Former Baidu Search Team's GEO Company Raises Tens of Millions, CEO Says Industry Enters "Technical Watershed": Self-Media Content Weight Drops to 10%
Qian Cheng, former Baidu Shanghai General Manager, serial entrepreneur. After leaving Baidu, he founded companies that were acquired by Ctrip and Didi.
In June 2025, he started again. This time, the direction is GEO (Generative Engine Optimization)—making AI tools like ChatGPT, DeepSeek, and Doubao prioritize recommending client enterprises when answering user questions.
The company is called Deep Cloud Sea, with a team of 70+, core members from former Baidu search department.
In its first year, it served over 3,000 enterprise clients with annual revenue in the tens of millions. On October 13, 36Kr reported they completed a tens of millions angel round led by HGTECH.
But what's really worth watching isn't the funding itself, but Qian Cheng's judgment about the industry: GEO is entering a "technical watershed."
He said that before H1 this year, simply publishing more content that could be crawled by AI could achieve AI visibility. But starting H2, the industry is shifting from traffic dividend era to technology-driven era. The threshold for measurable results has dramatically increased—only through technology, algorithms, and monitoring investment can you guarantee quantifiable delivery.
Even more striking: The weight of self-media content being cited by AI has dropped to 10%. The weight of authoritative sources continues to rise.
📌 Key Takeaway: Deep Cloud Sea, founded by former Baidu search team, completed tens of millions in funding. CEO Qian Cheng judges that the GEO industry is entering a "technical watershed." Large model search trace results have expanded from 10 to 60 entries, self-media content citation weight has dropped to 10%, while authoritative source weight continues to rise. Startup window is only 2-3 years left, industry is reshuffling. In 2026, Chinese generative search queries account for 41.2%, GEO market size expected to exceed 20 billion yuan.
Large Models Are Getting Smarter: From 10 to 60 Trace Results
The first key change Qian Cheng mentioned is that large model search trace capabilities are rapidly evolving.
Previously, when large models answered questions, they might only reference about 10 search results. Now? Expanded from 10 to a maximum of 60.
What does this mean? AI no longer just looks at top-tier content—it's digging deeper into more sources, cross-validating information credibility.
Even more significant: the trace result ecosystem is changing. Besides web pages, some AI applications (like Doubao) can now find Douyin videos and other content as reference material.
The challenge this brings to GEO: you can't just optimize text content anymore. Images, videos, audio—all could become AI citation sources. Multimodal content construction has become a new competitive dimension.
Self-Media Weight Drops to 10%, Authoritative Sources Rise
The second change Qian Cheng mentioned might be the most painful.
Previously, many GEO providers used this playbook: batch-register self-media accounts, use automation tools to publish articles, win by quantity. As long as there was enough content, AI would be "fed" the results you wanted.
But now, this trick doesn't work anymore.
The weight of self-media content being cited by AI has dropped to 10%.
Why? Because large models have gotten smarter. They no longer simply "believe whatever they see"—they evaluate source authority, credibility, and consistency.
If you publish the same article on 10 self-media platforms, AI considers this "homogeneous information" and won't give you 10x weight. Instead, if you can be cited on authoritative media, industry vertical platforms, academic databases, and other high-credibility sources, AI will give you higher weight.
This is why Qian Cheng says small GEO companies that lack technical capabilities, insufficient investment, and operational capabilities may gradually fall behind.
It used to be a competition of "quantity." Now it's a competition of "quality." It used to be "who publishes more." Now it's "whose sources are more authoritative."
How Big Is the Market? 20 Billion Scale, 41.2% Penetration
Qian Cheng's judgment is backed by data.
According to the latest CNNIC statistics, China's generative AI user base has reached 901 million, with 69.2% using AI search engines for information queries. DeepSeek's MAU exceeded 192 million, with daily search queries reaching 2.97 billion.
iResearch data shows that in Q3 2026, generative AI search queries accounted for 41.2% of the Chinese search market, doubling year-over-year. By Q3 2027, this proportion is expected to exceed 57%.
