OpenAI says ChatGPT growth is accelerating.
The company added 200 million users over just the past two months, jumping from 1 billion to 1.2 billion weekly active users in 60 days. Its search volume, it says, is approximately 20% of Google’s. Its users create 1.5 billion images with it per week. And its voice mode, between dictation and speaking with ChatGPT, now has 250 million combined weekly active users.
ChatGPT consumer product lead Adam Fry told me this during a brief fireside chat at an OpenAI-hosted dinner in New York City on Monday night. “The growth is just going big,” he said.
During our conversation, Fry also revealed plans to merge ChatGPT’s Chat and Work modes (so that confusing toggle up top will soon go away), agreed with feedback that ChatGPT-created restaurant signage looks too uniform, and addressed the cost to serve its personal assistant Dots.
Here’s a look at our full conversation, lightly edited for clarity and length. We also cover image generation, interactive interfaces, advertising, education, and public unease about AI:
ChatGPT’s search growth
Alex Kantrowitz: OpenAI has analyzed ChatGPT’s search volume and compared it with Google’s. What did you find?
Adam Fry: ChatGPT has been growing quite quickly. ChatGPT’s volume of search-like queries is now roughly one-fifth the volume of queries of Google.
It’s a pretty amazing statistic when you think about it, mainly because we started just about a year and a half ago. It’s hard to remember this, but ChatGPT at a time didn’t have any up-to-date information. You’d ask a question, it would say something like, “It’s past my training date,” and nobody had any idea what that meant.
You’re saying ChatGPT’s search volume is one-fifth of Google’s. Bing’s share of the search market is around 5%. Does that make you multiple times larger than Bing?
To say that more directly, ChatGPT has gone from one billion to 1.2 in just the last two months. That’s roughly 100 million users in 30 days, and another 100 million in another 30 days. The growth is just going big, and we’re staying very focused on how fast ChatGPT is growing.
What counts as a search? I might have 20 prompts back and forth in a ChatGPT conversation. Does each one count as a query?
Story continues below…
A MESSAGE FROM OUR SPONSOR
Move upmarket without a rebuild
Moving upmarket means saying yes to what enterprise buyers ask for first: SSO, directory sync, and audit logs. WorkOS gives you all three, so your team keeps building the product instead of the checklist.
Continued from above….
Adam Fry: What people came to ChatGPT for to start, if you go back about two years, is really writing help and coding help. Those wouldn’t be things that you really search for.
What we’ve seen over the last year is people have grown into all these other topics. Things like you’re looking up a person or people or a place or a thing, and it requires up-to-date information. That’s a key element.
Another big part is finance questions or health questions, where you really want to make sure it’s up-to-date information that’s quite factual, and we’ll do a web search for those questions.
It’s basically anything that a user would have traditionally gone to a traditional search engine for, but is now deciding to go to ChatGPT for.
Does each conversation with a search embedded count as one search, or does every question that requires a web search count?
Every time you ask a question and go search the web, that’s a search.
You used to work at Google. How would you say the nature of search is changing?
One of the things that people really love about ChatGPT is this personalization and this understanding of the broader journey.
If you look in the past, people would go for a Q&A box, and you’d ask a one-off question, like how tall is the Eiffel Tower. But there’s so much context in our lives around, “Maybe I’m taking a trip to Paris.”
It’s actually helpful for ChatGPT to respond, “Here’s how tall the Eiffel Tower is, but I know you’re going on a trip to Paris.” It’s that ability for people to actually maintain context over time that has led them to start using ChatGPT in this way.
Voice reaches 250 million weekly users
You’ve also announced 250 million weekly active users of voice input, including dictation and conversations with voice mode.
We are at 250 million weekly active users. What’s amazing about this is, if you look at those 250 million, what we’re seeing is it’s actually also incremental to the product.
It’s no longer just a novelty. People aren’t just trying out voice. We’re seeing them use voice input day to day in their daily life, and it’s grown so quickly.
Those 250 million users are using voice input, and that could just be a quick question or a long conversation. We’re seeing both behaviors. We’re seeing people just not want to type. You just hold it and start talking to it, or it turns into a longer conversation where people just sit with it.
The big reason for this is the frontier capabilities of our models that we’ve shipped. We shipped what we call GPT Live 1, and then we shipped GPT Live 1.1. The big change with this is that it’s much more natural to speak with.
It’s something we call backchanneling and interruptibility.
Mhhhm…
What you just did there is what the model does, because it’s really good at having a bidirectional conversation. That push to the frontier—we were the first to really release this at scale—has made it so much more natural for people to chat.
