The new storefront is a chatbot

A growing share of Chinese consumers no longer type a brand name into a search box. They ask Doubao, DeepSeek or Tongyi Qianwen what to buy, and whatever the model names first becomes the only option most shoppers ever see. That shift has created a new industry built entirely on gaming, or at least understanding, what a language model recommends: generative engine optimization, or GEO. Beijing-based PureblueAI is one of its most closely watched vendors, and its own year is a useful gauge of how much capital and anxiety this new battlefield is generating 1 2.

Three rounds, one year, unusual backers

PureblueAI has closed three financing rounds in roughly twelve months, an unusually fast cadence for a company founded only in 2024. A seed round disclosed in September 2025 was led by Innovation Works alongside marketing group BlueFocus, which described the deal as its first strategic AI investment since launching an all-in-AI strategy in 2023 1 2. An angel round followed in February 2026, led by Vertex China with Innovation Works and 36Kr returning as backers 3. A further angel-plus round closed in August 2026, this time pulling in a Shanghai semiconductor investment fund and brokerage Jiufang Zhitou alongside a launch of what the company calls a GEO digital employee, an automated agent meant to continuously rewrite and monitor a brand's presence across AI answers 4 5. Chip-focused and brokerage capital backing a marketing-optimization startup is itself a signal of how broadly investors are stretching to get exposure to the AI-consumption theme.

Founder and chief executive Lu Yang is a Tsinghua-trained marketer with more than two decades in the industry, most recently running marketing for ByteDance's Volcano Engine and its Doubao large model before starting PureblueAI 4. That pedigree, straight out of the platform whose chatbot now decides what gets recommended, is part of the pitch: a vendor that understands the model from the inside.

The roster that sells the problem

In July 2026, PureblueAI and 36Kr published the second edition of a joint AI Recommendation Power Roster, expanding on a first edition released earlier in the year. The exercise queries DeepSeek, Doubao, Tongyi Qianwen and Tencent's Yuanbao across five consumer categories, phones, appliances, cars, skincare and beauty, broken into nineteen specific purchase intents, then scores which brands the models actually surface 6 7. The stated finding across both editions is that appearing in an AI answer depends less on brand awareness than on whether a model can correctly parse what a product does and match it to a specific query, a very different skill from traditional search-engine ranking 7.

The headline numbers from the first roster still frame the pitch to prospective clients: PureblueAI says brands it has worked with saw AI-platform recommendation and top-ranking rates rise toward nearly 100 percent from a low base, alongside a jump in sales inquiries 7. With Doubao, Tongyi Qianwen and DeepSeek reporting roughly 382 million, 167 million and 130 million monthly active users respectively as of mid-2026, the addressable audience behind that pitch is not small 7.

A No. 3 spot that lasted one edition

What the roster also shows, whether intentionally or not, is how unstable that recommended slot is. In the sunscreen category, Anessa led the July ranking with a score of 50.76, ahead of Winona at 48.5 and Shiseido at 45.76, a spread of barely five points separating first from third 8. In sensitive-skin cleansers, Winona dropped from third place in the first edition to sixth in the second, just months later, even as it stayed inside the top tier overall 8. Midas Analytics could not independently verify PureblueAI's underlying scoring methodology, which the company has not published in full, but the roster's own before-and-after comparison is the clearest evidence yet that an AI-recommended position is closer to a rented ad slot than an earned one.

The bull case and the catch

The bull case for GEO vendors is straightforward: as AI answers replace search results and storefront browsing, whoever controls how a brand is described to a model controls a meaningful share of Chinese retail demand, and that service is worth paying for continuously. That is precisely why capital as varied as a PR conglomerate, a China-focused venture fund, a semiconductor investment arm and a brokerage has all bought in within a year.

The catch is the same data that sells the service. If a top-three skincare brand can slide to sixth within a few months without any change in the underlying product, the moat around any single GEO vendor's technique looks shallow, and the spend it demands looks recurring rather than one-time. For investors in PureblueAI and the wider GEO field, the roster is simultaneously the best marketing collateral available and the clearest warning that this business may need to keep proving itself every single month.

What to watch

Watch whether PureblueAI discloses a valuation alongside a future round, since none of its three raises to date has included one publicly. Watch whether Doubao, DeepSeek and Tongyi Qianwen push back on third parties optimizing for their outputs the way Google once policed search engine optimization. And watch the next edition of the roster itself: if the same handful of brands keep reshuffling the top ranks every few months, GEO risks becoming a permanent subscription cost for Chinese consumer brands rather than a one-time technology upgrade.