Content is based on the dialogue between Monica and Bennett on March 25, 2026.
Guest Introduction:
Bennett
CEO of Crescendia, building Marketing Agent to solve growth problems for outbound enterprises
10 years of AI product experience, previously served as AI Planning Director at vivo mobile
Monica
Today, I'm very happy to invite Bennett to our live broadcast. Bennett started as a product manager, previously worked at vivo for many years as a product lead, and has many years of AI product management experience. Now he has founded Crescendia, building Marketing Agent to primarily solve growth problems for companies expanding overseas, and is quite professional in GEO. Today, we've invited Bennett to discuss GEO-related topics with everyone. Please say hello, Bennett.
Bennett
Hello everyone, and thank you, Monica, for the introduction. I'd like to elaborate on my recent career direction and the thinking behind it. The reason I chose this track is that I previously worked in the mobile phone industry, responsible for Siri-like AI assistant products, which gave me a rather unique industry perspective. Mobile phone products inherently have a hardware perspective, involving the layout logic of various traffic entry points. Therefore, I've had long-term and in-depth thoughts on how to design a search-plus-AI model like Xiaohongshu, and how to differentiate vivo's Jovi-like products from Doubao and ChatGPT. At the same time, mobile phone products are highly correlated with the ecosystem, which is also an important industry entry point.
Based on this background, my perspective on GEO and SEO is also different. The core is to consider what kind of ecosystem is built among brands, users, and platforms through traditional search or AI assistants, and what the current industry rules are. This thinking stems from my tenure in the mobile phone industry, where I served as the head of the planning team, coordinating strategy, technology, and product aspects, and also completed internal financing work. Initially, I promised the company head that with an investment of 2 billion yuan, we could build a complete AI product system, from pre-training and product implementation to ecosystem construction, betting on the future of an industry.
Therefore, when the concepts of GEO or SEO emerged in the industry, we had already made relevant attempts. In my opinion, what we do revolves around the new AI-native ecosystem. It is based on this new ecosystem perspective that we design products, leading to some different industry insights. I hope today's exchange can bring some new inspirations to everyone.
Monica
Based on your past practical product experience, how do you think your current venture into GEO-related businesses is connected to your previous experience in the mobile phone industry, given the current industry stage? And why did you choose to start this venture and enter this particular track?
Bennett
Entrepreneurship itself is a final decision, and the prerequisite for that decision is clarity about what needs to be done. The core still comes from confidence in the endeavor and judgment of the industry's prospects. When I was previously responsible for AI assistant products, the core performance indicators were resolution rate and satisfaction. Only when these two indicators met standards could the number of users and user activity duration be effectively increased. All AI assistant products are built around these two indicators.
In the process of building AI assistants, I summarized three core tasks for AI: information acquisition, instruction execution, and emotional companionship. Among them, the information acquisition section is essentially the distribution of information services, falling within the scope of general advertising. And as an AI assistant product manager advancing this section, I felt a significant industry pain point.
The instruction execution section already has mature implementation models in emerging fields such as education and health, but the advancement of the information acquisition section has been fraught with difficulties. I believe everyone who has experienced various smart searches will agree that the core of information acquisition is online retrieval. When we first advanced this section, we deeply felt this problem.
At that time, we simultaneously interfaced with solutions from multiple platforms. Alibaba initially provided us with a combination of Quark and Bing, primarily to solve RAG technology implementation and online retrieval problems. Baidu and ByteDance also provided their respective solutions. However, actual testing revealed that these solutions could not adapt to mobile phone usage scenarios. Ultimately, we had to build our own adaptive system. When the logic diagram of the entire strategy was fully sorted out, I felt a huge impact. The complexity of the entire system far exceeded expectations, with various fallback strategies nested layer upon layer, making advancement extremely difficult.
It is precisely because of this that I had a strong thought at the time: what should AI-native search look like in the AI era? It certainly wouldn't be the current model of making simple patch-like modifications to traditional search with AI and then calling it. This was my core perspective when entering this field.
From the overall industry landscape, brands, platforms, and users are the three core entities. What fundamental changes has the emergence of AI brought to which party? The answer is that, besides these three entities, various intermediary service links may be reduced or even replaced. For example, many intermediary tasks can be completed through large models, and even the functions of various agent applications can be covered by large models. From an end-game perspective, large models may eventually become super applications, directly connecting users and brands.
If this end-game trend develops, future users may be monopolized by a few core AI entry points. User data and memories will be controlled by these entry points, and various websites and brands will ultimately be reverted to their original state, becoming mere providers of information and services. Of course, this is just the end state, and a long process of industry evolution will occur in between.
Looking back at the present from an end-game perspective, what unique value can brands ultimately possess? The answer is unique information and service content. In this process, the core industry contention lies in the evolution and competition between AI search and traditional search. However, the ultimate demand for AI will certainly not be individual Web search results, but a systematic long-context result.
Because an AI assistant, after sufficiently understanding the user, will ask targeted questions based on the user's memory, and the current Web pages facing humans can only carry very limited content. In this context, it is extremely difficult to answer users' diverse and highly personalized questions with limited search results. For example, ChatGPT can currently only retrieve 60 search results, 30 from Google and 30 from Bing; Gemini retrieves 100 Google search results. Such limited windows simply cannot meet users' personalized needs.
It is precisely because of this that I saw the industry opportunity for AI-native end-game websites at that time. That is, under the long-term vision of AI-native, brands need to build their own knowledge bases, and through these knowledge bases, develop the ability to build various index directories, allowing various intelligent search tools to access these AI-native official websites and knowledge bases to assist users in decision-making and purchasing.
The future service system should be structured where Web MCP provides information and knowledge bases, and UCP and ACP complete the purchase loop, ultimately forming a complete service system delivered to the customer. It is because we saw this industry opportunity that we decided to focus on future AI-native websites.
Of course, the current industry still largely relies on the Web. We will try to be compatible with the current form, as there is still a long way to go from the present to the AI-native end-game. We always approach current tasks with an end-game perspective, and because of this, the progress of related businesses has been quite good. I will share specific examples later.
How to Build Brand Knowledge Assets
Monica
How can brands efficiently build knowledge bases? In recent years, everyone has been talking about digital transformation. In the process of building knowledge bases, companies often pull out documents scattered across various office software and customer service software, which are disorganized and extremely difficult to manage. Are there any efficient construction methods?
Bennett
When discussing enterprise knowledge, we first need to clarify the enterprise's workflow: through which channels do the various knowledge within the enterprise ultimately reach customers? Currently, there are three main channels: search engines, AI assistants, and search media points. All three of these channels are undergoing a core change, which is the restructuring of the entire distribution service system by AI.
After transitioning from the internet industry to the mobile phone industry, I found that the core pain point for users today is not that products lack features, but that there are too many features, and information and services are too complex, leading users to be unable to find, know, or use them. Essentially, the distribution link has a problem. Humans inherently have natural physical limitations in the physical world, so efficient interaction through AI is needed to solve this core pain point.
So, the core problem AI needs to solve is that users can't find, don't know, or don't know how to use things. Based on this, we can see that GEO is a new industry rule brought by AI search, while SEO is the existing rule of traditional search. The boundaries between the two are gradually blurring. Observing the trend of SEO rule changes in 2025, we can find that the weight of content, personalization, and user behavior is continuously increasing, which is consistent with the core logic of GEO. This is because both rely on the foundational capabilities of large models to optimize product experience and solve users' resolution and satisfaction problems.
