AI assistants finally started talking about money. This summer, China's leading assistants put pricing on the table — Qwen began testing membership tiers, while Doubao went both directions: a paid Pro tier pulling directly from users' wallets, and a move to take commissions from local-life orders such as hotel bookings.
For years the big-platform playbook was uniform: win user scale first, figure out profit later. But GPUs do not honor long-termism. Under the weight of enormous infrastructure spend, even OpenAI — historically cautious about advertising — is now testing ads in its free and low-priced tiers. If the exemplar bends, there is no reason for Chinese giants to keep burning cash without monetizing.
Three Business Models, Three Walls
Consumer internet products monetize in roughly three ways: subscription (users pay for stronger capability), advertising (merchants pay for exposure), and commission (platforms take a cut after a deal closes). Subscriptions sell capability, ads sell placement, commissions sell outcomes. Applied to AI, each path collides with a wall.
Subscription tests the model itself. Even ChatGPT long kept more than 90% of consumers on the free tier before OpenAI pivoted hard to the B-side. In China the pattern repeats: Doubao, with 178 million daily active users, reportedly counts only a few hundred thousand paying subscribers. Ordinary users only renew when free quotas run out and the task is valuable enough to justify it.
Advertising poisons the product. In the search era, a keyword surfaced dozens of links, so a few ads left plenty to compare. An AI assistant may return just three recommendations — if one is a paid placement, user trust erodes. Perplexity has paused its sponsored-content push; ChatGPT deliberately separates commercial content from answers and insists ads do not influence results.
Commission hit a wall in under six months. OpenAI launched Instant Checkout with Etsy and Shopify in September 2025, letting users pay inside the chat and charging merchants a fee (4% for Shopify by early 2026). By March 2026 it de-prioritized in-chat checkout, calling the model "not flexible enough." Walmart's in-chat checkout converted at roughly one-third the rate of redirecting shoppers to its own site.
The One Thing Worth Most Is the One Thing You Cannot Sell
Here is the awkward position of the AI assistant: the most valuable asset is user trust in the answer — and that is precisely what advertising would destroy. The closer an ad sits to an answer, the more it poisons the trust that makes AI recommendations worth anything.
Bytedance clearly understands this. When news of hotel commissions spread, Doubao quickly clarified that its local-life business runs no paid promotion, that merchants cannot pay to influence recommendations or ranking, and that it charges only a channel fee after an order closes. The company was rushing to disavow the "AI recommendation equals paid ranking" association. Ads sell placement; commissions do not touch natural recommendations and charge only after a transaction. The interests differ — and in that chain, user experience and platform revenue are partly aligned.
Yet this is also why provenance and trust engineering matter more than ever. As AI increasingly curates and vouches for what people buy and read, verifiable content provenance becomes the foundation of trustworthy recommendations — a shift we examined in Claude's invisible watermark rollout.
Middlemen's Middlemen: The Ecosystem Wall
General AI assistants are not replacing the internet; they are being inserted as a new channel layer on top of existing transaction chains. The deeper problem is the incumbent web itself. China's super-apps sit behind ecosystem walls built by Alibaba, Bytedance and Tencent. The same hotel carries different prices, room types, membership perks and refund rules across Ctrip, Meituan, Feishu and Douyin. If an assistant cannot reach real-time supply or reviews from other platforms, its recommendations are bounded by what it can access — regardless of how neutral its algorithm tries to be.
Doubao's hotel channel fee — roughly 12% (11.4% software service plus 0.6% payment fee) — marks the first time a general assistant has independently priced a batch of AI-understood, AI-screened consumer intent. For Doubao, commissions can offset the massive compute cost of high traffic; for Bytedance's ecosystem, it feeds Douyin's local-life business new users and deal volume. Meanwhile Meituan's agent and JD's agents are wiring into Tencent and WeChat ecosystems. The intermediary now has an intermediary, and merchants pay real money for each new growth channel.
This is the machine-traffic economy in miniature: when assistants transact on behalf of users, the web's traffic is increasingly decided by agents rather than people — a dynamic we unpacked in the moment bots outnumbered humans online.
From "Helping You Choose" to "Buying for You"
Moving from "helping you choose" to "buying for you" demands more than a checkout button. Commission models require real-time supply data: whether a hotel room is free tonight, weekend surcharges, breakfast inclusions, cancellation windows. Hotel decisions lean heavily on guest reviews — and review data is the hidden moat of incumbent OTA platforms, not easily scraped by an outside assistant.
There is also a responsibility problem. When an assistant only chats, it can tell users to verify facts themselves. Once it closes deals and splits commissions, that verification burden cannot keep landing on the user. Perplexity's shopping assistant once showed a completed checkout for a tube of Walmart toothpaste, billed the user, then emailed hours later that the item was out of stock. Walmart's own Sparky assistant drew complaints about slow responses and weak answers.
The lesson for every team building AI commerce: recommendation quality and accountability must be designed into the business model from day one. The old internet lesson — take responsibility before the model is set — applies here with full force. If AI assistants are to become the next platform entrance, they will have to make money the hard way: by being genuinely worth trusting.
