Summary

ChatGPT remains the distribution winner, while Claude and Gemini are building differentiated use cases rather than merely trading share. ChatGPT is 2.7× Gemini on web, 2.5× on mobile, nearly 30× Claude on web, and almost 80× on mobile. Yet Claude’s prosumer focus and Gemini’s creative releases show that the market is expanding beyond a single “AI for everyone.”
Context could become one of consumer AI’s strongest moats. Anish Acharya noted ChatGPT’s 900 million sign-ups and the potential advantage of users bringing inference capacity with them. Group chats, developer prioritization, and an authentication layer Sam Altman hinted at could let users carry memory and tokens into third-party apps. The unresolved risk is persona separation: users may not want to mix identity and memory between work and personal life.
The monetization split is sharpening between Claude’s subscription-led model and ChatGPT’s Google-like consumer funnel. Claude’s ecosystem favors premium research, scientific, and financial tools; ChatGPT favors travel, nutrition, marketplaces, and consumer finance. Olivia Moore expects the latter could eventually monetize through ads and transaction cuts, a bull case “that isn’t yet showing up in the data.”
AI adoption is globally uneven, with access restrictions, workforce composition, and cultural trust shaping usage alongside model availability. Singapore ranks first per capita, followed by Hong Kong, the UAE, and South Korea; the U.S. is number 20, while Russia and China sit below 50. U.S. trust in AI was cited at roughly 32%, versus 50–70% in several leading markets.
Creative AI is consolidating around differentiated workflows, while AI-only social feeds have not yet shown comparable success. Commodity image generation is moving into ChatGPT and Gemini, leaving opinionated products such as Midjourney and Ideogram, while Suno and ElevenLabs have sustained top-20 or top-15 positions. Sora reached one million users faster than ChatGPT, but its exported content competed on other platforms against the best human work; “the emotional stakes” in an all-AI feed felt lower.
Agents have crossed the technical threshold, but distribution may determine who captures the horizontal market. Moore said OpenClaw would have debuted at number 30 on the web list, thought it had passed React and Linux in GitHub stars, and said it was acquired by OpenAI. New-user growth plateaued before it “fully escaped containment.” Manus reportedly ramped to $100–200 million ARR within 6–9 months, but its Meta acquisition illustrates why horizontal agents may benefit more from large-platform distribution.
Mainstream adoption will trail technical capability, with voice and memory likely serving as the bridge. Moore expects voice interfaces to spread to consumers within 6–9 months and argues that every AI company, and eventually every technology company, will become “an agented company.” Within a couple of years, a product that does not immediately know the user may “feel broken,” making onboarding itself obsolete.

Deep dive

1. Distribution is separating the general-purpose models

Moore’s starting point: six editions and three years of incredible growth still leave consumer AI “so early.” ChatGPT is the largest global AI product by far, yet only about 10% of the world uses it weekly; meanwhile, Notion says AI-first features may drive half its new ARR.
The headline dominance is stark: ChatGPT is 2.7× Gemini on web and 2.5× on mobile, nearly 30× Claude on web, and almost 80× on mobile. But Moore sees expansion, not simple bifurcation, as people assign different products to different jobs.
Claude is concentrating on prosumers through Claude Code, Cowork, Excel, PowerPoint, and premium research or financial data. ChatGPT is building for travel, nutrition, marketplaces, and consumer finance. Their app directories each have 200-plus apps, yet overlap by only 11%—a revealing measure of strategic divergence.
Gemini occupies a distinct creative corner: active and paying users are nearly perfectly correlated with releases of Veo 3, Nano Banana 1, Nano Banana Pro, and Nano Banana 2. Moore says NotebookLM’s greenfield status has enabled faster progress, while existing surfaces such as Sheets and Docs carry more inertia and management overhead.
Moore’s bull case for ChatGPT resembles Google: acquire broadly, convert some users to subscriptions, then potentially monetize the rest through ads and transaction cuts. Claude can remain subscription-led, but ChatGPT wants to be the “AI for everyone,” with monetization following distribution.

