2025年使用最广泛的100款AI应用

Hi, I’m Justine.

And I’m Olivia.

And welcome back to the a16z podcast. Today, we’re going to be discussing the Consumer AI Top 100 list. So, Olivia, let’s start because you are compiling the list and you’ve been doing this for a while. What is this list, and what’s the purpose of it?

This is our 5th time doing this list. We do it every 6 months. We started at basically the dawn of the consumer GenAI era, and the purpose is just to get a sense of what real consumers are actually using in AI.

As a sample set, we take every website and every mobile app in the entire world, and we sort them in descending order of usage. For websites, this is monthly visits globally. For mobile apps, this is monthly active users. We use Similarweb and Sensor Tower as our 2 data sources, and then we grab the top 50 from each of those sources that are AI-native companies.

You’ll notice that this captures usage, not revenue. It’s not paid usage only; it’s all usage, including people who are using the products for free. To me, that gives an even better—and more real—sense of what is capturing consumer imagination in AI.

Totally. I feel like a common question we often get as folks who are consumer investors in AI is, “What are people doing with AI besides—”

“What should I be doing?”

That too, yeah. I feel like these lists are a really helpful way for folks to get a sense of what some of the use cases are and what some of the products are that are getting really big and a lot of users—products that you might not know about. For a lot of people, it’s an inspiration list of what they want to try out.

Let’s talk about the new companies on the list, because that’s always exciting: Who cracks into the top 50 for web and the top 50 for mobile who weren’t on the list before? How would you describe the new companies? Were there any major changes?

The web list is the best way to track changes over time. The mobile list is a little bit less stable because the app stores are constantly updating their policies on what’s allowed.

For our first couple of mobile lists, it was just a wave of companies that were direct copycats of ChatGPT.

I remember that. Yes.

All the companies were called things like ChatGPT or Chat and Ask AI.

ChatGPT was a big one.

Exactly. Those were dominating the mobile list. This time, we had a lot of new companies on mobile as the iOS ecosystem has cracked down on that.

But on the web list, we still had 11 new names. That sounds like a lot out of 50, but it’s actually a big reduction from, I think, 17 new names 6 months ago.

Wow.

To me, that reflects the fact that the ecosystem is starting to stabilize a little bit more. There were still a couple of really exciting trends that popped out on this list that I would expect might continue and yield more new names.

Vibe coding was one of them, which I think burst into the forefront at the end of last year. We saw it a little bit on our prior list in March, but it really came into the forefront now. We have Bolt.new on our brink list, just below the top-50 cutoff, and then we have both Lovable and Replit on the web list.

In the past, a lot of the names on the list have been in 1 of 2 categories: general chat and companionship, or creative tools. I love both of those categories—I’ve spent a bunch of time in both of them—but it sounds like we’re starting to see things like vibe coding that are outside of those 2 categories, which were really the killer use cases of consumer AI for a long time.

We definitely are seeing more productivity as well, which we’ll talk about. I will say that companionship continues to dominate. We actually had 3 new companionship names: JuicyChat, Joyland, and DreamGF all made the list, alongside a bunch of others that have made the list before.

Character.AI was the biggest name. Janitor AI, SpicyChat, PolyBuzz, CrushOn.AI, A-Dot, and Candy AI were all on this list and have been in the past. That’s a pretty notable section of the list that is dominated by companionship names, which is pretty interesting.

For sure. There are also a lot of big companies that have products on the list, like Google. Maybe you can talk a little bit about that, because I think that’s been a surprise.

Google had a big 6 months, especially with its models. Even its consumer-facing products had a really big 6 months.

The way Google structured its domains in the past made it hard for us to include Google products on the list, since we needed to be able to independently rank their traffic. Now, for the first time, we were finally able to, and 4 unique Google properties made the web list.

Wow.

Gemini was number 2, right behind ChatGPT. It has just about 10% of ChatGPT’s traffic on the web, but it’s much closer on mobile. It has about half of ChatGPT’s traffic on the mobile app, which is interesting. It’s mostly Android users, which I think you would expect.

The other big debut for Google was a little more surprising to me. It’s actually AI Studio.

Which is Google’s developer-facing sandbox.

