Raised 1.2 billion yuan in financing, with its revenue surging by 300%: an AI advertising company has achieved explosive growth.

An AI advertising firm founded only two years ago has raised $180 million (approximately RMB 1.2 billion) in financing.

On September 15, Profound announced the completion of a $180 million Series D financing round, co-led by HSG and Kleiner Perkins, with a post-money valuation of $1.8 billion.

Profound is engaged in the currently hottest Generative Engine Optimization (GEO) business.

In the past, when consumers wanted to buy a computer, they would search on Google and Baidu; now, they may directly ask ChatGPT or Doubao: Which models of computers are most suitable for business people who travel frequently? Profound enables brands to know whether AI will mention them, how AI evaluates them, and how to increase their probability of being recommended.

Profound's revenue has tripled in the past six months, and it has more than 1,000 enterprise clients including Walmart and Estée Lauder, with over one-third of Fortune 100 companies using its services.

This highly profitable business in the United States presents a completely different landscape in China. Practitioners have long entered a K-shaped state: a large group of players are struggling to survive, while a small number of players are making huge profits.

Many enterprises are only willing to allocate 0.1%, 0.5% or 1% of their annual marketing budget for a trial first. For some enterprises that have reaped tangible benefits, their monthly GEO expenditure reaches 10 million yuan, and a number of top service providers have already "received more orders than they can handle".

Selling Software in the US, Selling "Press Release Distribution" in China 

Profound started with AI search monitoring, but has now expanded to the entire marketing workflow. This means that although both are called GEO, leading companies in the US are trying to build it into a set of reusable software and marketing infrastructure.

The situation in China is different.

JIAO Shidou, CEO of Moli Index, has long observed the domestic GEO market. He told Pencil News: "Many domestic GEO companies are essentially still providing services. After signing the contract, their main deliverable is to help clients create content and distribute press releases to media, and this model can hardly be regarded as a real technology business."

"Capital is more willing to invest in projects that are technology-driven, replicable and can scale up rapidly. If the business mainly relies on manual work to process orders one by one and lacks high technical barriers, it is naturally very difficult to raise large amounts of financing."

A large number of traditional public relations, advertising and SEO companies in China have transformed into GEO service providers. One practitioner NIU Xiaomeng (pseudonym) roughly divides the companies in the market into three categories: the first category is original advertising, public relations and SEO companies that have added GEO departments; the second category is native GEO startups with a scale of dozens of employees; the third category is more straightforward - they massively create and publish content to influence the information sources of large language models and obtain higher rankings.

Among them, the third type of companies may even find it easier to make profits.

Niu Xiaomeng's team belongs to the second category, taking the technical route, which requires R&D personnel, monitoring systems and compliant content production.

He estimates that the comprehensive cost of some companies that rely on a large amount of low-cost content for optimization may only be one-fifth or even one-tenth of theirs.

These companies can locate their content teams in third- and fourth-tier cities, and do not need to bear high R&D costs. When they get more clients, they just hire more people and publish more content. For a market that still takes "the volume of distributed press releases" as an important delivery indicator, this model has a price advantage instead.

Niu Xiaomeng said: "That's why native GEO companies like us that focus on technology are really struggling. The orders from small clients are too small to support the team; big clients have sufficient budgets, but for a 20-something-person company like us, we often can't even meet the entry threshold. If we really want to compete on low prices, we can't beat those companies that rely on low-cost press release distribution and content poisoning, and we are basically stuck in the middle."

There is a very contrasting result:

In the early stage of GEO, the companies with the strongest technology are not necessarily the ones that make the most money easily.

The situation Jiao Shidou has observed is similar.

He believes that there are already thousands of GEO service providers in China, but the entire market is highly fragmented. As long as the main revenue still comes from manual services and media distribution, this business can hardly be replicated infinitely like software.

This is also one of the reasons behind the huge valuation gap between Chinese and US GEO companies.

Niu Xiaomeng believes that the US market has begun to extend from pure GEO to a model closer to performance marketing, covering advertising, e-commerce and transactions.

A large number of projects in China are still at the stage of "whether the brand is mentioned by AI" and "whether the ranking has improved".

One side sells software, Agents and advertising infrastructure, while the other side largely sells manual services.

Investing 10 Million Yuan Per Month 

The biggest hurdle that China's GEO industry has difficulty crossing is to let clients see tangible results.

Traditional internet advertising has been working on this for more than 20 years.

When placing ads on Baidu, you can know how much it costs to get a click or a sales lead; on e-commerce platforms, you can see how many people enter the store and how many people place orders; live streaming platforms can even calculate clearly how many new followers a live stream gains and how many products are sold.

