#
With the widespread adoption of large AI models, the way users interact with information is undergoing a profound transformation. Generative Engine Optimization (GEO) has emerged in response, with its core objective being to optimize content to be more easily retrieved, understood, and cited by AI models, thereby gaining effective exposure when users ask questions. Unlike traditional paths that rely on expensive service fees, a preliminary GEO strategy can be established and validated through systematic manual operations. This article breaks down the process into four core stages.
Phase 1: Mining User Intent and Building a Query Framework
The starting point of any GEO strategy is to accurately understand the questions target users will ask AI in different scenarios. The key task in this stage is to build a comprehensive ‘query matrix’.
Defining Core Category Terms: First, it’s necessary to clarify the market designations for the product, including professional terminology, common names, and even slang. For example, ‘new energy vehicle’ versus ‘electric car’.
Discovering Real User Questions: Centering on category terms, you can collect real user questions through various channels. For instance, use the autocomplete suggestions in the search boxes of major search engines or content platforms (like Baidu, Douyin, and Xiaohongshu) to analyze the specific questions users are searching for. Additionally, a company’s internal customer service records are a valuable source for identifying frequently asked questions.
Categorizing by Decision Stage: The collected questions can be classified according to the user’s purchase decision funnel, which helps in prioritizing subsequent content strategy. Common categories include:
- Purchase Recommendations: e.g., ‘Which are the best new energy vehicles for home use?’
- Specific Scenarios: e.g., ’What’s a good electric car for long-distance commuting?'
- Price and Budget: e.g., ‘What kind of new energy vehicle can I buy with a budget of around 200,000 RMB?’
- Reputation and Reviews: e.g., ‘Is the Tesla Model Y worth buying?’
In the initial stages, it’s advisable to select a broad, general purchase-related question as a pilot case to run through the entire process of observation, implementation, and validation, thereby establishing a foundational understanding of the methodology.
Phase 2: Analyzing Source Preferences Across Multiple AI Platforms
Different large language models exhibit preferences for different information sources due to variations in their training data and Retrieval-Augmented Generation (RAG) strategies. Identifying these core sources is key to ensuring your content effectively reaches the AI.
The specific operation involves inputting the questions selected in the previous stage into multiple major AI platforms for testing, such as Doubao, Yuanbao, DeepSeek, Kimi, Tongyi Qianwen, and ERNIE Bot. To reduce the impact of randomness, it is recommended to ask the same question at least 3 times a day in a new chat window and observe continuously for 5 to 7 days to accumulate a sufficient sample size.
Key observation points include:
- Source Channels: Record the source websites cited in the AI’s answers, such as WeChat Official Accounts, Toutiao, Zhihu, Baijiahao, Sohu, as well as various vertical media or official websites. Frequently appearing channels are your core sources.
- Content Characteristics: Analyze the type (e.g., reviews, rankings, news reports, tutorials), format (text with images or video), and the title and content structure of the cited articles to inform future content creation.
- Brand Exposure: Check if your own brand is mentioned in the answers. If it already appears, you can list the corresponding question as an observation item and temporarily hold off on content deployment.
According to public industry research, different models exhibit certain source tendencies. For example, Doubao heavily relies on the ByteDance content ecosystem, Yuanbao prefers WeChat Official Accounts, and DeepSeek tends to favor comprehensive news platforms. However, these preferences are not static and require continuous empirical observation.
Phase 3: Creating and Structurally Publishing Content Assets
After identifying the core source channels and the characteristics of highly-cited content, you can proceed to the content creation and publication stage. The goal here is to produce ‘AI-friendly’ content.
Account Preparation: Prioritize creating accounts on the core source platforms discovered in the previous stage, choosing channels that support free registration for individuals or businesses, such as Toutiao, WeChat Official Accounts, Zhihu, and Baijiahao. At the same time, the content on your brand’s official website (e.g., product introductions, technical white papers) should not be neglected, as it serves as an authoritative source.
Content Creation Strategy:
- Content Direction: Creation should focus on answering real user questions rather than traditional brand promotion. You can analyze the AI’s Chain of Thought (CoT) when generating answers, extract keywords, and naturally incorporate them into article titles and body text.
- Content Type: Practical content such as rankings, comparative reviews, and tutorials are generally more easily adopted by AI as neutral sources than purely promotional brand content.
- Content Structure: Use a structured format that is easy for AI to parse, such as organizing information with clear subheadings, bullet points, numbered lists, and tables. This helps the AI model accurately extract key data.
- Publication and Distribution: Initially, you can adopt a multi-platform distribution strategy, publishing the same piece of content across multiple registered platforms. After accumulating some data, focus your resources on the most effective channels based on their actual citation performance to optimize distribution efficiency. Maintaining a consistent update frequency, such as daily updates during the launch phase, helps with faster indexing and inclusion by AI.
Phase 4: Performance Verification and Iterative Optimization
Content publication is not the end point. Continuous monitoring and optimization are essential to ensure the effectiveness of your GEO efforts. This stage forms a feedback loop.
After publishing content, you should use the same query phrases to check the various AI platforms daily, focusing on tracking two core metrics:
Content Citation Status: Check if the links to your published content appear in the AI’s list of citation sources. If cited, it proves that your current content strategy and channel selection are effective.
Brand Mention Status: Observe whether your brand name or product appears in the body of the AI-generated response, as well as its ranking and the surrounding context.
If your content is not being cited, you need to review and check its quality, degree of structure, and whether the publishing platform needs adjustment. Through this cycle of ‘hypothesize-execute-verify-optimize,’ you can gradually find a replicable GEO success model suitable for your industry and brand.
In summary, while this manual process faces efficiency bottlenecks when dealing with a large volume of keywords, it is a completely viable path for brands to explore the principles of GEO and validate the effectiveness of their strategies at zero cost in the initial stages. It provides a basis for deciding whether to invest resources in scaled-up operations later on.