At the same time, we also need to pay attention to changes in the social media landscape, such as the overseas X platform, where interesting forms of interaction have emerged: when someone posts an opinion on the platform, users will @Grok in the comments section, asking it to judge whether the opinion is true, and Grok will provide a corresponding answer. This form of interaction is essentially changing the logic of information distribution.
In fact, various communities are making similar attempts. Every user conversation is reshaping the logic of information distribution. This is a transformation process that all platforms will undergo after the emergence of AI. Returning to the core issue of enterprise knowledge base construction, we have not focused on the complex and scattered internal knowledge bases of enterprises, but rather on the content that enterprises have already accumulated on various platforms. Our core idea is to build an AI-compatible knowledge base based on the content that enterprises present externally.
Currently, our knowledge base content sources mainly fall into three categories: The first category is the company's own official website, as well as the official websites of leading competitors in the industry. If the client we serve is a leading enterprise in the industry, or its competitors are among the top 5 in the industry, we will collect the official website content of these five companies. Previously, during pre-training, we had experience in processing massive samples, and accumulated mature methodologies in sample cleaning and labeling.
In practical operation, we found that even for clients with very good SEO, their official website might have millions of pages, but only tens of thousands of them truly contain effective information. Therefore, after integrating the official website content of a top 5 industry enterprise, the effective page count is only in the hundreds of thousands, making the processing efficiency and difficulty manageable.
The second category is SERP search results, which is the core method of traditional SEO. We collect various keywords from search results and use third-party tools like Semrush to obtain data such as keyword DR weight for combination analysis. Through SERP search results, we can see not only the core players in the industry but also which vertical media have published relevant information. These are all high-quality content sources for the knowledge base.
The third category is social media platform content. Most social media platforms have complete APIs that allow access to relevant content under specific topic words. Some platforms also allow RSS subscriptions to obtain shares from industry influencers. For example, we have served some Web3 clients. Public online resources in this field are not abundant, but brand founders and industry influencers publish a lot of relevant content on the X platform. This content is valuable asset for building a knowledge base.
After collecting this content, we put it through a complex processing flow, including content cleaning, configuration, and indexing, ultimately completing the knowledge base construction. Essentially, our knowledge base construction is based on public data accumulated by enterprises on the three major platforms, rather than raw internal enterprise data.
After all, internal digital employees are responsible for sorting and processing internal data, while our core entry point is to integrate and optimize the public data that enterprises have already produced and presented to users. Currently, we have extensive practical experience in this area. If we expand to internal enterprise data organization in the future, we can discuss it further.
GEO Poisoning?
Monica
Before we delve into case studies, let's discuss a term that has been frequently mentioned recently – GEO poisoning. Can you, from your professional perspective, explain what GEO poisoning is?
Bennett
I have always been deeply involved in overseas market-related businesses. Many people have asked me why I don't do domestic GEO business. In fact, there are not many teams that can do overseas GEO business. Of course, after listening to my next sharing, everyone will find that what I essentially do is not traditional GEO business, but our method can effectively address GEO poisoning problems.
The reason I don't do domestic GEO business is primarily because when I was responsible for the information retrieval section of AI assistants, I felt a huge difference between domestic and international environments. Using domestic search engines versus overseas search engines for information retrieval yields vastly different integrated results. I once deeply researched the reasons behind this difference and found that to acquire information through search engines, the core is to grasp three dimensions: relevance, timeliness, and authority.
Everyone can ponder a question: what exactly is the authority of the Chinese internet? I believe many people will fall silent. But the definition of authority for the overseas internet is very clear: it's the DR (Domain Rating) weight of a company's official website. The development paths of overseas and domestic internets are vastly different. Overseas, mobile internet developed after the internet ecosystem was relatively mature. Therefore, based on the internet ecosystem, the development of various vertical communities has clear logic.
For example, many social media accounts overseas require the endorsement of a company's official website for verification. The entire ecosystem extends from the official website to vertical communities. And overseas companies operate their official websites with a long-term goal of over 5 years, so they are extremely cautious when publishing official website content.
I have a typical case around me: a company that made digital humans once provided application services for gray industries related to online dealers. Although it brought a large amount of traffic in the short term, it also caused serious negative effects and was eventually punished by Google, with related content being taken down. Users who originally learned about the company through its official website could not get information due to the content takedown, and then filed a large number of complaints against the company. The company suffered two rounds of continuous punishment, and the official website's authority and traffic plummeted, taking 6 months to recover. Even changing the domain name and starting a new account became extremely difficult.
The competition in the digital human industry is already fierce, and the industry window is very short. This incident directly caused the company to miss development opportunities. This case also confirms that a company's official website, as the core platform for content, is the foundation and endorsement of the brand. This is the core value of authority in the overseas internet.
Building authority on the overseas internet is extremely difficult. This is why, when doing overseas SEO, many people know that building official website authority is effective, but they can never calculate a clear ROI. Ultimately, bosses are more willing to choose advertising, because they can see short-term effects. However, doing SEO and building official website authority primarily involves producing better content than competitors and requires long-term operational accumulation. This long-term approach makes it difficult for many people to maintain patience.
However, the domestic internet ecosystem was directly impacted by mobile internet, and brands have not accumulated high-quality official website assets. You can observe the official websites of domestic enterprises; most of them primarily serve after-sales functions, or even become channels for user complaints, with sales conversion and brand cultivation functions largely absent, naturally leading to no traffic.
So where are domestic brands' assets accumulated? Taobao live streams, Taobao flagship stores, and other e-commerce platforms. The rules of these platforms are firmly controlled by the platforms, and brands do not truly have control over their assets. This leads to a serious problem: when a brand does not have its own authoritative platform, various KOLs and KOCs speak for the brand. Their evaluations of the brand vary, and AI has no concept of authority, so it cannot judge the authenticity of these evaluations, nor can it punish those who publish false information.
Essentially, it's because domestic brands haven't truly operated their core content platforms, leading to a domestic search system that only has two dimensions left: similarity and timeliness. In the first half of 2024, we tried to conduct relevant effectiveness tests in China and found that DeepSeek was being severely abused.
DeepSeek itself is a product of an AI lab. Its online retrieval function was built relatively simply, without significant human investment in commercialization. It primarily relies on some domestic search companies to provide integrated solutions, with only simple optimizations internally. It doesn't undergo major updates every one or two months like ChatGPT, nor does it continuously iterate functions like Doubao.
This led to a result: publishing 10 related pieces of content on platforms like NetEase News could achieve good search results on DeepSeek. We found at the time that publishing 1 piece of content per week on such platforms could continuously generate traffic. The core reason for this phenomenon is that the domestic search system lacks the concept of authority, and can only compete on text relevance and timeliness.
Before March 15th, when I was talking to some domestic practitioners, I asked them about their current content publication volume, and the answer was shocking: 1700 articles published in one day, and the content could go live in 3 minutes. In my opinion, the essence of GEO poisoning is to generate traffic by publishing a large quantity of content with high text similarity that won't be penalized, relying on timeliness. This is the core logic of poisoning.
To penalize such poisoned content, the core mechanism is that users, misled by false content, file complaints, and the platform then punishes the publishing entity accordingly. The foundation of this logic is the concept of authority. However, in the domestic ecosystem, even if false content is discovered, after blocking the publishing account, the perpetrator can immediately create new accounts to continue publishing content, continuously flooding traffic. The ultimate result is that, due to the lack of authority, no entity vouches for the quality of the content.
In contrast, ChatGPT also went through a similar phase. I believe everyone has had relevant experiences or heard industry voices. When optimizing ChatGPT previously, Reddit content was a core optimization handle because Reddit content has very authentic user feedback.
