Core conclusions:In this case, Shenzhen, a smart wearable company, took 4 months to increase the number of monthly inquiries from 10 to more than 40. The number of inquiries from AI sources accounted for 45%, the cost of obtaining customers was reduced by 50%, and 8 new overseas brand customers were added.
1. Customer background
A smart hardware company in Shenzhen, Guangdong Province focuses on smart wear, and its product line covers smart bracelets, smart watches, sports headphones and other forms. The company has a complete structure, software and algorithm team. The product iteration speed is among the fastest among similar factories. Almost all overseas revenue previously came from the Amazon platform, relying on the platform's natural traffic and advertising space to achieve scale.
After a certain volume is achieved, the ceiling begins to appear. Transactions on the platform are increasingly dependent on advertising bidding and prices. The same product is repeatedly compared on the page, and brand premiums cannot be achieved; more importantly, almost all those who come to the door from the platform are C-end consumers, and the overseas brand customers, chain channel operators, and regional agents that the company really wants to connect with are simply inaccessible in the platform system. Management decided to do something longer-term: build its own English-language website and content assets so that overseas customers can find themselves when proactively looking for suppliers.
2. Problem diagnosis: Why does the original path fail?
After taking over the project, we first made a round of diagnoses and concluded that the company's product capabilities are far stronger than its online expression capabilities, with three obstacles in between.
- Obstacles 1 are that the Independent Station is in name only.There are only a few product drawings and a company profile on the site. There are no technical parameter descriptions, no application scenario pages, and no information to download. Overseas customers and purchasing managers will close them after taking a look.
- Obstacles 2 are that the content is not accepted by the big model.Nowadays, more and more overseas purchases will first ask tools such as ChatGPT and Perplexity to "recommend several smart wear suppliers in China" before making formal inquiries. The names given to these models come from the web pages they have read and are willing to quote. The company has almost no technical content to quote on the entire network, so the model cannot mention it at all.
- Obstacles three, the page cannot be read by machines.The site has no structured data, no FAQ, and no documentation for AI crawlers. Product parameters are scattered in pictures and PDFs. Even if the big model captures the page, it cannot accurately determine what the company does and what specifications it can achieve.
Simply put, in the past, it used to compare prices on a "people-looking" platform, but now the decision-making entrance for overseas buyers is migrating to "asking AI". As the entrance changes, the writing of the content must also change.
3. What have we done?
- GEO optimizes complete deployments.Add llms.txt description files to the site, product and company structured data tags, and FAQ pages for high-frequency purchasing questions, allowing large models to quickly and accurately understand who the company is and what it can offer.
- Construction of AI-readable product content.Write down what originally existed only in the mind of engineers: what forms it can make, what sensor solutions it supports, what level of waterproof level and battery life, and what application scenarios it is suitable for. The content is written according to the standard that "the model can be directly quoted", with complete paragraphs, clear conclusions, and verifiable data.
- Reconstruction of English Independent Station.Reorganize the information architecture, put technical advantages and R & D capabilities in a conspicuous position, and supplement information such as certification, production lines, and proofing cycles that buyers really care about.
- Multi-platform content distribution.The same batch of technical content is rewritten into different forms and distributed to industry media, technical communities and Q & A platforms to expand the probability of being captured and cited by large models.
4. Execution rhythm
Month 1: Build the base
Complete site diagnosis, information architecture adjustment, structured data and llms.txt deployment, and rewrite the core product page. There are basically no inquiries this month, and the goal is to let machines understand the company first.
Month 2: Layout content
Concentrate on producing technical content and FAQ, covering the most commonly asked issues such as specifications, certification, minimum order quantity, and delivery dates, and simultaneously distribute them on multiple platforms. In this case, inquiries from AI sources began to appear from the second month.
Month 3: Calibration
Regularly ask questions about mainstream large models and record which questions mention customers and which ones do not. The question that has not been mentioned is the content gap, and the text is filled in according to the gap. This is the biggest difference between GEO and traditional SEO: the optimization object is the answer of the model, not the ranking in search results.
Month 4: Receiving inquiries
After the number of inquiries in the first three months increased, the question changed from "how to come from" to "how to receive them." We helped sort out the inquiry classification, information package, sample policy and English quotation template, so that the initial communication with overseas brand customers would not fall through.
V. Outcome data
- Monthly inquiry volume: fromof 10raised tomore than 40
- Proportion of AI source inquiries:45%
- New overseas brand customers:of 8
- Customer acquisition costs: Reduced in this case50%
- Inquiry structure: From focusing on scattered small orders to focusing on brand customers and channel providers
It should be noted that these numbers correspond to the industry in which the company is located, its original content base and investment pace. Changing to other categories, the cycle length and improvement rate will be different.
6. Which factories can this method be replicated to?
GEO's approach is most suitable for one type of factory: products have independently developed components and complex parameters, and customers need to "understand" before purchasing rather than "compare prices". Typical industries include consumer electronics, smart hardware, industrial sensors, instrumentation, energy storage and charging equipment, drone parts, precision hardware, etc.
Conversely, if products are not differentiated at all, customers only look at prices, or transactions rely heavily on platform price comparisons, then the leverage of content optimization alone is limited, and product positioning needs to be solved first.
7. Common questions
Q: Does doing GEO mean you don't have to do traditional SEO?
A: No, the two are superimposed. SEO solutions are "found in search results" and GEO solutions are "quoted and recommended by large models." The basic SEO skills such as site structure, page quality, and external links also affect the judgment of the big model, so they are usually done together.
Q: How long will it take to see the effect?
A: In this case, inquiries from AI sources began to appear in the second month, and the proportion stabilized at 45% in the fourth month. The length of the cycle depends on the original content foundation of the industry. If the site has almost no technical content before, it will take more time to prepare the process in the early stage.
Q: Can our own team do it?
A: You can do the content production part yourself, but someone needs to continuously monitor the answer changes of the large model and make directional adjustments. The difficulty is not in writing an article, but in judging "which question has not been covered." This requires tools and a stable execution rhythm.
