OpenAI had passed 50 million subscriptions "several months back" and that ads reached more than a billion dollars of annual recurring revenue within 200 days. For the ads product itself, see our claimed-versus-verified post on ChatGPT ad targeting.
Again, all of these are statements, not filings. A useful habit: treat each as a data point to compare against the next official OpenAI release, not as a base for a budget.
Can Dots pay for themselves?
This was the most interesting exchange for builders. Kantrowitz asked about estimates that serving a $100 Dots subscriber might cost OpenAI $3,000 in compute. Fry did not confirm or deny the figure. He said "Dots are on the compute-heavier side" and that OpenAI wants to "dial back the compute needs" over time. His case for the economics rests on three things:
- Newer models are cheaper to serve. He pointed to GPT-6.1 Sol being "much more efficient" than Astra, which matches the picture in our cost-efficiency analysis of Sol and Astra.
- Dots will move to "more efficient, agentic models" instead of the best frontier model for every step.
- Subscriptions, credits and ads together should cover the cost.
Asked about Elon Musk's complaint that Grok Bot's serving cost was unsustainably high, Fry said there are two factors, cost and revenue. We looked at that claim in the Grok Bot serving-cost post. He stressed OpenAI's diversified revenue (subscriptions, credits and ads) and said he is confident a Dot "is going to be able to pay for itself."
On access, Dots stays gated to the top plan. Kantrowitz said he upgraded to $100 a month to get it. Fry had "nothing to share tonight" about a broader release date. Our earlier coverage of the Pro $200 usage cut for Dots shows how tight the allowance already is.
Fry also described the trust side. Dots are deliberately conservative about permissions: Kantrowitz said his Dot would not accept a credit card. Fry called it "a whole new level of trust" and said OpenAI wants to be "a trusted player". For a concrete case of where trust broke, read the unapproved email incident.
What does the Chat and Work merge mean?
Fry agreed that asking users to pick Chat or Work "is too much to understand." The plan is a single interface that works out whether you want a quick answer or a long agent task. He called it "real research" because the model must infer intent. His timeline was two words: "Coming soon."
If you build on the ChatGPT apps platform or write guides for teams, expect screenshots and onboarding steps that mention the toggle to go stale. OpenAI has already been testing a combined desktop layout, described in our tabs-merge report.
What this means for builders
Distribution is real, but unaudited. If ChatGPT really serves 1.2 billion people a week, apps built inside it reach a very large audience. Check any pitch that depends on the number against OpenAI's published developer metrics, not an interview.
Search habits are shifting. If a fifth of lookup-style queries now happen in an assistant, content that is easy for models to cite matters more. Keep pages fast, factual and crawlable. OpenAI's search crawler behavior is something site owners can check in their own logs.
Plan for simpler interfaces. Fry named simplicity as one thing a rival got right, pointing to Meta's Muse app. Fewer toggles means more of the work of choosing a model and effort level moves from the user to the product. Our comparison of proactive agents shows how each lab handles that choice.
Cost still shapes access. The reason Dots sits on a $100 tier is compute. When Fry says efficiency will improve, that is a bet. Price your own agent features on today's costs, not on promised ones.
How to check these numbers yourself
- Look for an OpenAI statement or blog post that repeats the 1.2 billion and 250 million figures, and note the date and wording ("weekly active users" is not "monthly" or "registered").
- Compare with third-party panels and web traffic trackers, which usually measure web visits, not app users.
- Ask what the denominator is whenever a share-of-Google number appears.
- Watch for Google or Alphabet commentary. Alphabet reports earnings quarterly and has so far not published a daily search count.
- Revisit this post: we will add an update section if OpenAI publishes a methodology or a filing.
Where the interview is thin
A few topics got only a line or a deflection. Fry did not give a date for Dots on cheaper plans. He did not give a cost-to-serve figure. He did not explain how OpenAI deduplicates users across devices and logged-out sessions, a standard question for any weekly active user count. He also did not say whether the 1.2 billion includes people who only use voice dictation. These are the details that decide whether 1.2 billion is a clean count or a generous one.
The interview also touched education, advertising and public unease about AI. On the last point, Fry said polling is negative because AI "is going to change the nature of how work gets done," and that perceptions improve once people use it. That is an opinion, not data.
Bottom line
The most useful reading of this interview is as a snapshot of how OpenAI wants its consumer business described: fast growth, search-sized usage, voice and images at scale, and an agent product that will get cheaper. Each part is plausible, and some parts match earlier OpenAI statements. None of it is independently verified yet. Quote it as "OpenAI says," link the interview, and wait for numbers that come with a method.
Figures and statements are accurate as of November 3, 2026, as reported by Big Technology. Check OpenAI's own pages for updates.