This industry turning point was specifically on October 13, 2025. I remember it very clearly because we had just taken on a new client's case, and Reddit content at that time had an extremely significant effect on ChatGPT's optimization. A major industry adjustment for ChatGPT at that time was to increase the number of Google search results retrieved to 100. This was a landmark event in the industry, meaning that for every search query, ChatGPT would retrieve 100 Google search results.
It's important to know that in traditional SEO optimization, ranking in the top 20 is already very good, and long-tail content beyond the 30th position generally receives no user attention. However, Reddit content has genuine user feedback across various platforms. Therefore, when ChatGPT retrieved 100 search results, a large amount of Reddit content appeared in the first 10 pages, which gave practitioners optimizing Reddit content excellent results. This behavior was essentially GEO poisoning for ChatGPT.
In response to this situation, ChatGPT made two major adjustments. First, Google reduced the number of search results it retrieved from 100 to 10. This had a huge impact on ChatGPT: either bear 10 times the cost and 10 times the response time, or adjust its search strategy. ChatGPT ultimately chose the latter, introducing a 30+30 search strategy, meaning 30 Google search results plus 30 Bing search results.
This adjustment directly led to a large amount of Reddit's long-tail content being untrievable, and Reddit's traffic began to decline significantly. But this was not the most fatal blow. ChatGPT also discovered GEO poisoning behavior in the industry and began to optimize from a product level. From the brand's perspective, they certainly hope to influence or even deceive AI through various means to gain more traffic; but from the perspective of an AI assistant product manager, this causes great anxiety, because if AI provides false information to users, it will cause users to lose trust in AI, ultimately losing the opportunity to compete in the core AI assistant track.
At that time, ChatGPT's optimization strategy was very clever. The first point was to build a logical chain judgment mechanism. There were two dimensions for specific judgment: first, whether there was relevant brand information in the pre-training data. I will share specific examples later. Some brands' unique selling points are only provided by that brand. Even after referencing 10 relevant search results, with 4 of them being the brand's content, ChatGPT still wouldn't recommend that brand.
The reason is that ChatGPT, through inference, found that there was no relevant information about the brand in the pre-training data, and there was a possibility of being deceived. Even if it considered the brand's content to be somewhat reasonable, it would not recommend it, which led to the brand not being able to obtain stable exposure.
The second point is to combine the search result weights of Google and Bing for judgment: Google's search results are primarily high-authority content, while Bing's are primarily long-tail content. When combined, if a piece of content appears in both results and has third-party official website endorsement, ChatGPT will prioritize its recommendation. If there is only third-party content without official website endorsement, the exposure rate of that content will be very low.
When multiple pieces of content appear simultaneously, ChatGPT will prioritize recommending content with official website endorsement. Simply put, it uses the dimension of authority to effectively correct false information and improve the authenticity of content recommendations. Returning to the core issue of GEO poisoning, my view is: domestic GEO poisoning is essentially due to the lack of the concept of authority, with a large amount of discourse power held by KOLs and KOCs.
These KOLs and KOCs can arbitrarily evaluate brands and will not be punished for publishing false information, unless it involves defamation and is subject to legal sanctions. When the internet is filled with false content, AI has no concept of authority and cannot judge the authenticity of the content at all, and humans also find it difficult to distinguish. This is the core problem caused by the lack of authority.
In contrast, ChatGPT, after a series of adjustments, increased the dimension of authority in content recommendations. Currently, overseas GEO poisoning has significantly decreased. Most of the previous practitioners were opportunistic, and ChatGPT's adjustments have brought the industry back to the most difficult task of building authority. This is my understanding of GEO poisoning.
Case Study 1 - AI Native Office Software:
Achieving 4x Ad ROI Growth with Precise Traffic in Practice
Monica
Next, let's move on to today's core case study sharing session. What insights and knowledge does Bennett hope to bring to everyone through the sharing of these three cases?
Bennett
I primarily want to share from a product manager's perspective, combined with my practical experience, how I would manage growth and operations for a brand in the AI era, hoping to offer some new perspectives.
Monica
The first case is an AI-native office software. Bennett, please introduce this product and its status when you took on the case.
Bennett
The core function of this product is AI PPT creation, primarily targeting overseas markets and competing with industry giants like Genspark. The product itself has certain strengths and has accumulated some traffic. After the emergence of ChatGPT, many users asked questions related to AI PPT creation on ChatGPT. The brand saw this as a huge industry opportunity and tried to enter the GEO-related growth track.
However, our operational approach is different from all other GEO teams; we simply don't do any AI visibility optimization. Here, I need to explain why. I believe that when everyone currently engages in GEO-related businesses, the first thing they encounter is various AI visibility optimization methods. But if you've seen the backend of an AI assistant, you'll find the core problem.
The backend of an AI assistant categorizes search queries based on user profiles, and the repetition rate of these search queries is extremely low, which is astonishing. Imagine, almost no search queries are repeated within a single day, meaning traffic is highly dispersed. In such a context, how do you select representative samples for monitoring search queries?
Through serving clients and our own practical experience, we found that AI-driven traffic is relatively evenly distributed across five to six thousand pages, leaving operators with no effective operational leverage. Therefore, I believe that any AI visibility optimization in the intermediary stage is ineffective; only the brand itself can effectively categorize search queries.
A brand can clearly attribute a certain type of search query to a specific user intent and can determine the proportion of traffic for that intent. Only the brand can do this effectively; other service providers simply cannot. Therefore, I believe that, apart from ChatGPT and Gemini releasing GSC-like products, it is not very meaningful for other third-party service providers to optimize AI visibility, because the entire chain is disconnected.
Next, I will outline the complete funnel logic of AI search: after a user issues a search command, AI first rewrites the command, converting it into a general search command, removing invalid information, and then combines it with the user's memory to fill in what the user did not clearly state or expressed inaccurately.
After rewriting the instruction, AI performs intent recognition to determine the type of information the user wants to obtain, and each category has corresponding slot concerns, such as whether the user has requirements for timeliness, subject, or region. Only by making these judgments can AI answer user questions as accurately as possible, then execute the search instruction, retrieve relevant results, and finally piece together the answer to deliver to the user.
This doesn't even touch upon technical issues like rank and chunking. Setting aside these technical details, we can see that each platform's AI search strategy differs, which is why everyone feels that the user experience of various intelligent search tools is not ideal, because their underlying search strategies are very crude.
Products from major companies like Doubao offer a better user experience, primarily because they integrate a large amount of search technology, achieving AI-driven search. This system is extremely complex. This also confirms my view: apart from platform providers, it's almost impossible for other third parties to optimize AI visibility.
So, in such an industry context, what can third-party service providers do? My first thought is to deeply bind with brands. Either bind with platforms or bind with brands. The core leverage is the brand's knowledge assets. We also advanced this office software case based on this idea.
Of course, in this case, we haven't yet presented a complete knowledge asset building solution. The client's core demand at that time was also very clear: not to achieve exposure for specific search queries, but to see how much traffic AI could bring to the brand. Therefore, we determined the core strategy: no limitation on search queries, no limitation on AI visibility, with traffic as the core monitoring indicator, to count the number of users who entered the brand's official website from ChatGPT with UTM parameters.
In three months, by publishing AI-friendly content on the brand's official website and web links, we brought 10,000 unique visitors (UVs) to the client, accounting for about 3% of the brand's total UVs for that month. This case occurred in the first half of 2025, when the proportion of traffic brought by AI to brand official websites was generally around 1%. We increased it to 3%, achieving effective traffic growth.
