Why China's AI Models Won't Just Recommend Infant Formula — And What That Means for Foreign Maternity Brands in 2026

China's infant formula and maternity market is worth roughly $110 billion by 2026 estimates, with China alone accounting for close to 40% of global infant formula demand. It is also one of the only consumer categories where China's AI search models — Doubao, DeepSeek, Kimi, Qwen — visibly hedge. Ask Doubao for the best running shoes or a hotel recommendation and it answers directly, often with a shortlist. Ask it which infant formula to buy for a newborn, and the tone shifts: caveats about consulting a pediatrician, references to regulatory approval, and citations to parenting forums instead of a confident brand pick.

That hedging is not a bug in the model. It is a signal about how GEO works differently in safety-sensitive categories — and it's a signal foreign maternity and infant-care brands have largely ignored while they were busy building their Xiaohongshu and Tmall storefronts.

A category where AI models default to caution, not brands

China's AI ecosystem has a long institutional memory of the 2008 melamine contamination scandal, which permanently reshaped how Chinese consumers — and the platforms serving them — treat infant nutrition. That history shows up in model behavior today. Where DeepSeek and Qwen answer a luxury handbag query by citing SMZDM, Zhihu, and Xiaohongshu influencer content, the same models handling an infant formula query lean harder on sources with institutional authority: hospital-affiliated parenting content (丁香媽媽 and similar), regulatory disclosures from China's State Administration for Market Regulation, and long-running parenting forum threads rather than single-post recommendations.

This matters enormously for GEO strategy because it inverts the usual playbook. In most categories, brand-controlled content — product pages, KOL placements, short-video demos — is the fastest way into an AI model's citation graph. In infant and maternity nutrition, brand-controlled content is exactly what these models are trained (and increasingly fine-tuned) to treat with skepticism. A glowing product description from an official Tmall flagship carries less citation weight than a pediatrician's Q&A thread that happens to mention the same product neutrally.

The domestic content flywheel foreign brands are missing

Doubao's dominance in China's AI market — ByteDance reports roughly 79% user penetration with a Net Promoter Score of 54.4, well ahead of Tencent's Yuanbao (42% penetration, NPS 28.3) and DeepSeek (37% penetration, NPS 36.1) — is inseparable from Douyin's short-video ecosystem. Domestic infant formula brands like Feihe (飞鹤) and Yili have spent years building exactly the kind of high-volume, algorithmically favored short-video content (breastfeeding-transition tutorials, pediatrician cameos, factory-transparency videos) that Doubao's retrieval layer is tuned to surface.

Foreign brands selling into China through cross-border e-commerce — Aptamil, Wyeth Illuma, a2 Milk, Bubs, Else Nutrition — largely rely on a different content ecosystem: import-compliance documentation, daigou (代购) reseller networks, and expat parenting communities. That content is real and often trusted by actual buyers, but it is structurally thin in the exact places China's AI models are now pulling citations from. The result is a visibility gap that has nothing to do with product quality or even brand reputation among consumers — it's a gap in whether the brand shows up in the sources these models already treat as authoritative for this category.

How the platforms diverge on baby-category queries

This divergence means a single "China GEO strategy" document is close to useless for a maternity brand. The same product needs different evidence assets depending on which model a parent happens to be using — and unlike categories like fast fashion or electronics, parents in this category are unusually likely to cross-check an AI answer against a second or third source before buying, which raises the stakes of not being cited at all.

A longer research cycle actually favors brands that show up correctly

Most GEO advice optimizes for speed: get cited fast, get cited often, ride the algorithmic wave while it lasts. Infant and maternity nutrition breaks that assumption in a way that should be reassuring to foreign brands willing to do the slower work. New parents in China do not make a single AI query and buy — the category's purchase cycle typically spans multiple research sessions across weeks, often starting during pregnancy and continuing through a child's first two years as formula stages change. That means a brand doesn't need to win a single high-traffic query moment the way a fast-fashion brand needs to win a flash-sale search spike. It needs to be present, consistently and neutrally, across the dozens of smaller queries a parent runs over that research window: ingredient comparisons, stage-transition questions, allergy and sensitivity concerns, import authenticity checks.

This is actually good news for brands that get the authority-content strategy right, because it means the visibility gap described above is a solvable, compounding problem rather than a one-time ad-spend arms race. A brand that seeds five or six pieces of genuinely neutral, well-sourced Chinese-language content into the right forums and compliance channels this year will likely still be showing up in model citations eighteen months from now, because that content keeps getting re-crawled and re-cited as parents keep asking variations of the same questions. Compare that to a sponsored KOL placement, which has a shelf life measured in weeks.

What "authority" content actually looks like in practice

The brands making progress here are not the ones buying more Xiaohongshu placements — they're the ones changing what kind of content they produce. That typically means three shifts. First, publishing full import and safety-certification documentation in Chinese, not just English compliance PDFs buried on a corporate site, and getting that documentation picked up by the trade and consumer press that Baidu ERNIE and DeepSeek already index as authoritative. Second, working with — rather than around — the hospital-affiliated and pediatrician-run content channels that dominate citation sources for this category, since a single mention in a well-trafficked 丁香醫生 or 丁香媽媽 thread carries more downstream citation weight than a dozen brand-run posts. Third, treating cross-border authenticity as a content topic in its own right: parents researching foreign formula brands are disproportionately worried about counterfeit product moving through daigou channels, and AI models increasingly surface authenticity-verification content as part of the answer — brands that publish clear, searchable guidance on how to verify a genuine purchase are answering a question the model is already trying to help with.

Takeaway for Brand Marketers

If your brand sells infant formula, baby food, or maternity products into China, treat AI visibility in this category as an authority problem, not a content-volume problem. Sponsored placements and flagship-store copy will not move the needle the way they do in beauty or sportswear GEO campaigns, because the models are actively discounting brand-originated content for exactly this category. Instead, prioritize three things: getting your import registration and safety certification status documented in places AI models already index as authoritative (regulatory filings, compliance news coverage); seeding neutral, comparison-style presence in long-form parenting communities that DeepSeek and Qwen already cite, rather than promotional posts; and treating Doubao separately with a genuine short-video content investment if that platform's consumer reach matters to your China strategy, since its retrieval behavior rewards a completely different content format than the other three models. Brands that keep funneling budget into the same Tmall-and-KOL playbook that works for other categories will keep losing citation share to domestic competitors who built their forum and short-video presence years ago — not because their product is worse, but because the AI models literally can't find neutral evidence that it exists.

Related: see hubGEO's brand tracking data for how individual companies are scoring across China's six major AI models, and our earlier coverage of the Doubao–Douyin content flywheel that underpins this same dynamic in other consumer categories.