That hit the top 10 as well. And then right below the top 10 was NotebookLM.

Wow.

At number 13. I take your “wow” as a little bit of a surprise, which was a surprise to me too, and I think a bit of a narrative violation. They went viral almost a year ago, in September 2024, but they’ve actually kept up their traffic every month, with flat or increasing traffic, with the exception of this last month of summer, when I think a lot of the academic usage caused it to decline—but just barely.

One of your favorites also made the list: Google Labs, down at number 39.

This is the consumer-facing sandbox for Google models, whereas AI Studio is a developer-facing sandbox. Google Labs includes Veo 3, the amazing new video model that you yourself have probably driven a big portion of the usage for.

Spent a lot of money on Veo 3. Exactly.

And driven many visits, for sure. Google Labs includes a bunch of other products, like Doppl, which is its outfit-try-on product. It includes Portrait, which is a coaching product, and Whisk, the image-generation sandbox, which is very cool and fun to play with.

Exactly. Project Mariner, which is its agentic browser sandbox.

Interesting. But my hunch is that it’s largely Veo 3, because traffic to Google Labs spiked 15% in the month that Veo 3 was released, which was really exciting.

Let’s talk about the Chinese AI companies. We’re seeing both new startups coming out of China and big Chinese companies releasing models or consumer-facing properties. How are these ranking on the list, and what do you think has driven their growth?

You could argue that Chinese companies show up in 2 or even 3 pretty interesting ways on the list. The first are companies—or AI products—that are built for China and used in China.

And you might say, “Why?”

It’s because a lot of non-Chinese AI products, like ChatGPT and Claude, are banned in China. They have to follow specific regulations and policies and data-capture rules that those companies aren’t willing or able to follow if they’re going to operate in China.

China doesn’t have a lot of the products that we have, but it has its own. A couple of companies that made the list are Qwen, Alibaba’s AI assistant, on both web and mobile; Doubao, ByteDance’s AI assistant, also on both web and mobile; and Kimi, another general AI assistant from Moonshot AI.

All of them rank in the top 20 on the web, with the majority of their usage in China. That reflects the fact that China is the biggest country in the world by population. Even if they’re only in one place and don’t have competition from the rest of the world, they can still get a lot of usage.

The second category of Chinese companies we’re seeing are startups that are developed in or around China but are built for the rest of the world. In many cases, you can’t even use them in China. These are products like DeepSeek, video models like Hailuo and Kling, and products like Krea for images.

I would say the concentration of these is especially strong in image and video, which is a space that you know well.

Yeah, I think image and video have seen a lot of great models from China, particularly when it comes to animating from a single frame or image. ByteDance has also been doing a lot more around things like Seedance, which is a really great video model. They released that, and it actually outscores Veo 3 on some of the arenas.

One of the really interesting things that we can’t disentangle in this data is that I think a lot of Chinese companies have taken the strategy in the U.S. of distributing their models through U.S. properties. There are companies like Krea or Hedra that host various models and have subscriptions where you can access a bunch of the Chinese models at once, whether it’s closed ones like Kling or MiniMax, or some of the open-source ones like Wan or Qwen, which a lot of people are using and really liking.

Some of them will also have their own consumer-facing properties. Others will distribute through either the application layer or, honestly, through folks like Fal.ai and Replicate—more developer-facing platforms where you can do one-off runs or have an API to use them in your application.

That is probably yet another way that Chinese companies are showing up on the list in a way that, as you said, isn’t measurable right now.

But the last trend that we were seeing with Chinese companies was products that are developed in China and used both in China and abroad. Manus is a great example. They just announced a $90 million annualized run rate. And if you look at their traffic, it's number 1 Brazil, number 2 the U.S.; number 3 surprises me.

Yes. A lot of AI traffic comes from Brazil. More free traffic than paid traffic.

I was going to say yes.

But a lot of traffic. So we're seeing both this domestic AI ecosystem in China start to mature, I think, for the very first time on this version of the list, and then we're seeing the same thing we've seen on other versions of the list, which is Chinese companies exporting their AI products abroad.