In the process of advancing this case, we made several core discoveries: First, the traffic brought by AI is extremely precise. Users brought by ChatGPT with UTM parameters had a payment rate nearly 2 times that of Google Ads, and the average payment per user also reached 2 times that of Google Ads. According to the ROI calculation method in Martech, the ROI for this part of the traffic was more than 4 times that of Google Ads, indicating a very impressive conversion effect.
But secondly, the dispersion of AI traffic remains a prominent issue. To increase the overall traffic brought by AI, content needs to be produced around thousands of topics. Such large-scale content production, if published only in blog form, can easily be deemed duplicate content due to low quality. Therefore, how to produce enough high-quality content on the official website has become a core industry challenge.
However, it is undeniable that as long as high-quality content production is done well, AI traffic will bring significant growth to brands. We also validated these core conclusions through this client, which is the most valuable discovery in this case. As for the specific implementation details, many practitioners have shared relevant information; I mainly want to offer some differentiated industry insights.
Monica
Are the operations you just mentioned essentially the same as doing SEO? What do you think is the core difference between GEO and SEO?
Bennett
As a practitioner who started as an AI assistant product manager, I have never deliberately distinguished between SEO and GEO. I only focus on one core point: what is the ultimate performance metric for users, whether it's a search engine or an AI assistant? After the emergence of large models, with their capabilities of intent understanding, memory, execution, and reasoning, the core ultimately remains solving users' resolution rate and satisfaction problems.
Therefore, as long as content is created around resolution rate and satisfaction, SEO and GEO are merely different evolving versions of industry rules. As long as the produced content can bring traffic to the brand, it naturally conforms to both SEO and GEO rules. Therefore, throughout our entire operation process, we hardly study the specific rules of SEO and GEO. The core is to think based on the logic driven by large models, recalling what kind of data labeling and deduplication training methods were used to clean data when pre-training large models.
We apply this pre-training method to the construction of industry data, and then make the constructed industry data compatible with Web forms to build Web authority. This is our core operational approach. Therefore, I am not a traditional SEO expert and have not invested much effort in the professional rules of SEO. Instead, I start from first principles and work with the team to build knowledge bases, exploring what kind of content structure is more friendly to AI, and replicating the methodologies used for AI assistants to experiment. This is our core practice. Therefore, I don't particularly care about the specific rules of SEO and GEO; I just found and summarized some traffic attribution problems during practical operations.
Monica
You just mentioned that the payment rate and average payment per user for AI traffic are both 2 times that of Google Ads. What is the core logic here? Is it because users brought by AI are more precise, or because their quality is higher?
Bennett
The core reason is that the users brought by AI are precise enough. It was only recently, after working on brand growth, that I came into contact with many marketing veterans and learned a lot of marketing concepts from them. Previously, when working on internet products, I was generally familiar with Alibaba's AIPL model and ByteDance's current 5A user model. The core architecture of these models is basically consistent, all outlining the complete user journey from awareness to repurchase: awareness, interest, comparison, purchase, repurchase, or evaluation.
Everyone can ponder a question: what impact has the emergence of mobile internet had on this journey? Mobile internet essentially skips the awareness, interest, and comparison stages. Many brands need to rely on influencer content to achieve user brand awareness and cultivation, while the brand's official website or core platform becomes a pure conversion channel.
Brands in e-commerce should deeply understand this. In the early stage, they rely on investing in influencers and advertising to cultivate interest. When users enter the brand's independent platform, they directly make a purchase, primarily serving only the conversion function. The essence of this phenomenon is that the user's awareness, interest, and comparison stages have shifted.
What is the awareness stage? It's primarily about users understanding "what" the product is and "how to do" things. The interest stage is when users are attracted by the product's core selling points, essentially the brand's ability to capture core keywords. The comparison stage is when users compare products from different brands, focusing on the difference analysis between brand A and brand B.
When we built content for this office software, we did so around this user journey. The emergence of AI has brought a fundamental impact to this journey, because we can clearly see that content with high resolution rates and satisfaction levels is professional, complex, and systematic.
In multi-turn conversations with AI, users will form a classic dialogue logic: first, state their needs and problems encountered, then ask AI to provide solutions, then ask if the solution is optimal and can solve their own problems, then ask AI to recommend multiple products and explain the reasons for adaptation, and finally generate a purchase intention.
In this multi-turn dialogue process, AI helps users complete the entire process from awareness to cultivation. When users finally enter the brand's official website through the link provided by AI, they have already reached the final conversion stage. Therefore, AI traffic has a higher conversion rate, primarily because the traffic is precise enough, and AI has already implanted the initial intent for the user.
I've also had such an experience recently. In our startup process, we need to build many engineering systems, such as a crawler knowledge base. Previously, this was done by a team of hundreds of people, but now we need to build it from scratch. I consult Gemini and ChatGPT to find high-quality suppliers. After testing, I found that the suppliers recommended by AI are all very good and can provide us with a lot of high-quality content.
Looking back at this process, what entities are most impacted by AI? Besides brands, KOLs will also be greatly affected. When I was working on the AI assistant backend, I found that in fields where content is missing, such as medical aesthetics and healthcare, AI plays a more important role because AI can provide users with a complete logical chain of reasoning. This capability will have a huge impact on traditional KOL cultivation models.
Returning to the question itself, the reasons for the high conversion rate of AI traffic are, first, AI's precise identification of user intent, which solves the problem of scattered traffic; and second, AI helps users complete the entire journey from knowledge construction to cultivation and conversion, significantly shortening the user's decision-making process. After being tormented by AI PPT-related problems for a long time, users consult ChatGPT for solutions. ChatGPT recommends this office software to them. Users then enter the official website and directly place an order, even purchasing a quarterly card. The core reason is that AI has already clearly explained the product's suitability and core advantages to the users.
During internal sharing, I also talked about this industry trend: future AI assistants will directly complete the service loop, without the need for third-party service providers. The AI assistant itself will be an e-commerce portal, and all user purchases can be completed within the AI assistant. Future software stores and APP stores will achieve a closed loop within the AI assistant, and this office software case is a valid verification of this logic. This is also the core reason why AI traffic is precise enough.
Monica
From a user's perspective, when I use AI and ask it to recommend a product, I deliberately challenge and question the AI. At this point, to please me, AI often changes its recommended answer. This is completely different from the logic of KOLs, whose recommendations are very purposeful and who will firmly defend a certain product. AI's core goal is to satisfy user satisfaction, and it will continuously adjust its answers. Do you think this phenomenon poses certain risks? For example, if AI initially recommends product A, but after the user questions it, AI then recommends product B, will this affect the brand's traffic and conversion?
Bennett
The core of this phenomenon is still driven by AI's resolution rate and satisfaction, and the underlying risk depends on the user's understanding and trust in AI. For advanced users, who clearly know that AI has hallucination issues, they will challenge and question AI to prompt it to provide more precise answers. We call this process "exchanging tokens for results"; users continuously ask questions, and AI continuously optimizes its content.
However, this process also reflects that for AI to achieve a high resolution rate and satisfaction, it needs to meet two core conditions: first, AI must understand the user well enough and have enough user memory; second, the information AI obtains must be accurate enough to provide users with high-quality solutions. The former is influenced by GEO, while the latter is influenced by the user's interaction time with AI and memory context. Both need to cooperate to improve AI's service capabilities.