Market size? IDC and CCID Consulting jointly estimate that China's GEO service market was about 6.2 billion yuan in 2025, expected to reach 20.3 billion yuan in 2026—a year-over-year increase of over 227%.
Enterprise adoption is also rapidly increasing. iyiou Think Tank research shows that 31.7% of enterprises have deployed GEO, and 44.2% plan to deploy within 12 months—totaling 75.9% of enterprises have incorporated GEO into their digital marketing strategy.
In high-ticket industries like finance, healthcare, and B2B enterprise services, leading enterprise GEO coverage exceeds 63%.
Window Only 2-3 Years Left, Industry Reshuffling
The third judgment Qian Cheng gave might be the most important: The startup window has only 2-3 years left.
Why? Because the industry is moving from technical lowlands to technical highlands. In the early days, you could capture traffic dividends just by creating content and publishing articles. But now, you must have technical foundation, algorithm capabilities, and monitoring systems to provide quantifiable delivery.
Deep Cloud Sea's approach is "reverse engineering": observe what trace checks large models perform before giving results, reverse-engineer algorithm matching, and determine which sites are more likely to be cited.
They've developed a proprietary GEO marketing algorithm engine and built an 8-layer enterprise knowledge graph (entities, relationships, attributes, events, rules, FAQs, tags, training constraints). Around this foundation, they've applied for multiple software copyrights and patents.
Qian Cheng says GEO won't completely replace SEO, but will become standard for enterprise AI marketing. The industry is in a reshuffling phase, and the rules are clear: early technology development, win by quantity; mature technology development, win by quality.
Small companies that only batch-publish articles, lack technical capabilities, and can't provide quantifiable results will be eliminated. Professional companies that understand large model underlying logic, can provide authoritative source construction, and can do continuous monitoring and iteration will survive.
Traditional Media's Awakening: No Longer "Working for Others"
The GEO industry's changes affect not just enterprises, but also media.
On October 14, an analysis article pointed out that traditional media is undergoing a transformation from SEO to GEO. Xinhua Net launched a GEO intelligent agent platform, The Paper launched a "Trusted GEO" communication service, Hunan Daily launched "Smart Eye Trusted GEO," and the People's Political Consultative Conference Network launched a trusted space data governance GEO platform.
On the surface, traditional media has finally found a new path. But the article raises a concern: from pleasing search engines with SEO, to opening accounts on various platforms, to now pleasing AI with GEO, traditional media seems to always be "working for others."
A Reuters Institute report shows that from November 2024 to November 2025, referral traffic from Google search dropped by 33%, and is expected to decline by an average of 43% over the next three years.
Even more painful: ChatGPT used specific news organization content in 54% of answers but almost never cited sources. A joint test by 22 European public broadcasters found that Gemini and Grok 3 cited fake or broken links in over half their answers.
Facing this situation, publishers are shifting focus from generic, service-oriented content to original investigations, in-depth analysis, and human stories. Research shows publishers plan to increase investment in original investigations (net score +91) and contextual explanatory reporting (+82), while cutting AI-replicable general news (-38).
The media competitive landscape is showing a "dumbbell" structure—one end is media organizations deeply cultivating AI-irreplicable exclusive investigations, in-depth analysis, and other high-value original content; the other end is media organizations relying on automated, scaled production; and the large number of daily news producers in the middle will be squeezed.
Conclusion: Technical Watershed, Which Side Are You On?
Deep Cloud Sea's funding is just a microcosm of the GEO industry.
Qian Cheng's "technical watershed" is essentially a signal: this industry is moving from wild growth to regulated development, from traffic dividends to technology-driven, from quantity competition to quality competition.
For enterprises, choosing a GEO provider can no longer just look at "how many articles can be published," but must look at "can they provide authoritative sources," "can they do continuous monitoring," "can they quantify results."
For media, GEO is not another round of "working for others," but an opportunity to redefine their value. From "adapting to algorithms" to "defining standards," from "traffic-oriented" to "trust-oriented," from "content supply" to "value anchoring."
The window is only 2-3 years. Which side you stand on determines whether you can catch this wave.