And then, to talk about another big trend—which you touched on earlier—vibe coding: We're probably all seeing the tweets from various vibe-coding companies that are like, “We went from 0 to $100 million or even more in ARR in a really short period of time.” They have millions of users, kind of faster growth than I think we've seen in many other categories in a very long time. I'm curious: How did you see that reflected on the list, and how did that change since the last edition?

Yeah, so 6 months ago it was just Bolt on the list. Now Bolt is actually on the brink list, and Lovable and Replit both made it onto the main list. To me, I was particularly interested to look at not just traffic for these, but also revenue and retention. Lovable recently announced $100 million in ARR.

But it's always easy to say, especially for consumer properties, “Yeah, a lot of people are maybe signing up for a trial,” but they're canceling. They're not going to stay on. Retention is low.

We actually also, for the vibe-coding products, looked at revenue retention across another data provider, Consumer Edge. So that looks at a cohort of everyone who signs up in month 0: How much are they paying, and what percentage are still paying in month 1, month 2, and month 3?

We put some of this in the report, but what we found for many of the leading vibe-coding platforms is they actually see 100% or above revenue retention in the first 3 months, and then maybe flatten out a bit below 100%.

Which, as a point of comparison, is very strong. We see that for—I mean, it's very rare to see that in anything, especially a consumer or prosumer product. It might suggest that there's some enterprise usage happening there, with prototyping and stuff within companies.

Enterprise usage, or even just solo vibe coders: people building projects who do get a low-cost plan to try it out and then find themselves actually publishing and using something, buying more credits, and upgrading to the next plan. Which is exciting.

It's interesting, though, on that point: For both Lovable and Replit, we can independently see traffic to the sites where things you or I make are hosted. They're hosted on lovable.app.

Yeah. So you can see traffic for everything people are making with Lovable and Replit, and then separately, traffic to Lovable and Replit, which is the traffic from the people making things.

And in both cases, traffic from the people making things is much higher than traffic to the things that they're making. Yes.

Which suggests one of 2 things. One, it could be that people who are making serious, high-traffic websites are buying domains for them—a custom domain—so they no longer pop up under the Lovable or Replit domain in this data.

So their traffic would not get included. Or it could be the case that people are building almost personal software—software for themselves—with these vibe-coding platforms. So they might publish them and not get a lot of hits, but they're extremely valuable to the individual user, or maybe the user's family and friends.

And I'm curious: Have we seen anything yet in vibe coding coming to mobile? Like, are there?

Not yet.

Interesting.

There's a couple of startups that we've seen, but none of them ranked high enough to get a spot on this list this time.

Yeah, it'll be super interesting to see the next edition of the list if any of those start popping up in the top 50. It's hard to displace the companionship and the AI girlfriend ones for sure.

But maybe a vibe-coding one can do it.

Yes. No, it's actually a good point because, for the first time, on the point of displacement, we ranked what I'm calling our AI All-Stars. So this is across 5 lists: What are the companies that have made the list every single time? Which is hard to do because this dates back 2 years now.

So there are 14 of these companies. So we have ChatGPT, not a surprise. Perplexity and Poe all made the list

in what I call kind of general LLM assistance.

Right.

Then we have Character AI in companionship. They've made it on both web and mobile. We've got Midjourney, Photoroom, Leonardo, Cutout Pro, VEED, and ElevenLabs in creative tools. We have QuillBot and Gamma in productivity. And then we have Hugging Face and Civitai in model hosting.

And so those have made the web list for every single one.

Web list for every single—every one of all 5 versions we've done of this.

Now, it was funny looking back at the very first list because I think we had this question of whether it was only going to be companies that invest tens of millions of dollars into proprietary models that are able to hold consumer attention.

But now, when we look at these 11 All-Stars, more than half of them are actually hosting or using other people's models or are model aggregators, right? And so I think that speaks to the fact that in consumer AI, the UI and the product experience matter just as much as the model, especially when so many amazing models are now API-available or open source.

If you think of some of those properties, too, some of them have built really deep and interesting workflows around creating content, even if they don't have their own model. Some of them are also starting to see what I would say are sort of the first kind of non-data network effects in AI, especially things like Hugging Face and Civitai, where you have these communities where people are hosting tons of models or LoRAs or datasets, they're commenting on them, they're ranking them, they're spinning up many apps that you can only access on that website.