For advanced users with a deep understanding of AI, they will consume tokens to get better service results. But for most ordinary users, AI can already provide a solution to quickly find precise information. Even if AI changes its recommended answer, ordinary users will quickly make a decision to cultivate interest because their trust in AI is high enough. Only when users are misled by AI's false information will they start to repeatedly question and push AI, and AI itself is constantly iterating and improving.
This is like the chasm theory in the industry. Users who know how to use AI will consume tokens to get better results. For most ordinary users, the AI experience is already far superior to traditional information acquisition methods, because these users were previously easily misled by KOLs or false information, and AI can provide them with better information, allowing them to cross the chasm of information acquisition. This is the first point.
Secondly, user memory plays a crucial role in AI recommendations. Everyone can ponder a question: if AI first recommended a certain brand to a user, and the user had no negative feedback after using it, then in subsequent recommendations, AI would consider that the previous service achieved a high resolution rate and satisfaction, and would continue to recommend that brand and related products to the user. This phenomenon is very critical; it is essentially about seizing the traffic high ground in the AI era, where the first-mover advantage will be very significant.
Some brands are already using this rule for operations, for example, by adding a button on their official website to guide users to "Ask ChatGPT all questions about this website." When users click it, they enter ChatGPT with specific search instructions, strongly constraining ChatGPT to prioritize retrieving content from that brand's official website. This user action leaves a memory in ChatGPT, and subsequently, AI will continue to recommend relevant content from that brand to the user. When I saw this operation, I realized that many brands had already started using AI's memory rules for growth.
AI itself is also constantly evolving and optimizing. From a user's perspective, it's already difficult to distinguish truth from falsehood in internet content. As a male user, I like to read content about cars and digital products on Xiaohongshu, but the evaluations from various KOLs are mixed, making it impossible to judge the true situation of the products. For example, the movie "Pegasus 3" in 2026 received very poor reviews, so I chose to watch "Blade of the Immortal," but after watching it, I found it wasn't as good as everyone said, which put me in a dilemma of content discernment.
In this context, when users consult AI, their trust in AI is much higher than in KOLs. For most ordinary users, AI can provide a seemingly better option based on objective public facts, which is already sufficient. Therefore, we cannot resist the development trend of AI. Most users will directly feel the improvement in experience brought by AI, and once AI implants brand memory in users' minds, it will form a huge first-mover advantage for the brand that was first imprinted. At least for a period of time, this user will become a potential customer for the brand.
When a brand enters the user's repurchase stage, AI will continue to recommend that brand to the user following the previous recommendation logic, unless the user has an unsatisfactory experience with the brand and seeks a new one. From the 5A user conversion logic, as long as a brand completes the first four stages, subsequent repurchases depend on the product's inherent strength. For brand marketing and growth, completing the first four stages already means success. Everyone must pay attention to this, as it is a new brand growth model in the AI era.
Monica
You mentioned the need to deal with a massive number of search queries. So, in GEO operations, how much content does a brand need to produce to achieve a qualitative leap?
Bennett
Regarding the volume of content, there's no fixed standard. Different industries have different needs, and the core is to find the suitable content volume for your brand through testing. I can share a practical method that our team has been using. SEO and GEO practitioners should be familiar with the concept of landing pages. I wasn't familiar with these marketing concepts before, but I deduced and summarized similar methods through my own reasoning and understanding, and only later realized they were marketing landing page strategies.
Our core idea is: first, build the brand's knowledge base, then based on the knowledge base, guess the questions users might ask, generate corresponding search queries, and then produce content around these search queries. Through data monitoring, we determine which content brings AI traffic and which brings organic traffic. With traffic as the goal and content as the result, we deduce the form of content production from the results. This is our core logic.
The premise of this logic is that the brand possesses high-quality core content. If the brand lacks any high-quality, timely, and absolutely accurate content for indexing, then subsequent content production will fall into the trap of GEO poisoning, which is very fatal. Therefore, the early stage of core content creation must be heavy, and then continuously optimized through testing to ultimately find high-quality content that brings traffic to the brand. This is the real industry opportunity.
For brands, don't expect platforms to open up search data. Platforms will eventually convert this data into sales tools for commercial monetization. So, how should brands act? The core is to present their highest quality content based on two dimensions: "what the brand has" and "what the user needs."
"What the brand has" refers to the brand's various functions, such as Function A, Function B, Function C, etc. These functions serve as topic words, around which corresponding search queries are generated. Then, this content is structured with interlinking relationships. This is what I call a landing page, known as a hub-and-spoke model in marketing, where a central page is accompanied by multiple landing pages, forming a content matrix.
The core assessment dimensions for search engines are timeliness, authority, and text relevance. As long as we produce content around general search queries, and through the interlinking of landing pages, we can increase the authority of the web page, because search engines will consider interlinked content as high-quality content, rather than fragmented content published solely through blogs.
Every time we generate a new article, we aggregate it into the core landing page to ensure continuous updates, thereby meeting the three core assessment dimensions of search engines. After practical implementation, we found that this method is very effective. The proportion of ChatGPT crawler visits to these landing pages can reach over 80%, indicating that this content is very friendly to AI, while the average proportion of AI crawler visits for ordinary websites is less than 10%.
Through this method, both the SEO and GEO effects of the brand can be effectively improved. Therefore, I always believe that SEO and GEO are fundamentally interconnected. The core of both is the conversion from knowledge to traffic. As long as the content can bring traffic to the brand, it means the content complies with industry rules.
When I communicate with many brand marketing teams, I find that most marketing teams don't understand what knowledge assets are, which is a very frightening thing. I would ask them, what is the high-quality content that can bring traffic to the brand? Most teams cannot provide an answer, which indicates that they have not paid attention to the connection between content and traffic at all.
Therefore, the solution we provide to brands is primarily divided into four steps: First, crawl and manage brand knowledge to build a dedicated knowledge base. Second, generate content for the brand based on the knowledge base and conduct traffic tests. Third, aggregate the successfully tested content to continuously increase the content's authority. Fourth, continuously diagnose the differences in the quantity and quality of knowledge assets between the brand and its competitors for specific topic words, providing continuous content building suggestions for the brand.
We have built four core agent tools around these four steps, and the practical results are very good. This method is effective for GEO to some extent, but we found that Google provides higher exposure, directly pushing brand content to higher search rankings. Therefore, in my opinion, SEO and GEO are essentially the same thing; the core is the conversion from knowledge to traffic. As long as the content brings traffic, it means it complies with all industry rules.
Therefore, for those practitioners who repeatedly ask about the specific rules of SEO and GEO, I actually cannot give a specific answer. However, if we are discussing the conversion from knowledge to traffic, I suggest everyone must insist on producing high-quality content, find a suitable content direction through testing, and then continuously delve deeper. This is the core secret of brand growth.
Monica
In practice, what is the definition of a quality asset? Are there specific criteria or standards for evaluation?
Bennett
This is a core secret of our team, but I'll briefly share the core idea. First, we crawl the collected content and remove irrelevant advertising content. Our team processes up to 500,000 pieces of content per month, so the token consumption is very high. After initial content cleaning, we get two types of core content: the first is Markdown-formatted content, primarily high-quality text information; the second is JSON-formatted content, primarily raw web page information like description and meta tags.
Next, we extract summaries from this content, piece them together into complete long summaries, and label the content to build an enterprise content index. After indexing, we re-label the original content. The core of this step is to save tokens, as crudely labeling all original content would result in a large waste of tokens.
