TL;DR
- "AI traffic" is not one channel. Different AI surfaces are emerging with different commission structures, adoption curves, and per-user transaction patterns.
- A recent commission change in China illustrates the pattern clearly. One platform now charges a meaningfully higher commission for orders that start in an AI conversation than for orders that start in a content feed, on the same settlement system.
- The growth-planning implication is to budget per surface, not per category. A team that budgets for "AI" as a single line item will misread where its actual return is coming from.
Once a team accepts that AI-mediated shopping is real and growing, the next question is usually a budgeting one: how much attention, and how much cost, does this channel deserve? That question gets harder to answer honestly once you notice that "AI shopping" isn't a single channel at all — it's a set of surfaces, each with its own economics, and they're already starting to diverge in ways a single line item can't capture.
What recent data shows about diverging AI-channel economics?
In China, one of the largest AI-native apps began charging a higher commission — reported as roughly 12% versus 8% — on local-service orders that start inside its AI-conversation entry point compared with orders that start inside its content-feed entry point, on the same merchant and the same settlement system, according to 钛媒体's reporting from mid-August 2026. The stated rationale, per that reporting, is that a shopper's intent expressed directly to an AI conversation carries more information density than a feed-driven browse-and-guess interaction — and the platform priced that difference into its take rate. On the very same day, according to 晚点LatePost, a competing AI platform launched the opposite strategy: a free, open-access model aimed at drawing early enterprise adopters. Two platforms, same week, two opposite bets on how to monetize AI-mediated demand.
Why does one company's pricing decision matter outside its own market?
No Shopify brand needs to operate in that specific market to learn from the pattern. The lesson isn't the exact percentage — it's that once an AI surface can observe a shopper's actual intent (not just infer it from browsing behavior), that surface has a reason to price access to that intent differently than it prices access to passive attention. As more Western AI shopping surfaces mature past their early, mostly-free adoption phase, the same logic is likely to show up: conversation-native discovery may simply carry a different cost structure than feed- or search-native discovery, once platforms start monetizing it deliberately.
How fragmented does the underlying landscape already look?
QuestMobile's June 2026 data on China's AI-native app market, cited by 晚点LatePost, shows just how unevenly adoption is spreading even within one country's ecosystem: one app reached 382 million monthly active users (+172.1% year-over-year), a newer entrant reached 167 million (+5,792.9%), a third app's monthly active users fell 20.3% to 130 million, and a fourth reached about 50 million (+100.9%). No single platform holds a stable, dominant share — the ranking is still actively reshuffling month to month. A brand that built its entire AI-discoverability plan around today's leading platform would be planning around a landscape that could look different by the time the plan is executed.
What does this mean for how a growth team should track spend and return?
| Assumption | Why it breaks down | Better approach |
|---|---|---|
| "AI traffic" is one line item | Different AI surfaces have different intent signals, cost structures, and audiences | Track access, retrieval, and referral per surface, not as one combined number |
| The current top platform will stay on top | Adoption data shows month-to-month reshuffling even among well-funded apps | Revisit which surfaces matter on a recurring cadence, not once a year |
| Free access today means free access later | At least one major platform already moved from flat access to intent-based pricing | Build a plan that still works if a currently-free surface starts charging for qualified access |
| More AI traffic automatically means more efficient growth | A large user base doesn't guarantee meaningful per-user transaction volume | Judge a surface by realized outcomes, not by its headline user count |
Why does a large user base not guarantee meaningful commerce volume?
The China data offers a useful caution here too: one of the platforms with the largest user base (over 200 million daily active users, per reporting) was estimated to generate only about ¥10 million in daily ecommerce transaction volume — a per-user daily transaction figure of well under one yuan. Scale in usage and scale in commerce outcomes are not the same measurement, and a growth team that equates "big AI platform" with "big AI revenue opportunity" risks over-investing in reach that hasn't yet converted into transactions worth chasing.
How should a Shopify growth team actually plan around this?
- List the AI surfaces that plausibly send you traffic today, even informally — general-purpose assistants, platform-native shopping features, and any vertical-specific AI tools in your category.
- Track each surface separately using the staged model — access, indexing, retrieval, referral, attribution — described in last week's piece on measuring AI-mediated shopping, rather than folding every surface into one "AI" bucket.
- Revisit the list on a quarterly cadence. Given how quickly adoption rankings shifted in the data above, a surface that's negligible today may not stay that way, and vice versa.
- Build the plan to survive a pricing change. If a currently-free or low-cost AI surface begins charging for qualified intent, the plan shouldn't depend on that cost staying at zero indefinitely.
Where does Agentic Page fit?
Agentic Page's role in this picture is upstream of pricing and platform selection: it keeps a merchant's product information structured and current so that whichever AI surfaces a growth team ultimately decides to prioritize, the underlying product data is ready to be read accurately by any of them. It does not set or influence what any AI platform charges for access to shopper intent, and it does not pick which surfaces will matter most for a given brand — that remains a judgment call for the merchant's own growth planning, informed by the kind of staged tracking described above.
Related reading: How Merchants Should Measure AI-Mediated Shopping · How AI Answers Are Becoming a New Discovery Channel · How Agentic Page Creates an AI-Readable Discovery Layer
Curious which AI surfaces are already sending you traffic, even informally? Learn more about Agentic Page or book a demo to see what's observable in your own store's data.