That’s still only about 20% of the user base. From the outside, that looks small. Why is it exciting from the inside?
It’s very exciting for a few reasons. We see from the data that when people start using voice, they stay around using voice.
One of the things, transparently, we can do a better job of is bringing more people into discovering it. We see that the number of people who discover it could be a lot higher. But once they do, it’s a really new way for people to interact with AI.
We’ve made a lot of improvements to it too, where we’re just going to see the number go up. You have background mode, so now by default, if you close the app, it’ll show, “Hey, the conversation’s still going.”
If you want to walk around the house or you’re putting away the groceries, you can still have a chat. That kind of continuity has also changed how much people use it.
Dots, permissions, and trust
Dots launched on September 29 and is about a month old. How’s it doing?
It’s doing great. We talk to people about their Dot, and the first thing that everyone says is, “I love my Dot.”
One of the reasons we’re seeing this is the proactivity, its ability to reach out to you and figure out things. The biggest use case I see all the time is, “Did I forget something today that I was supposed to do?”
You can just say to your Dot, “Remind me of things at the end of the day that I may have forgotten. Somebody emailed me or messaged me,” and it’ll do that.
This Dot changes the paradigm of the relationship with AI, where it’s actually helping you and giving you time back.
You’re more conservative with permissions than some competitors. I tried to give my Dot my credit card, and it wouldn’t take it. With the creator of OpenClaw in-house, why are Dots’ permissions more limited?
This is a space where we want to be very thoughtful and conservative because it’s a new space where you’re entrusting your Dot to take care of things on behalf of you that matter and are meaningful.
We’ve just talked about how you go to this 1.2 billion user growth and asking questions. Now we’re asking people, “Can you ask ChatGPT or your Dot to help get things done?” But when you ask it to help get things done, it’s a whole new level of trust.
We’ve built up a tremendous amount of trust in giving the right answers to your questions, and we want to make sure that when we build this new product, it has a high degree of trust.
It’s a long game. We think this is going to be a new market for a very long time, and we want to make sure that we’re being this trusted player in the market to start.
I initially thought Dots would be available on my $20 plan, then realized I had to spend $100 a month to get it. I upgraded, but I’d like to downgrade at some point….
We hope to keep earning your business and make the value good enough.
It’s a really good point. ChatGPT’s goal more broadly, which I think is unique for us in the market, is we are focused on bringing intelligence to everyone safely across the entire world. For Dots in particular, there’s no difference.
We don’t have anything to share tonight about the date of when Dots will go to a broader set of users. But you can rest assured we’re working very hard, and there are a couple things we have to get right.
Over time we’re going to work on more efficient, agentic models that are able and capable of doing things for you.
The first one is, right now we wanted to make sure that your Dot is actually really helpful. We started with some of our best models, but over time we’re going to work on more efficient, agentic models that are able and capable of doing things for you.
The second thing is around proactivity. We’re still dialing in: How much do I message you to help you get things done? We can always play around with that to understand how much usage should a user use to have a successful Dot.
We’re going to work on both, and our goal and ambition is to bring it out to all users.
Can personal agents pay for themselves?
How much background computing does a Dot require? I’ve read estimates that serving a $100 subscriber might cost OpenAI $3,000 in compute, and mine seems to read every email that arrives. How often is it checking connected data sources?
Right now, Dots are on the compute-heavier side, and this is one of the things we want to make sure that we can dial back over time. But the goal is to give you the best product.
This is our premium, always-on personal assistant. Once we nail that, we’ll figure out how to dial back the compute needs.
The thing we’re seeing also from our models, as we train frontier models, is just how—I can’t even keep up with how quickly we have a new model, and each new model is more efficient.
If you looked at 6.1 Sol when it came out relative to Astra, much more efficient. I’m very excited about the next set of models we have coming too that’ll be even more efficient that we can use.
Elon Musk said the cost of serving Grok Bot was unsustainably high relative to revenue. Is efficiency the key to winning against competitors, or is there more to it?
There are a couple of factors that go into it. One is the cost side, but the other is the revenue side.
On ChatGPT and on our consumer business, we’ve been extremely proud of our ability to diversify different revenue streams in a way that’s growing really fast.
To give you another stat, we announced several months back that we had passed 50 million subscriptions, and the number has grown considerably since then.
On ads, within 200 days, we reached more than a billion in ARR, and that business is growing extremely quickly.
We also have credits. Rather than paying for subscriptions, users can buy more credits. Between credits, subscriptions, and ads, I’m confident that the value of a Dot is going to be able to pay for itself.