Through this method, a core effect can be achieved: every viewpoint forms a tree-like hierarchical structure, and every viewpoint can be mapped to the original content. The core is to solve the content ranking problem. After the index is built, we judge the timeliness of each piece of content, and then, based on customer needs, customize absolutely correct core content.
Based on the absolutely correct core content and timeliness standards, low-quality, duplicate, and outdated content is removed. At this point, multiple content categories are obtained, each with a corresponding content volume. For particularly important, fixed content, we perform knowledge fusion, meaning 10 related articles are integrated into 1, making the content semantically smoother and logically more coherent. This method is also our core method when training large models; we don't directly feed an entire book to the large model, but rather perform knowledge fusion first, then model training.
The entire content production process is collaboratively completed by four agent tools: the first agent is responsible for producing core content, the second agent is responsible for matching relevant images, the third agent is responsible for building URL and anchor text link relationships, and the fourth agent is responsible for comparing the content with the absolutely correct content and supplementing it. After completion, the QC stage is initiated to perform the final quality inspection of the content. The entire process consumes a very large number of tokens, but it can effectively guarantee the quality of the content.
After content production, we continuously monitor external and internal citations of the content, as well as the AI traffic feedback generated by the number of links under that topic, and the traffic data for each sub-page. After observing for a period, we eliminate content with poor traffic performance and test new content. Through such continuous iteration, the content meets the requirements for quality, authority, and timeliness simultaneously. After practical implementation, we found that this set of methods is very effective, and customer satisfaction is also high. However, the overall operation difficulty is relatively high, and we are currently gradually scaling up, and the agent tools are also under considerable pressure.
Monica
It sounds like you've invested a lot of effort and cost into creating high-quality content for brands.
Bennett
Yes, the core is to do a good job of knowledge cleaning and organization. Once this step is completed, a brand that wants to build its own Wikipedia has already completed 70-80% of the work.
Monica
So, essentially, doing GEO and SEO is no longer just simple website decoration, but rather requires rebuilding a brand's house from the ground up, starting with the underlying content infrastructure.
Bennett
Exactly, this is also the core reason why our operations are relatively heavy. In my opinion, only by building a solid underlying content infrastructure can subsequent traffic growth be achieved. However, many practitioners do not understand this; they simply modify the format of existing content and even question our logic for format optimization.
Therefore, from the outset, we did two very significant things for brands, which many clients initially didn't understand but later found to be highly effective: First, building knowledge assets. Although clients didn't have a clear concept of knowledge assets and couldn't directly see the content differences, the subsequent traffic conversion results made them acknowledge this work. Second, refined traffic monitoring. We implemented traffic segmentation monitoring at the CDN level, allowing us to clearly track traffic from different channels like ChatGPT, Gemini, and Google, as well as the traffic source and conversion situation for each page.
Through traffic monitoring, we can uncover many core issues: for example, some crawlers are for pre-training data, whose purpose is to crawl content for large model pre-training. We monitor the access patterns of these crawlers. Some crawlers are for AI search, and we track their access proportions. In practice, we found that the crawler access proportions vary significantly across different pages. This is how we continuously optimize the brand's content layout.
In short, we first perform very heavy traffic monitoring and knowledge management, then relatively light content production, and finally create high-conversion traffic cases for brands. This is our core operational approach.
Case Study 2 - Video Editing Software:
How a New Brand Achieves 0-to-1 Cold Start with Incremental Information and Multimodal Assets
Monica
Let's move on to the second case study, a video editing software.
Bennett
I will directly share the core content with you. In this case, we implemented a 0-to-1 cold start growth strategy. When we first started, I was also confused: how should a new industry and a new brand build knowledge assets? During the process, I discovered many industry pain points and tried many new strategies. Ultimately, I found that this set of strategies was consistent with my previous core logic for AI assistants.
The core of search is to solve two things: covering long-tail information and managing timeliness. General knowledge has basically been included in the pre-training data of large models. Therefore, to achieve a brand's cold start through search, the core is to create incremental information for the brand, which is also the core competitiveness of new brands and new products.
We also validated this logic through this case: when brand authority is insufficient, we can only exert effort in two dimensions: timeliness and incrementality. Therefore, for new products to do well in GEO and SEO, it is essential to extract unique selling points compared to competitors, use them as incremental information, build a content matrix around this incremental information, and then use the method I shared previously to produce high-quality content for testing. Through a reasonable content logical structure, we can capture traffic related to unique selling points.
Here's a practical tactic: prioritize Bing's traffic feedback, because Bing is more sensitive to content from low-authority brands. Through monitoring, we found that in Google's search results, content with high DR (Domain Rating) accounts for 70-80%, making it difficult for small businesses to gain exposure. However, in Bing's search results, content from low-authority brands can rank in the teens, which presents a huge industry opportunity for new brands and small businesses.
The second practical point is to create AI GEO video content. Multimodality is an important development trend for AI. For example, when asking questions in Doubao, Doubao will push relevant Douyin videos to users. This multimodal presentation can effectively improve user resolution and satisfaction. However, AI's current token consumption is already very high, and AI does not directly read video content; instead, it extracts video summaries and subtitles. Therefore, we can now provide a service for brands: directly converting a high-quality text article into a video script to create video content.
This service has shown good practical results. Although it cannot yet bring a huge increase in traffic volume, it can provide brands with a unique advantage in materials, which is crucial for new types of brands. For example, this video editing software competes with industry giants like Sora and CapCut International. Direct competition offers no opportunity; it can only achieve differentiated competition through unique selling points and unique content. This is the biggest leverage for a new brand's cold start. Finding users who value the brand's unique selling points through testing is the core takeaway from this case.
Monica
Did the brand provide you with its unique selling points?
Bennett
Yes, the brand will provide its unique selling points, and we will require the brand to commit to ensuring that these selling points have been implemented or are about to be launched. Otherwise, if AI-driven traffic enters the official website and finds no corresponding features, it will lead to a large number of user complaints. In the AI era, user behavior is infinitely amplified, because AI's core purpose is to cater to users and meet their needs.
If users complain or are dissatisfied with a brand, the brand loses not just an AI traffic entry point, but an entire growth strategy, because AI will attribute this negative feedback to the brand itself and subsequently reduce its recommendations for the brand. Therefore, brands must ensure the authenticity of their content and avoid deception, especially in overseas markets where brand authority is hard-won and must be cherished.
In the AI era, authority represents AI's trust in a brand. AI itself also has "inertia"; it will trust high-authority brands more and prioritize recommending their content. Therefore, after a brand provides its unique selling points, we first conduct small-scale tests. If we find that users are interested in that selling point, we immediately delve deeper into content and build a complete content system. This is very friendly for a brand's 0-to-1 cold start.
At the same time, you can try to seize material channels that have not yet been noticed by competitors, such as YouTube and TikTok. The content on these platforms has not yet been fully indexed by search engines, especially YouTube content. Creating content there yields very good results.
Monica
This case is quite unique. When you took over, the brand hadn't even built an official website, right?
Bennett
Yes, the timing was also critical. The brand signed the contract in October 2025, and on October 13th, ChatGPT updated its search strategy, directly invalidating the Reddit tactic I had prepared for the brand. Therefore, we had to return to the core idea of knowledge asset construction. However, this brand had no accumulated knowledge assets and was far behind its competitors, making direct knowledge asset competition impossible.
After extensive testing, we found that the only way to find opportunities was from the dimension of incremental information, which also stems from the perspective of knowledge assets: the brand achieved traffic cold start by seizing incremental information. This is the core key to this case.
Monica
If a brand doesn't even have an official website, it means its SEO foundation is zero. Can such a brand still directly do GEO?
