CHAPTERS
- 0:00 – 1:20
What the Consumer AI Top 100 measures (and how it’s built)
Justine and Olivia lay out the purpose of the Consumer AI Top 100: understanding what real consumers actually use beyond headline apps. Olivia explains the methodology, data sources, and why the list focuses on usage rather than revenue.
- •Fifth edition of the list, published every six months since early consumer gen-AI days
- •Universe: all websites/apps worldwide, ranked by usage, then filtered to AI-native products
- •Web metric: monthly global visits (Similarweb); mobile metric: monthly active users (SensorTower)
- •Top 50 web + top 50 mobile; captures free + paid usage to reflect consumer interest
- 1:20 – 3:29
New entrants, ecosystem stabilization, and why web is the better change signal
They discuss what’s new this cycle and why the mobile list tends to be more volatile than the web list. A key theme is that churn among top products is decreasing, suggesting the consumer AI landscape is starting to stabilize.
- •Mobile rankings are less stable due to app store policy shifts and enforcement
- •Earlier mobile lists were dominated by ChatGPT-copycat apps (e.g., “ChatGTP” variants)
- •This edition: 11 new names on the web list—down from 17 six months prior
- •Stabilization implies fewer sudden one-month spikes and more durable winners
- 3:29 – 3:57
Companionship keeps dominating—and expands with new names
Companionship remains one of the strongest and most persistent consumer AI categories. They highlight multiple new companionship apps that broke into the list alongside recurring incumbents.
- •Companionship remains a major share of top consumer AI usage
- •New companionship entrants: Juicy Chat, Joy, and R Dream
- •Returning/major names: Character.ai (largest), Janitor, Spicy Chat, Polybuzz, Crushon, Adot, Candy AI
- •Companionship’s staying power makes it hard for other categories to displace top spots
- 3:57 – 4:41
Big Tech arrives: Google’s breakout moment on the web list
Google’s domain structure previously made it difficult to measure individual product traffic, but this cycle multiple distinct Google properties qualify. They break down which Google products ranked and what that implies about consumer and developer adoption.
- •First time four separate Google properties could be independently ranked and included
- •Gemini ranks #2 on web, behind ChatGPT; ~10% of ChatGPT web traffic
- •Gemini is much closer on mobile (~half of ChatGPT), likely driven by Android distribution
- •Google’s presence signals Big Tech can compete not just with models, but with distribution and product packaging
- 4:41 – 6:24
NotebookLM, AI Studio, and Google Labs: sandboxes go mainstream
They dig into Google’s surprising entries beyond Gemini, including developer and consumer sandboxes. NotebookLM’s sustained traffic defies expectations that virality would fade quickly, while Google Labs appears boosted by new video capabilities.
- •AI Studio (developer sandbox) lands in the top 10—strong developer interest
- •NotebookLM ranks #13 and sustains mostly flat/increasing traffic since 2024 virality
- •Google Labs (#39) acts as a consumer-facing sandbox; likely driven by Veo 3 usage
- •Veo 3 release correlates with a ~15% traffic spike to Google Labs; other Labs tools include Doppel, Portrait, Wisk, Project Mariner
- 6:24 – 7:52
Chinese AI products for China: domestic assistants climb the rankings
They describe how China’s regulatory environment shapes a parallel consumer AI ecosystem. Several China-focused assistants rank highly due to population scale and the absence of U.S. assistants inside China.
- •ChatGPT/Claude-style products are restricted in China; local alternatives fill the gap
- •Top China-focused products: Quark (Alibaba), Doubao (ByteDance), Kimi (Moonshot AI)
- •These assistants rank in the top 20 on web, driven primarily by usage within China
- •High usage can come from single-country adoption when the market is massive and competition is limited
- 7:52 – 9:39
Chinese companies exporting AI: global apps, video models, and distribution channels
They outline a second pattern: China-developed tools aimed at global users, especially in image/video. They also note that some Chinese model usage is ‘hidden’ inside U.S. aggregators and developer platforms, making it hard to fully measure.
- •Globally oriented Chinese products include DeepSeek, Hailuo, Kling, SeaArt
- •Particular strength appears in image/video (e.g., single-frame animation capabilities)
- •Distribution strategy: Chinese models accessed via U.S. apps/aggregators (e.g., subscriptions bundling multiple models)
- •Developer channels like Replicate/Fal can obscure end-model origin in consumer usage data
- 9:39 – 10:18
China-built products used both domestically and abroad: the Manus example
A third China-related trend is products that find traction both inside China and internationally. Manus illustrates how AI adoption hotspots and international usage patterns can be surprising and not purely driven by paid users.
