Top 100 Consumer AI Apps, Explained
Summary
Consumer AI looks mature by traffic but remains fluid by revenue. Only 11 products were new across the combined web and mobile rankings, the series low, yet 29 of the 50 products ranked by consumer card spending appeared on neither traffic list. Olivia Moore’s conclusion: consumer AI is currently a “power user game,” and traffic alone misses many products with real monetization.
Personal agents are the emerging product cycle, but their headline growth has not yet translated into mainstream adoption. Instinct reported 100,000 users growing 10% daily, with 40% connecting a credit card within three weeks and spending more than $1,000 on average in month one; Olivia recalled Muse figures of about 500,000 downloads and 250,000 active users in roughly 12 days. Yet Muse’s approximately 5 million U.S.-and-Canada downloads over about 22 days remained far below Threads’ 16 million, partly because agents are personal utilities without the same urgent network effect—and because high serving costs make aggressive growth unattractive.
Broad awareness masks an exceptionally concentrated market. Nearly half of Americans report using AI and roughly 25% believe they interact with it almost daily, but only about 4.5% pay according to the cited Yipit panel. Within that cohort, the top 10% generate more than half of revenue, the top 1% generate 20%, and the top 1% spend $903 per month versus a $25 median: “The top 10% of users are driving this market.”
The mass-market unlock may be helping people invest time, not merely save it. Today’s biggest spenders over-index toward developer, productivity, and creative products such as n8n, Granola, Higgsfield, and Manifold; early agent communities likewise cite coding and technical automation as their leading use case. Olivia cites a point from Eugenia, one of the firm’s directors: “Most people aren’t looking to save time; they’re looking for ways to invest it”—the behavior behind entertainment, social media, and other high-frequency consumer franchises.
Subscriptions finance today’s expensive inference, but ads and transaction fees are the more plausible route to internet-scale adoption. Roughly 85% of the web list monetizes through subscriptions, 62% also uses credits or token payments, and only 13% offers ads or another model where users do not pay directly. Olivia therefore does not necessarily want the paying share simply to expand: “We’ve moved beyond the need for everyone to buy an AI subscription,” while lower inference costs should eventually reopen advertising and commerce models.
OpenAI’s early advertising ramp shows both the upside and the trust risk of conversational intent. The episode cites an annual advertising expenditure of $1 billion as of August and 1.2 billion weekly active users. Elena argues that live conversations and accumulated context could enable segmentation even better than Meta; Olivia says this knowledge could support higher prices and conversion. But a poorly timed health or weight-loss ad could destroy the intimacy that makes the product valuable; the winning implementation must feel “like a natural fit in the conversation.”
As models improve, durable value is moving toward interfaces, proprietary context, and multiplayer networks. ChatGPT still leads consumer usage and revenue, but specialists such as Suno, ElevenLabs, Midjourney, Granola, WhisperFlow, and Town retain differentiated workflows or power-user economics. Town illustrates the prospective moat: an email that sounds “99.9% like me” needs seconds of correction, while one that sounds 85% right takes more than 10 minutes—and those accumulated personal playbooks are difficult to migrate.
Deep dive
1. Stable traffic conceals the beginning of an agent cycle
The seventh Top 100 report had to move beyond browser traffic because AI is no longer principally a dialog box. Adding consumer card data changed the picture: just 11 products were new across web and mobile, but 29 of the 50 spending leaders appeared on neither traffic ranking. Olivia’s read is not stagnation so much as establishment—incumbents are growing while meaningful businesses emerge below traffic’s line of sight.
OpenClaw captures how quickly an agent wave can crest and mutate. It would have ranked highly just before the previous cutoff, then disappeared this time after traffic collapsed; Olivia says she understands that its team was acquired by OpenAI, while Josh still credits its open-source foundation with turning AI from a productivity aid into something that “would actually help us get things done.” People bought Mac minis, connected personal data, and asked, “Wait, what could these products actually be?”
The freshest consumer assistants arrived largely within the last three weeks or month, too late for the formal rankings. Poke was an early pioneer, Tomo already has a large audience, and Olivia identifies Muse and Instinct as the biggest recent developments—products she could finally feel “comfortable giving to a family member” rather than tools reserved for technical enthusiasts.
Elena’s Threads comparison supplies the adoption check. Olivia recalled that Muse had about 500,000 downloads and 250,000 active users in approximately 12 days; Elena later put Muse at around 5 million U.S.-and-Canada downloads over roughly 22 days, versus about 16 million for Threads across the same markets and window. Josh’s rebuttal: Threads needed immediate critical mass, whereas a personal assistant can deliver value alone—and high service costs give providers reason not to push distribution as aggressively.
2. Trust and discovery are harder bottlenecks than agent capability
Agent ecosystems already create platform leverage and conflict. Elena says Amazon will not let Muse browse and purchase on its site, while Muse has hundreds of other partnerships, including Shopify commerce access. Josh expects markets and networks of businesses reachable through agents to become transformative, even before person-to-person network effects emerge.
The obstacle is that usefulness and intimacy rise together: “The more useful the product is to you as a consumer, the more it knows about you.” Email, personal disclosures, and credit-card access make assistants powerful, but also recreate the separation users maintain between work and personal ChatGPT accounts. Josh argues that software must learn human rules about which entrusted information may be shared and which must remain protected.
Capability may advance faster than consumer willingness. David Pulan’s Assistant Benchmark reportedly tracks more than 170 agents—possibly over 200 by the next morning—and organizes a community of more than 1,500 early users; their leading use case was coding and technical automation, even for consumer assistants. That helps explain reports that supporting some users costs hundreds or thousands of dollars per month, versus tens of dollars for more conventional consumer behavior.