So it's pretty cool to see both of those make the list over and over again and show that you can maybe have network effects in AI.

Yeah, I think you could argue, too, ElevenLabs is the same, where people can publish their voices on ElevenLabs for other people to use.

The voice libraries—I hear that all the time from people—that one of the reasons they love using ElevenLabs is there are thousands and thousands of voices you can choose from, and that's because so many folks have uploaded voices to the voice library.

You could even argue the same is true for ChatGPT or Midjourney, where the more user data they get, the better models they can train, which then makes the product better and attracts more users, and it's a whole virtuous cycle.

Yeah, I think that is the natural assumed network effect for AI: more users, more data, more feedback, better models. But I know a lot of folks are wondering, where are the network effects beyond that? Which is why I was particularly interested in the Hugging Face and Civitai of the world, things like the ElevenLabs voice library.

Yeah, we're even starting to see in some cases—Gamma is a good example of this on the list; ElevenLabs is another good example. Same with Photoroom. These consumer AI properties are almost graduating to enterprise usage with things like team plans, templating, and design libraries, and then it becomes something that can spread organically through a team and that a whole organization can benefit from.

The more you invest into building out, say, your templated slide deck on Gamma, the harder it is to turn and move on over to another platform. So, yeah, I think that's one thing that a lot of people didn't expect with a lot of these, especially creative tools or productivity tools.

What I try to tell founders all the time is, like, there are so many AI enthusiasts now. Well, some are tourists, but some have real use cases. Often with companies like ElevenLabs, we will see one individual sign up, try it out, and see if they like it or not. Then that individual actually becomes their own self-serve enterprise sale.

They bring it into their company, they make the case for it, and then it's so easy for them to go to the business and say, “Hey, we actually have 10 or 15 or 20 users on individual plans. We should have some sort of enterprise-level contract here.”

Which makes it a lot easier for these AI companies to grow than the traditional top-down sales that enterprise software had to do, where you had to go in and get approval from 10 different people and then convince the employees to use the product, versus this bottoms-up adoption that we're seeing with a lot of these premium AI products that start in consumer. It feels very cool and unique.

Yeah. And that in turn supports more consumer growth because then someone on a team will adopt it because their other team member had it, and then they'll bring it home and use it for a project and spread it to their friends and their kids. And, yeah, I expect we'll continue to see more prosumer enterprise representation on this consumer AI list.

So, as you've mentioned, this is the 5th edition of the list. I'm curious, reflecting back on all of the past editions, what are your biggest takeaways, and do you have any hot takes or predictions for what we're going to see in the next few lists?

Yeah, I would say looking at the list now and then looking back, I really can appreciate now that the first 1 or 2 lists were total chaos, in the sense that half the list was new every time, if not more.

And the traffic was fluctuating so wildly every month. There’s often a 2-week-to-1-month delay between when the data happens and when we get it through the data sources, and so founders would be contacting us, angry: “Hey, I’m actually ahead of this other property now. I should be number 2 instead of number 5,” just because the data sources had a little bit of a lag. Things were so volatile back then; things were changing so rapidly.

I would say, as you said, companionship and general LLM assistance like ChatGPT, and then creative tools were kind of the 3 categories that were working 2 years ago. Anything else that popped up on the list, in many cases, was pretty random and would spike 1 month and be gone the next month.

I would say we’ve had a lot of stabilization since then. By the third edition of the list, we started seeing things that weren’t just image and text.

We started seeing video. We started seeing music, which was really exciting. The 4th and 5th editions of the list still have some breakouts, but we’re also starting to see the list kind of normalize.

Right? You said 11 new on the web this time versus 17 last time.

Yes, exactly. So, even list over list, there are fewer newcomers. There’s a clear list of all-stars, and the newcomers that we are seeing are kind of clustered around many of the same themes, like vibe coding and companionship on this edition of the list.

I’m bought into this all-stars concept now. We should maybe make jerseys for it or something.

I know. And I’m very—

Proud of them.

I hope that they are proud of themselves, and I hope that they’re rooting for themselves.

Send them trophies or something.

We should. We absolutely should.

I would say when I’m looking at the next 5 lists, there are a couple of things that I have top of mind.

