At the 2026 Seed Awards stage on August 7, brands usually wait for gold cases. What CMO Zhiheng and tech VP Fengdi pushed harder was a product pack named NEXT: not “AI can do everything,” but a merchant funnel cut into three stages—see people, influence people, win people.

Net Economic Society’s write-up of the event is blunt: seeding is stretching from marketing into the operating chain, and AI is starting to sit in R&D and media decisions. The product question is not “Xiaohongshu also has AI.” It is how the platform turns community person–product signals into SOPs merchants can run in a back-office chat.

Mechanism 1: a three-stage funnel, not a pile of AI buttons

NEXT’s core is not one chat pane. It is re-ranking the expensive decisions in a merchant’s day by funnel stage.

In “see people,” AI insight lands in trend spotting, new-product development, reputation and category analysis, content strategy, long-horizon ROI diagnosis, churn cohorts, and more—plus five role agents: data analyst, media strategist, asset designer, creative planner, ad buyer. Tasks that once meant dashboards and a week of judgment start compressing into dialogue-generated reports.

That is a different SKU from “give merchants a general LLM.” A general model answers “what do young people like?” NEXT bets that answers must sit on Xiaohongshu notes, search, and comments, then feed the next content and media loop. The awards narrative stresses that lived experience stays hard to replace—AI makes insight finer and measurement sharper. The community is still the fuel; the model is the pipe.

“See people” also pulls R&D and assortment decisions back into the platform’s data domain. If merchants still open SKUs with external research plus gut feel, Xiaohongshu is only a media field; if assortment questions are asked inside NEXT, audience packs, pain words, and rival reputation flow naturally into ads and content systems. The platform wants more than model calls—it wants assortment decisions born at the top of its own funnel.

“Influence people” bridges content and ads: insight → inspiration → production → diagnosis, including smart K-selection and custom briefs, then media results to falsify the strategy. Three efficiency numbers shown on stage are very product-shaped: overall adoption of creative-platform generations 72%, launch-to-ads rate 57%, ad review pass rate 98%. The gap between adoption and spend still matters: generating is not the same as worth burning budget; a high pass rate means the platform is willing to let AI output into review, not only demos.

Read the three numbers together. 72% adoption means “usable”; 57% launch-to-ads means “willing to burn” still lags—the gap is risk appetite and test discipline on the budget owner. 98% pass rate lowers friction that AI creative never enters the ads system. For agencies, Brief and first-draft work is easiest to commoditize; for brands, time saved must be reinvested in what to test and what to kill, or adoption only piles mediocre assets.

“Win people” moves closer to money: app promotion, lead ops, off-platform trade sync, on-platform closed-loop commerce. DM upgrades from passive replies to industry-configurable paths; the public line is roughly 1 extra lead per 5 users talked to. A component ecosystem lets maps, floor plans, reports, and similar services enter DM. Off-platform, combo campaigns with ~46% new-customer GMV share are used to argue seeding–conversion sync; on-platform, the story shifts toward long-term user value and less obsession with same-day ROI.

“1 lead per 5 chats” is an efficiency line for scripting: it does not prove AOV or close rate—only that configured DM paths can lift lead density. Industry components (maps, floor plans) keep high-intent actions inside the conversation instead of dumping users to external forms. The ~46% new-customer GMV share in combo campaigns argues seeding can shape new-customer mix—if brands stretch attribution beyond same-day ROI.

Together, NEXT sells layered interfaces of an operating system: insight outputs conclusions, content/media outputs assets and tests, conversion outputs leads and repurchase metrics. Merchants need not understand models—only how to ask along the funnel.

Mechanism 2: lower the ops bar to deepen occupancy, not sell another ad slot

NEXT’s commercial logic is not “one more impression product.”

Classic seeding monetization runs exposure → interest → on/off-platform conversion. Merchants’ dearest cost is often people—strategy, creative, media, DM ops. NEXT uses role agents to chew those labor steps. The pricing narrative is closer to productizing agency work: fewer hires or fewer agency rounds, in exchange for longer back-office dwell, denser data write-back, and a more measurable media loop.

That also explains launching the pack at the awards, not only burying features in the ads console. The awards are a brand classroom: three-year totals above 1,500 cases and 1,000+ brands (public figures) mean a cohort already fluent in seeding language. NEXT compresses that methodology from award stories into clickable paths.

For the platform P&L, deeper occupancy means audience packs in insight reports flow back to ad accounts; generated content flows into spend; DM leads hang on lead and commerce components. Token/compute cost hides inside “a week becomes an hour.” Merchants feel throughput; the platform feels operating data spinning faster inside its own funnel.

The counter-intuitive bit: many platforms make AI “write notes for you.” NEXT keeps note-writing in the middle of “influence,” and adds stages before and after—see people (R&D and assortment) and win people (leads and LTV). The sell is not copy speed; it is decision throughput. Agencies that only sell copy and media buying feel commoditization; those that own industry components, closed-loop conversion, and hard-to-standardize judgment can sit as a value layer on top of NEXT.

Another tradeoff: the fuller the funnel, the deeper the path dependence—switching platforms means retraining “see people” questions and data habits. Moat language for Xiaohongshu; a governance question for multi-platform brands deciding whether assortment brains live here too.

Shipping also needs a clear line on who owns a wrong report. If dialogue insight errs, does the brand blame the model or the community sample that trained the signal? If media scales budget off a report and conversion fails, is NEXT a decision tool or a draft opinion? The public pack shows efficiency numbers and scene lists, not liability boundaries or a default human-review path. Treat NEXT as an auto-decider instead of a draft accelerator, and 72% adoption can become 72% blind follow.

Set against a possible user-facing dialogue shopping guide (reported elsewhere; not officially confirmed), “win people” could appear on both merchant and user sides. The platform would then need to stop two systems from double-tapping the same intent—and stop creators from feeling notes only stock a guide shelf. What this piece can nail now: Xiaohongshu’s AI commerce story has already moved from “help write notes” to an ops-funnel operating system.

Who should do what

Role	Math to run	Don’t mythologize
Brand / merchant ops	Which stage lacks headcount—insight, media, or DM; can NEXT replace weekly agency decks; who owns test/kill discipline after adoption	72% adoption = growth
Agencies	Which deliverables get commoditized by the five AI roles; differentiation via industry components and closed-loop conversion	Platform AI kills the category overnight
Community-commerce tool builders	A three-stage funnel is a clearer SKU than a chat plugin; leads and on/off-platform sync are the paid points	Only racing image/title generation

Seeding remains community trust capital. NEXT’s job is plugging that capital into a billable operating line. The funnel is clear; next is whether merchants really hand R&D and media calls to the reports in the chat.

Sources

• Xiaohongshu Seed Awards: seeding extends into ops; AI joins R&D and media (100EC)
• Xiaohongshu’s second growth engine (36Kr, context)