- •Manus is used in multiple geographies and reports strong revenue scale (90M annualized run rate cited)
- •Traffic split example: #1 Brazil, #2 U.S., #3 China—Brazil as a standout AI-traffic source
- •Free usage can drive significant traffic, influencing rankings
- •Signals maturation of a domestic China ecosystem plus ongoing export success
- 10:18 – 12:08
Vibecoding takes off: traffic, revenue, and retention surprises
They examine vibecoding’s rise from a niche trend into multiple top-list placements. Beyond traffic, they discuss revenue retention data that suggests these products may be unusually sticky and potentially crossing into enterprise-like behavior.
- •Shift from prior list: Bolt moves to “brink,” while Lovable and Replit enter the main web list
- •Lovable’s rapid revenue growth is paired with cohort-based revenue retention analysis (Consumer Edge)
- •Many vibecoding tools show ~100%+ revenue retention in months 1–3—rare for consumer/prosumer
- •Strong retention hints at real ongoing value: prototyping at work or sustained solo builder usage
- 12:08 – 13:35
Where are vibecoding’s end-users? Builder traffic vs. project traffic
They use domain-level measurement to compare builder activity to consumption of created projects. The imbalance suggests either successful projects move off-platform to custom domains, or many creations are personal/internal tools with limited public traffic.
- •Ability to measure traffic separately: platform (builders) vs hosted outputs (created apps/sites)
- •Observed pattern: builder traffic is much higher than traffic to what’s built
- •Hypothesis 1: high-traffic projects shift to custom domains, disappearing from platform-domain counts
- •Hypothesis 2: many creations are personal software—valuable to the creator but not broadly visited
- 13:35 – 14:32
AI All-Stars: the products that stayed on every list (and what it implies)
To understand durability, Olivia introduces “AI All-Stars”—companies that made every web list across five editions. The set reveals that lasting consumer attention is not limited to companies with massive proprietary model spend.
- •14 companies have appeared on every web list across five editions
- •General assistants: ChatGPT, Perplexity, Poe; companionship: Character.ai
- •Creative tools: Midjourney, PhotoRoom, Leonardo, Cutout Pro, Veed, ElevenLabs
- •Productivity + hosting: QuillBot, Gamma, Hugging Face, CivitAI
- 14:32 – 16:34
UI, workflow, and network effects beyond ‘better models’
They argue that consumer AI defensibility increasingly comes from product experience, community, and content ecosystems—not just proprietary models. Examples include community-driven model libraries and marketplaces that create compounding value.
- •More than half of all-stars rely on hosting/aggregating others’ models—UI matters as much as models
- •Workflow depth can differentiate even when underlying models are commoditized via APIs/open source
- •Non-data network effects: Hugging Face/CivitAI communities (models, LoRAs, datasets, rankings, mini-apps)
- •Marketplace-style effects also appear in ElevenLabs voice libraries; classic data flywheel still exists for ChatGPT/Midjourney
- 16:34 – 18:30
From consumer to enterprise: bottoms-up adoption and ‘prosumer’ expansion
They describe how successful consumer AI products are graduating into team and enterprise usage through self-serve, bottoms-up adoption. Templates, shared libraries, and team plans increase switching costs and help products spread inside organizations.
- •Prosumer/enterprise features: team plans, templating, design libraries (Gamma, PhotoRoom, ElevenLabs)
- •Adoption path: individual tries → proves value → pulls into team → triggers enterprise contract
- •Bottoms-up motion can outperform traditional top-down enterprise sales
- •Enterprise usage can feed back into consumer growth as tools spread across coworkers and into home projects
- 18:30 – 25:14
Biggest takeaways and predictions: verticalization, Grok’s debut, and what’s next
Looking across editions, they note early chaos has given way to clearer category clusters and more persistent winners. They predict more ‘verticalized’ usage across assistants, growth in reliability-focused productivity tools, and emerging categories like health, edtech, finance, and AI-native social.
- •Early lists were volatile; now fewer newcomers and clearer recurring “all-stars”
- •Users increasingly choose different assistants for different tasks (verticalization within general tools)
- •Grok debuts strongly (#4 web) and Meta AI begins showing up on web; competition heats up beyond ChatGPT
- •Next waves: prosumer productivity tools (spreadsheets, decks, email), plus potential breakouts in health, edtech, personal finance, and AI-native social
- 25:14 – 26:02
Closing: explore the full list and share what’s missing
They wrap by encouraging listeners to view the full published list and discuss favorites. They explicitly solicit audience input on products that should have ranked but didn’t, setting up curiosity for the next edition.
- •Call to action: click through to see the complete Consumer AI Top 100 report
- •Prompt for comments: favorite products that made the list
- •Prompt for debate: surprising omissions and daily-use tools not represented
- •Expectation-setting: next list arrives in six months