Elon has subscriptions too. This is clearly an expensive product to serve.
One thing we continue to bet on at OpenAI and ChatGPT more broadly is that the models will continue to get much better and much cheaper, way faster than any of us think.
If you look back at the historicals on it, it continues to be true. From what I’ve seen, I see no slowing down of that.
What Meta gets right
What is Meta doing right with Muse? It has spent a lot of time atop the App Store charts.
There are a couple of things they’ve done well. The first is people have been craving simplicity in their products.
One of the things we have stated publicly as well is, at times AI products can be too complex. We’ve announced we’re merging ChatGPT’s Chat and Work. We’re making it much more simple in the products. You don’t have as many toggles or sliders between efforts and different model pickers.
One of the things they’ve done well is this simplicity. You just go in and you can ask a question. You don’t have to think too much about what you’re doing. That’s one thing that seems to have resonated with users, which we also believe in.
Second, there’s something really nice about being able to access your agent across many different channels. We just announced and launched Dots on iMessage.
Whether you talk to your Dot on email or iMessage or within the ChatGPT app, the ubiquity in getting your assistant out there is going to be key. Those are two things that, if I had to be nice to our competitors, I would say were two big ones.
So Alexandr Wang’s meme-tweeting strategy isn’t something you’re going to emulate?
I have a different tweeting style.
Merging Chat and Work
ChatGPT still asks me to choose between Chat and Work, even though some people call this technology AGI. Will that distinction go away?
Couldn’t agree more. We have to solve this. Asking users to choose — we’ve heard repeatedly from all of our users it’s too much to understand. When should you go to Chat? When should you go to Work?
It’s solvable. There’s real research to solve to make sure the models can understand what your intent is. Are you asking for a fast, quick reply? Are you asking me to go do a long agentic task?
We’re going to solve it, and the plan is to bring Chat and Work together.
So one interface will work out whether I’m asking for a conversation or a task. When?
Coming soon.
Images and interactive interfaces
You told me the majority of ChatGPT’s 1.2 billion users are using image functionality. What are they doing?
Today, 1.5 billion images are created per week. It’s hard to even wrap my head around that number. They’re doing a wide range of things.
One of the most interesting things they’ve started to do is really use it for productivity use cases.
The first thing it started with—I don’t know if you saw the ’80s trend, but it was quite fun. You could take a photo of yourself and say, “Put me and my family in ’80s outfits.” I did this for my family, and it was a really fun poster.
But what we’re also seeing is businesses for their menus, or posters for birthdays. People are using it almost in a productivity-like fashion to create visual diagrams of what they would like to see.
We’ve actually brought the latency down on images as well, so it’s gotten a lot faster. It went from a very high number of seconds to getting much faster. As it comes down even further, the number of images people are creating and the number of use cases is going to grow a lot.
My feature request: When a restaurant makes a menu or sign, prompt it for more specificity so it doesn’t look like every other AI-generated restaurant sign. Or offer a shuffle button for a different style.
I agree with the feedback, and we’ll take it to the team.
I’m seeing ChatGPT generate little interactive interfaces within its answers. What’s the direction for Intelligent UI, and how are people using it?
I am extremely excited about Intelligent UI. If you go back a couple years, these chat apps were all text, and now that’s changing. They’re going to be visuals. Intelligent UI and our first attempt at this with GPT-6 Intelligent UI is really just the beginning.
As you go to the extreme, what you can imagine is, let’s say you’re using voice mode, and you just say, “Show me how offside works.” Immediately it should show you a visual diagram, and you say, “No, I don’t quite understand it.” It should just repopulate.
That’s where the future of this technology goes. It’s almost like you’re talking to a canvas that will update and change.
People are finding it helpful for a lot of things. My favorite one is—it’s hard to admit that I had no idea how hearts work. The last biology course I took was in high school. I just asked it, “How does the heart work?” and it instantly gave me a beautiful interactive diagram of where blood flows and how it goes through.
I’m really, really bullish on this for learning in particular.
You’ve described useful applications in finance, health, and learning, and ChatGPT has 1.2 billion users. If people are getting so much value, why is public polling on AI so negative?
The reality with AI is it is going to change the nature of how work gets done. For a lot of people, that is something that can feel scary at times, and we’ve all felt that.
But I remain confident that when people actually use the product and they use AI, it very quickly changes how they perceive it. We’re at 1.2 billion, but there are still a lot of people who haven’t used it.
The more people use it, get familiar with it, how it can actually help you in your life, it’s going to feel a lot less scary.
the restaurant sign question was the best one in here