Our partner Anish wrote an article about how AI products are kind of verticalizing even within these general products. You’ll go use Gemini for one thing, Claude for one thing, ChatGPT for another thing, and Perplexity for another thing. I would expect that to continue. We’re not seeing that this is necessarily a race that ChatGPT is running away with, especially on mobile. People have lots of different use cases, right?

Grok—we didn’t talk about Grok yet—but they had a big debut on both the web and the mobile list this time around, which is super exciting.

They popped up near the end of the period.

Exactly. In terms of what drove it, I imagine Grok’s companions and its image and video features are what drove a lot of the growth in the Grok app.

Exactly. Yeah. Grok actually debuted in the top 5, at number 4 on the web list, which is pretty surprising and exciting. Incredible.

We also saw Meta AI start to make a little bit of a dent this time around—not on the mobile list, but on the web list as well.

They have a web app. I should look into this.

Yeah, exactly. So we’re still seeing things heat up even in this space, where you might guess that ChatGPT has kind of run away with it.

For sure.

I would say the other thing I’m expecting to see in the next 2 lists in particular—

On the other end of the spectrum from creative tools, where hallucinations are the feature, there are prosumer tools where hallucinations are a real problem, right? These are things that can make spreadsheets and slide decks for you, build financial models, and answer emails. We’re finally seeing both model reliability and workflow and UI—

Get to a point where these products are exploding in revenue, and I think they’re going to explode in usage as well, for sure.

So I’m excited to see more in the prosumer product category on the next version of the list. Manus made it this time. Perplexity Comet, as its own domain, I would expect might make it. Genspark is another one that I might expect to see on future versions of the list.

I think one really interesting thing that people sometimes don’t realize about more of these professional-ish use cases—financial modeling, as you mentioned, or coming up with a presentation, doing a formal pitch or document, or whatever—is all of the evolution that we’re seeing on the foundation model side.

Grok 4, the new version of Claude, and GPT-5 aren’t just sort of in their bubble on the side, with the app layer separate. A lot of these applications are powered by all of those models. So as those models get better at things like math, logic, and reasoning, they have fewer issues with hallucinations. That naturally supercharges all of these products that are in the productivity, prosumer, accuracy-is-important space.

Exactly. And I imagine we’ll reach a point where it tips them over: “This was cool, but it was inaccurate too much of the time, so I can’t really use it,” to, “This is now good enough that it can do a lot of stuff for me, and I am willing to use it.”

I think that’s a great example. Even things like the GPT-5 release being very good at consumer health questions. We might expect to see a wave of new and more reliable consumer health products on the web or mobile on the next version of the list. Yeah, I was thinking about what categories we didn’t see a lot of that we might expect to see in the future.

More edtech, I think.

Edtech, for sure. Edtech is another one where it’s important not to have a ton of hallucinations.

Personal finance is one where we haven’t seen a lot on either web or mobile, but I expect to see a bunch more on mobile since it’s such a big category there.

Social. Yep. I think there have been a lot of people wondering when we’re going to see the first big AI-native social platform, which some could argue is now Facebook, which has largely been taken over by boomer AI images. I personally think there’s going to be at least 1—probably multiple separate ones—and honestly, Grok is sort of making a case there and being a social platform.

Health is a great category, as are some of these more structured coaching, therapy, and wellness types of products that I think a lot of people use ChatGPT for today but probably deserve to exist in separate products that will hopefully make the list in the future.

Yeah, I agree. My biggest takeaway, actually, from every version of the list is that, as with everything else in consumer, there’s so much randomness. If we were able to perfectly predict the next great consumer product, there’d be hundreds of people out there building it.

Just like vibe coding 9 months ago was a completely unexpected category to pop up on the list, I’m sure there will be 1 or 2 more in the next few editions, which is really exciting to think about.

Amazing. So, we’ve got to stay tuned for the 6th edition, it sounds like.

Exactly. Click through to check out the full version of the list, which is out now. We'd also love to hear in the comments what your favorite product or company is that made the list this time. Or if you have any products you use all the time that you were shocked to find did not make the list, we'd particularly love to hear about those. And we'll see you in 6 months for the next one. [Music]