But GEO can hardly achieve this at present.

"Clients will ask me: What exactly can you bring me? Is it product sales, traffic diversion or customer acquisition?" Niu Xiaomeng said, "It's very difficult for us to promise these results now."

For example, a brand has placed ads on Douyin, done content marketing on Xiaohongshu, and purchased GEO services at the same time. A consumer first sees the product on Douyin, then asks Doubao about it a few days later, and finally completes the purchase. Which channel should this order be attributed to?

Therefore, even if enterprises think AI traffic is worth paying attention to, they do not know how much budget they should allocate to GEO.

Some clients Niu Xiaomeng has contacted only allocate 0.1% to 1% of their annual marketing budget for a trial. Some small enterprises even only purchase the service for one month to optimize several keywords, and the service stops as soon as the project ends.

One important reason why enterprises are cautious is that the effect cannot be accurately measured.

But some players have already started to find ways to calculate this account clearly.

Jiao Shidou has contacted a consumer electronics enterprise that started to invest in GEO in August and September last year.

In order to independently measure the effect of GEO, this company has specially developed a product for the GEO channel. In this way, the sales generated by this product can be relatively independently attributed to the GEO channel, making input-output ratio easier to quantify.

After the effect is clearly measured, the budget has continued to increase. Jiao Shidou revealed that the amount this company invests in GEO every month now "may have reached nearly 10 million yuan".

This also shows that enterprises do not lack budgets, but they need to see tangible results first. Whether GEO can develop from a "trial" to a long-term investment depends on whether the effect can be clearly measured.

This is the most notable set of K-shaped contrasts in China's GEO industry at present.

Most clients are still testing with 0.1% of their budget; a small number of companies that have realized the closed loop of monitoring and conversion have dared to invest nearly 10 million yuan per month.

So the problem may not be that enterprises have no money. Brands need to know how AI views them, whether it mentions them, and how they differ from their competitors in different large models such as Doubao, Yuanbao, DeepSeek and Kimi.

A large number of monitoring products have emerged in the market today. Niu Xiaomeng observed that Tencent has also begun to try to build a similar GEO monitoring platform. He estimates that there may already be more than 100 companies developing similar monitoring products in China, but there is not yet a platform widely recognized by both the demand side and the supply side, just like what Newrank was for WeChat official accounts in the early days.

Jiao Shidou also believes that "visibility and quantification of perception" will eventually become a layer of infrastructure in the AI marketing era.

Large Language Model Companies Also Want to Profit from GEO 

In the short term, low-cost press release distribution may be very profitable. But Jiao Shidou believes that this window of opportunity will not exist forever.

The reason is that large language models are also evolving.

He observed that after some models adjusted their information sources this year, some previous optimization methods soon became invalid.

For example, Doubao has begun to cite more content from its own ecosystem, and the form of information sources has expanded from text to video. The original delivery system built on text and image content may need to be completely rebuilt suddenly.

"Some service providers are finding it more and more difficult to operate, while a number of top teams have received more orders than they can handle. Companies that used to acquire clients through low prices, channels and rapid expansion of sales teams may see their original delivery system become invalid once the model changes its rules; in contrast, companies that truly master model, marketing and data capabilities will benefit."

Large language model developers have also started to close loopholes.

Jiao Shidou told Pencil News: "In the past, you could publish a random press release, and after it was indexed by the large language model, the model's view on a brand might change.

It is different now. When a large language model sees an article, it will first judge whether it is a marketing article. If it is, the model may not adopt it, and even conduct cross-verification."

Because the large language model developers themselves may also want to take a share of this market. Niu Xiaomeng told Pencil News: "Internet platforms like ByteDance have indeed begun to explore GEO and its commercialization."

Because a regular GEO company lacks almost all the resources that ByteDance has. It has AI entry like Doubao, advertising system like Ocean Engine, and transaction scenarios like Douyin E-commerce and local life services.

Niu Xiaomeng judges that for internet platforms like ByteDance, it is not particularly difficult in technology to connect large language models to existing advertising systems, set different links for different projects, and track subsequent conversions.

Doubao has previously tried to access product and local life scenarios. After users search for products and services in AI, they can even directly enter the transaction process.

Google has long been doing similar things: pushing AI search from "providing answers" to "doing business": products, advertisements, brand Agents, shopping carts and payment are being connected into the same full path.

This article does not constitute any investment advice.

This article is from the WeChat official account "Pencil News" (ID: pencilnews), written by XIN Xiao, and authorized for release by 36Kr.