3. A tiny cohort currently carries consumer AI economics
Usage is broad but payment is narrow. Nearly half of Americans report using AI, around 25% think they encounter it almost daily, and approximately 4.5% pay under the cited Yipit panel; other sources put penetration near 2%–2.5% or somewhat higher. The paying share has roughly doubled in a year, but it remains only a sliver of the addressable audience.
Concentration persists even inside that sliver. The top 10% of paying users account for more than half of revenue; the top 1% contribute 20%, against 16% from the bottom 50%. At the individual level, the top 1% spend $903 per month on personal credit cards while median spending is $25—evidence that current revenue measures intense productive use, not ordinary consumer adoption.
n8n, Granola, Higgsfield, and Manifold were overrepresented among the biggest spenders because users employ them to code, create, manufacture, and sell. Josh sees “a whole new generation of creators” paying personally to make software, art, video, and content they previously lacked the skills to produce: not simply saving time, but bringing more ideas to life.
Olivia says developers and coders are naturally attuned to this opportunity and that it remains a specific worldview. She cites Eugenia’s framing that “most people aren’t looking to save time; they’re looking for ways to invest it,” which helps explain the pull of Netflix, TikTok, YouTube, and social products. Josh thinks creation can bridge the gap—the person who never mastered Adobe Premiere can now express an idea—but the opportunity remains very early.
4. Subscriptions are temporary scaffolding for broader business models
Current monetization is historically unusual: about 85% of the web list uses subscriptions, another 62% sells credits or tokens, and only 13% offers ads or another non-payment model. ChatGPT has already entered the global top 20 consumer subscriptions in roughly three and a half years, but almost every other member is media or a large, frequently used platform such as Amazon or Uber.
Olivia’s “controversial” position is that she does not necessarily want the 4.5% paying share simply expanded. Most consumers neither need a business-building coding tool nor want to buy software; the industry instead must revive advertising and transaction fees. Josh says subscriptions became the early default because inference was costly: “If it’s going to cost me a lot to provide the service, at least let me charge money.”
That cost base reverses an old consumer-internet playbook. Companies once built density for five years because marginal service costs were near zero, then monetized liquidity; AI companies can face uncontrolled spending if they grow before understanding unit economics. Falling inference costs and lower-cost models should reopen familiar models, but until then even promising products may rationally suppress growth.
OpenAI illustrates the potential transition. The episode cites an annual advertising expenditure of $1 billion as of August and 1.2 billion weekly active users; Olivia argues this is “real density,” reached far faster than historical ad launches. ChatGPT Login and a possible future wallet could deepen conversion, while the vertical product OpenEvidence—said to reach 50%–60% of U.S. doctors—offers a smaller but exceptionally valuable audience.
5. Advertising wins only if conversational intent survives monetization
Josh finds commercial discovery increasingly natural as users ask ChatGPT for travel, purchases, and decisions rather than facts alone. Researching New York produced useful event ads; a separate conversation about a detachable valet-key ring eventually surfaced a suitable product from a Detroit factory he had never encountered. He received no ad in that exchange, but says he would have welcomed one that improved the answer.
Elena argues that live dialogue could eventually enable segmentation “even better than Meta,” because expressed intent arrives alongside accumulated knowledge of the user. Olivia says that knowledge could allow higher prices and conversion, but the same intimacy creates the failure mode: inserting a dubious weight-loss product into a personal health discussion “doesn’t build trust with AI.” Olivia praises OpenAI’s rollout because the ads she has seen are clearly labeled and fit the conversation rather than interrupting it.
Elena leaves an unresolved investor question: what are advertising unit economics when a query may involve expensive coding, image generation, or prolonged inference? The panel has no answer. Its narrower claim is that discovery-oriented queries provide attractive inventory, while the economics will depend partly on the computational burden required to produce the surrounding experience.
6. Model reach and product value are separating
ChatGPT remains dominant, running at roughly 6 times Claude’s web usage and 2 times Gemini’s, with about 3 times as many U.S. paid subscribers as either. Yet Claude has overtaken Gemini in U.S. paid subscribers despite Google’s installed-base advantage. Anthropic’s subscription intensity is different too: about 7.5% of subscribers use a maximum plan above $100 per month, versus roughly 1% for ChatGPT and Gemini.
Specialists retain openings where labs have not concentrated. ElevenLabs and Suno rank near the top in traffic and spending for voice and music; Olivia questions whether labs pursuing coding agents and AGI want Suno’s intellectual-property headaches. In images, Images 2.0 and Google’s Nano, Banana, and Bio versions have absorbed a large portion of high-end consumer traffic, pushing Midjourney out of traffic rankings—yet Midjourney reappears in revenue because power users still pay for “taste, aesthetic sense,” and precise control.
Video remains a “wild frontier,” with Olivia arguing that Chinese companies able to train with any data have an advantage in synchronized visual and audio generation. Meanwhile agents can operate creative tools directly: Josh let Astra and 5.5 compete head-to-head on animation after giving an agent permission to install what it needed. The product opportunity is to package those expert experiments into interfaces that are easy to use and can become products over the coming year.
This is Josh’s case for the software layer. Town connects Slack, email, and Google Drive, then creates manuals for who Olivia is and how she sounds; that context makes it more than “just a harness.” Beyond productivity, the open map includes dating, recruiting, social AI, shopping, home buying, gaming, and entertainment—multiplayer or marketplace categories where conversation alone is insufficient and where “lost time” may ultimately create the largest consumer companies.