Bennett
Of course. I've repeatedly emphasized that there's no clear distinction between SEO and GEO. The core of both is to provide high-quality content to users, and the emergence of AI has further blurred their boundaries. Therefore, brands don't need to operate according to the fixed concepts of SEO and GEO; they can simply focus on high-quality content.
Monica
This might be a misconception for many practitioners, who believe they need to spend a lot of time building their SEO foundation before doing GEO. It seems that's not the case; GEO can be started from day one of a brand's creation.
Bennett
Yes, through practical experience, we found that whether it's an established enterprise achieving dual growth in SEO and GEO traffic with this method, or a new brand's cold start, simply rewriting SEO content or directly publishing GEO-friendly content can yield good traffic results.
The core reason is that GEO itself places more emphasis on content quality, and its content structure is compatible with SEO. The logic of keywords is also just an expansion of more and larger search queries based on SEO. Therefore, the brand's core operation is simply to abandon the traditional crude content production methods of SEO and produce sufficiently high-quality, structured content for GEO. Essentially, it still revolves around high-quality content, and the core is to compete on content effectiveness.
Traditional SEO, based on Bert technology, can only perform simple content summarization, calculating text similarity and vector weights. Large models, however, possess reasoning capabilities and can judge the quality of content. Therefore, current brand growth no longer relies on traditional content structures. Although web-based content still needs to consider authority, the most crucial aspect is building content quality. As long as content quality is high enough, both SEO and GEO will bring traffic to the brand, because all platforms, at their core, have integrated large model capabilities and are transitioning towards AI.
Audience
If a new brand doesn't have an official website and directly does GEO, will its authority be very low?
Bennett
What is the most crucial way to increase a brand's authority? We need to consider a question: for a specific keyword, does increasing authority through a landing page or through a single blog post bring a greater boost? The answer is clearly a landing page. Therefore, if a large amount of authority converges on a core landing page, the search ranking for that keyword will quickly improve. This is also the most effective way to increase a brand's SKU authority, and we have validated this logic through practical experience.
Therefore, new brands don't need to overly focus on early-stage authority issues; they can directly focus on high-quality content and landing pages. First, find effective traffic growth methods through testing. Also, don't invest a lot of effort in building complex landing page systems early on, as you must first test high-quality content. Otherwise, a landing page built with significant effort might not bring any traffic, leading to wasted effort.
Our team's unique method is to use agent tools to improve the efficiency of landing page construction: a core topic page is paired with twenty to thirty related pages, building linking relationships, and a landing page for one topic can be completed in a week. We will continue to optimize the agent tools to enhance their processing capabilities and achieve batch landing page construction.
This method can effectively solve the content production efficiency problems faced by most enterprises, because traditional landing page construction involves long development times, and the difficulty of content aggregation and competitor analysis is high, making the process very complex. Our method can significantly shorten the cost and process of landing page construction, enabling rapid building. Brands can verify this through traffic data; AI traffic is extremely sensitive to high-quality content. Prioritize AI traffic feedback and continuously advance with this method. Essentially, it's a simple logic of testing and iterating around high-quality content.
Case Study 3 - 3D Generation Model Product:
Batch Construction of High-Authority Landing Pages and Efficient AI Crawler Conversion Loop
Monica
Next, let's move on to the third case study, a 3D generation model product.
Bennett
This 3D generation model product has already built relatively high-quality landing pages, and these landing pages have brought a certain amount of traffic to the brand. The brand recognized our methodology, so we tried to use the entire workflow to build landing pages and produce content in batches for the brand.
This week, we will complete the construction of the third topic's landing page. The initial process was quite challenging because, as a startup, our engineering capabilities were somewhat lacking, and we invested a lot of time in refining the style of the landing pages. Here's another difference in perception I'd like to share: if you truly want to achieve traffic growth, you must de-emphasize style and prioritize content.
However, this brand has high requirements for the style of its official website. Even when attaching relevant content at the bottom of the landing page, it demands aesthetic appeal, so we invested more time in refining the style. But the final effect was very significant; the landing page brought more traffic in just three days than pages the brand had been working on for a long time.
And the completion of landing page construction is not the end. If a landing page performs well in terms of traffic, new content needs to be continuously produced for updates, and more external links need to be built to continuously optimize and form a closed loop of traffic growth. Combining this case, I will summarize the core strategies for different growth stages of a brand:
First, in the 0-to-1 cold start phase, quickly test to find the brand's core landing pages and capture core topic words. After capturing core topic words, further capture keywords related to user groups and scenarios. For example, for this 3D generation model product, we created content such as "how gamers use it" and "how designers use it." This is because AI assistants, when making search recommendations, will personalize them based on user profiles and usage scenarios, which is the core logic of AI search.
However, much of the content from many brands still remains at the level of "what features does the brand have," while users are genuinely looking for "solutions for a specific group of people." This type of customized content is more likely to be recognized by users and AI. Therefore, in the 0-to-1 stage, the core is to test with high-quality content to find effective traffic growth points. Never test with low-quality content, as it will directly lead to a brand's traffic collapse, which is not worth it.
Second, in the 1-to-10 growth phase, once effective landing pages and content are identified, systematically build a content matrix, add weight to landing pages, and continuously operate, update, and maintain content. Third, in the 10-to-100 mature phase, if a brand wants to capture the TOP1 or TOP2 position in the industry, it needs to seize the highest traffic strongholds in the industry, create a large number of high-authority external links through high-quality content, discard low-authority external links, and even try to build high-authority content like Wikipedia.
At the same time, brands can try to cooperate with government agencies and universities. For example, software brands can offer free benefits to university students. University domain names have very high authority, which can effectively increase the brand's content authority. Through this method, landing pages can achieve higher search rankings for corresponding topic words. And since AI has a vast number of topic words, brands can continuously create new topic landing pages. By improving landing page rankings, the overall Dr authority of the brand's official website can be boosted, which is more efficient than traditional SEO.
In my opinion, once a brand's DR authority is improved through this method, combined with high-quality GEO content, it can seize the first-mover advantage in the AI era, and the brand's traffic growth trajectory will significantly increase. This is also the industry opportunity we see. We are currently validating this logic for this 3D generation model client, and have already completed the construction of landing pages for three topics, with the first topic showing very significant traffic results.
We monitored that ChatGPT's crawler visits to this landing page accounted for over 80%, which is an extremely exaggerated figure, as the average across the entire network is less than 10%. We will continue to observe the traffic performance of this landing page. At the same time, we found that registered users who entered the brand's official website through ChatGPT accounted for 10% of the brand's total registered users, whereas our previous best monitored data was only around 5%. Therefore, this brand has the opportunity to obtain more high-quality traffic from ChatGPT and other AI assistant platforms.
Of course, currently we have only built landing pages for three topics, so we cannot draw absolute conclusions. However, I always believe that the core logic of brand growth is very simple: what kind of content to produce and in what form to present it to bring traffic to the brand. This is the brand's best practice. In the AI era, the conversion from knowledge to traffic, and finding the content form that suits oneself, is the brand's true industry barrier.
And the various agent tools in between are merely means to improve efficiency, not core competitiveness. Brands must be clear about how many knowledge assets they possess and which knowledge assets are most sensitive to traffic. Focus on the result of knowledge-to-traffic conversion, without overly focusing on intermediate process indicators. This is the core point I want to share with everyone today.