2. Memory and geography are creating new forms of segmentation

Context may stop being easily portable. Moore sees lock-in emerging through ChatGPT group chats, developers shipping first to the largest audience, and an authentication layer that could let users bring “your memory and your tokens” into third-party products.
Anish Acharya’s extension of the thesis: with 900 million sign-ups, ChatGPT could supply personalization and inference capacity, sparing developers that cost while deepening platform lock-in. Moore’s caveat is identity collision—enterprise adoption helps familiarity, but users may resist mixing professional and personal memory. Acharya joked, “Don’t cross the streams.”
Geography supplies a second source of divergence. China’s combined ChatGPT and Gemini usage is only 15%, supporting Doubao, DeepSeek, Yuanbao, and Kimi; Russia has a parallel ecosystem around GigaChat, Yandex, and DeepSeek, whose second-largest market is Russia. These are the report’s two “huge outliers.”

3. Adoption follows trust, while creative AI rewards differentiation

Per-capita adoption puts Singapore first, then Hong Kong, the UAE, and South Korea; the U.S. ranks number 20, with Russia and China below 50. Moore connects the leaders to tech-heavy, white-collar workforces, while large U.S. employment categories such as retail and transportation remain comparatively untouched.
Cultural confidence matters too. Moore cited U.S. trust in AI at roughly 32%, versus 50–70% across several higher-adoption countries; U.S. monthly usage is around one-third, while some smaller European or Eastern European markets reach 40–60%. India remains an opportunity because many local languages are still poorly served.
Early creative tools benefited from hallucination because surprise and imperfection could look “beautiful or original.” Commodity memes and marketing images can now be handled by ChatGPT or Gemini, leaving standalone image products such as Midjourney and Ideogram to compete through opinionated aesthetics or workflows unavailable in a prompt box.
Music and voice remain more defensible: Suno and ElevenLabs rose into roughly the top 20 or top 15 and held position. Video looks less likely to produce “one model to rule them all”; Moore says Chinese models are very strong because they can train on any data, with Seedance 2.0, in her view, head and shoulders above U.S. models so far. That favors aggregators such as Krea that let users switch models.

4. Sora and agents show where capability outruns product behavior

Sora’s launch was enormous: number one in the U.S. App Store for 20 consecutive days and one million users faster than ChatGPT. Acharya estimated that reaching number one now probably requires about 150,000 daily downloads. Sensor Tower cited 3 million dials, although monthly downloads fell from roughly 6 million in November to 1.5 million.
Cameos made likeness-sharing and friend-generated memes unusually compelling, but exportability weakened Sora as a destination. On TikTok, Reels, or YouTube, its best clips competed alongside the best human work; inside an all-AI feed, “the emotional stakes” felt lower. Licensed fan videos involving beloved characters and entertainment figures might offer a narrower moat.
OpenClaw missed the ranking because February fell outside the data window, but would have entered the web list at number 30. Moore said she thought it was number one in all-time GitHub stars, passing React and Linux, while sign-up traffic flattened—evidence that it remains exceptional for technical users but has not “escaped containment.” OpenAI acquired it, and Moore hopes it will productize OpenClaw for mainstream consumers.
Acharya’s pushback—worth keeping: OpenClaw’s cross-model operation may be central to its value, making ownership by a single model provider potentially constraining. Moore agreed keeping it model-flexible would be smart, while expecting productized and vertical “OpenClaw for this” variants to proliferate.

5. Horizontal agents need distribution; consumer adoption needs invisibility

Manus was Moore’s consumer-agent breakthrough: reliable autonomous work across email, browsing, slides, and spreadsheets. She cited a reported ramp from $0 to $100–200 million ARR in 6–9 months; Meta acquired it for more than $2 billion during the period covered by the list. Its engineering was “3 to 6 months ahead,” but horizontal capability eventually invites competition from platforms already holding enterprise contracts.
The report’s web-centric methodology is increasingly incomplete. Cursor, Granola, voice dictation, and Claude Cowork live primarily in dedicated desktop apps, where they can interact with files and remain ambient; revenue could therefore become a useful companion ranking. Even browsers face inertia: Comet’s peak download-page traffic was five times Atlas’s, yet neither has found the mainstream “killer” feature.
Teenagers reveal the likely endpoint: more than half admit using AI for homework, 38% for creative work, 16% for casual conversation, and 12% for emotional support. Moore expects agents to disappear as a category—every AI company, and eventually every tech company, becomes “an agented company”—while voice spreads within 6–9 months and memory makes traditional onboarding feel broken within a couple of years.

Everything You Need to Know About Oil A tour through the media landscape with TSOH's Alex Morris