If I were to start a product now and focus on brand growth, even if the company's business is not suitable for ToB, and there are many customers and high service pressure, I would still adhere to this core idea. I suggest that when starting a brand growth venture, you must clearly think through the core strategies for three stages:
In the 0-to-1 cold start phase, prioritize producing high-quality content. Even if a brand lacks its own content, it can research competitors and industry high-quality content to provide a foundation for AI-generated content. The generated content must be substantial, have clear logical relationships, and avoid fabricating false information. Even for科普 (science popularization) content, as long as it's high-quality, AI will recognize its value. But if you make up things that don't exist, AI will capture negative information, which will be detrimental to the brand. Brands should continuously test according to their product rhythm, find target audiences and keywords interested in the brand, and build content around unique selling points.
In the 1-to-10 growth phase, rapidly expand the quantity of content, and build it vertically rather than horizontally. It is crucial to create core landing pages and use them to capture core topic words. In the 10-to-100 mature phase, advance to higher-authority traffic strongholds, proactively publish high-quality content under specific topic words, implant brand memory in AI, and form a first-mover advantage.
When a brand's high-quality content enters the pre-training data of large models, the brand truly enters a competitive advantage stage against its competitors. Conversely, if a brand falls behind its competitors and there is no relevant information about the brand in the pre-training data, it will be extremely difficult to catch up later, even with significant investment in content.
Therefore, in the AI era, as a practitioner, I have a deep anxiety: if I were in charge of a brand's marketing team, I would be extremely concerned about first-mover advantage, because this advantage will be infinitely amplified and is crucial for the brand's long-term growth. That concludes the core sharing of this case.
Monica
Thank you very much for Bennett's sharing. Combining the three cases just shared, what commonalities do you think they have? And for different vertical sectors, what other considerations are there when doing GEO and SEO?
Bennett
The brands in these three cases are in vastly different growth stages: some are in the 0-to-1 cold start phase, some are in the 1-to-10 growth phase aiming for a top 5 industry position, and others are in the 10-to-100 mature phase aiming for the number one spot. However, these brands all share a common advantage: a deep understanding of AI.
Brands with a deep understanding of AI will pay more attention to the construction of knowledge assets and the refined operation of traffic. Their marketing teams are also relatively more professional and can understand the methodologies we share. The most basic point is that these brands have very well-structured tracking systems, complete data analysis systems, and continuously monitor traffic brought by AI, which is crucial.
Practitioners in the AI industry have already recognized this trend and begun to seize relevant traffic strongholds, but the real industry opportunities are actually in traditional fields. The official websites of many e-commerce brands are completely unprepared for AI, with the awareness and interest stages of the AIPL model entirely missing. They lack content design for cultivation, and it coincides with AI dominating the traffic landscape, continuously impacting traditional KOL cultivation models.
At this point, brands need to re-examine what the core value of their official website is in the AI era. If the official website merely serves as the final transaction landing page, then this function should be standardized and integrated with AI as soon as possible, leaving the conversion process to AI. If a brand wants to capture traffic and defend its official website as a core stronghold, it must transform as soon as possible, attempting to build a closed-loop service from the entire AI chain. This approach can significantly improve the brand's current SEO and GEO effects. This is my current observation, and it also confirms that most categories have not yet realized the value of AI; only AI-related categories are rapidly expanding. Therefore, I hope practitioners in other categories can communicate and discuss more, discovering their brand's strengths for deployment, as first-mover advantage is especially important in the AI era.
End-Game Thinking on GEO
Monica
How do you view the end-game of GEO? Will it lead to stronger industry monopolies in the future? Because if AI forms user cognition, it will generate stronger trust and loyalty for brands it first encounters, creating deeper user memories. Will new products entering the market later face greater competitive challenges as a result?
Bennett
Regarding the end-game thinking of GEO, I believe there are two core points. First, intermediary links in the industry will be compressed as much as possible. The system I am currently building is a Marketing Agent, and I'd like to use this live broadcast to break it down in depth. When such agents can complete marketing tasks at lower costs and higher efficiency, marketing practitioners need to develop a sense of crisis, as this will inevitably become an important direction for cost reduction and efficiency improvement in the industry.
Second, in the end-game industry trend, many intermediary agencies that provide specific services may gradually disappear. You can imagine the end-game form of GEO: AI cannot replace human content production capabilities; humans will continue to produce various types of content, but user needs will become increasingly personalized. The task of matching content production with user needs can only be completed by AI. So, the future industry relationship will be a model of human-AI collaboration and AI-human interaction.
The core opportunity for brands in the future lies in whether they can defend their core position and build exclusive agents to interface with various AI platforms. If a brand cannot complete this transformation, future personal AI assistants will not guide users to the brand's official website to place orders, and the brand will not be able to convey brand value to AI and users' personal assistants through its official website. Ultimately, it will be integrated by platforms like ChatGPT, becoming a basic service plugin in the platform ecosystem.
By then, brands will not need to build official websites or conduct marketing; they will only need to focus on product R&D and synchronize product information with the platform. The platform will screen brands through various means, requiring them to provide true and accurate product information. If the information is not up to standard, the brand will face severe penalties.
At this stage, the development direction of brands will diverge significantly: either they have the ability to build a category-specific content system and occupy a core industry position, or they can only become a basic service provider within the ecosystem. Of course, this end-game scenario involves many assumptions and cannot be precisely predicted at present; I am merely providing an initial explanation.
Therefore, marketing practitioners need to view this trend with a more positive mindset. The earlier a brand builds a high-quality content system and establishes its exclusive agent ecosystem, the greater its chances of success. If a brand chooses to respond passively, even if it is prepared to abandon its own stronghold, it still needs to actively embrace new ecosystem models, otherwise it will fall into an awkward, indecisive position.
Wanting to seize the industry traffic trend but not actively integrating into the new ecological stage is the most fatal problem. In my opinion, capable brands will definitely build exclusive agents to recommend the brand to various AI platforms through their own agents, conveying core brand advantages, such as core product selling points, a complete knowledge base, and core brand values, to achieve brand communication through an AI-to-AI model. Brands with insufficient capabilities will ultimately become a plugin in the MCP, only providing basic knowledge base services for the ecosystem.
Monica
This point is crucial. In the future, for brands to effectively interface with AI assistants, the best way is to build their own AI agents, allowing communication between similar entities.
Bennett
Exactly, information interaction between AI and AI is the only way to achieve full debate and verification of information. If AI's core goal is solely to cater to users, it will prioritize satisfying user needs even if it identifies false information. So, in the future, to convey true and accurate brand content, the AI-to-AI model will be a better solution.
Monica
I'm very happy to have had this exchange with Bennett. From the initial sharing of underlying logic, to specific case studies, and finally to the end-game thinking of the industry, many fresh perspectives have been brought. Lastly, Bennett, would you like to share a few more words with everyone?
Bennett
In the process of advancing related businesses recently, I found that many practitioners share a common feeling: doing anything in the AI era, the core logic actually boils down to first principles, which are very simple. All seemingly complex industry plays and operational processes, when dug deeper, reveal their core logic clearly.
The so-called growth methods and product development logic are all like this. When I first entered this field, I was also ignorant of much of the content, but I insisted on approaching work from first principles. AI is a field full of possibilities. Everyone can try to quickly implement their ideas through various vibe coding and agent tools.
As long as you are willing to tackle core problems thoroughly and meticulously, you can find many opportunities in this field. This field has also completely changed my perception. Initially, I just wanted to do something small and refined, but as the work progressed, I gradually discovered greater industry value, and the business has developed well. I hope to have more opportunities in the future to share fresh industry insights and practical experiences with everyone.
No comments yet.