Mile Search: More answers, AI customer service agents from multiple enterprises respond in parallel

When search results only return a dozen or so enterprises and GEO only provides one or two, users are essentially playing a blind box guessing game. Milai Search aims to first expand the pool of available answers, then enable AI customer service representatives from multiple enterprises to respond simultaneously, turning the process of consulting each enterprise one by one into a single query.

By Milai Search Team

When a user searches for "packaging machinery", traditional search engines may return a dozen enterprises, while some AI search or GEO products only provide one or two. On the surface, the answers are more refined, but it is harder for users to judge: are these all the available enterprises? Do those not recommended simply not exist? In the end, users are not selecting enterprises, but choosing from the limited number of answers provided by the platform, which is more like playing a blind box guessing game.

Fewer answers mean fewer choices

Over the past decade, search engines and content platforms have gradually become the main entry for enterprises to acquire customers, but the right to allocate traffic has become increasingly concentrated. To be visible in search results, enterprises often need to face competition from paid search rankings, SEO, GEO and short video content at the same time. For a large number of small and medium-sized enterprises, the problem is not the lack of good products, but the lack of an affordable and sustainable display method.

Users have a more direct experience. Searching for a clear product or service is originally intended to see more real suppliers and then make comparisons among them. However, users often only get results of several to a dozen enterprises, and in the era of AI recommendation, sometimes only one or two are left. What users see is not the complete market, but a slice filtered by the platform.

According to data from the State Administration for Market Regulation, by the end of 2024, there were about 61.226 million registered enterprises nationwide, of which about 4 million had official websites and about 1.5 million were conducting search engine promotion. A large number of enterprises do not lack official websites, but their official websites have not become a stable customer acquisition entry.

"Users do not lack one answer, but enough real answers. Platforms should not make decisions for users, but put more enterprises in front of users," said Mi Ang, head of the Milai Search project.

Expand the pool of answers first

The first priority of Milai Search is to maximize the results of official websites, products and services. When a user enters "packaging machinery", the system first determines that the user is looking for manufacturers and products rather than principles or tutorials, and then accesses the Milai official website database; when searching for "the principle of packaging machinery", it will call general search capabilities such as DeepSeek and Yuanbao to return knowledge, tutorials and materials. Official website results and knowledge results are no longer mixed in the same list.

To ensure that the results are sufficiently accurate, Milai does not support fuzzy search, nor does it split keywords into synonyms for guessing. Instead, it requires a complete and clear expression. Keywords can be submitted manually by enterprises, or automatically analyzed and recommended by AI based on website content; each keyword corresponds to one official website page and one video, and users can switch between official website mode and video mode.

According to the project team, currently searching for "packaging machinery" can return the official websites of about 950 enterprises. Milai hopes that users can first see enough real enterprises and then decide which ones to focus on, instead of being narrowed down to a very small range by the platform at the very beginning.

In terms of sorting, Milai divides the results into five tiers from Z0 to Z4. The tier determines the general display interval, but within the same tier, the results are sorted randomly every time you refresh. Merchants compete for tiers and content, instead of getting fixed rankings by constantly raising prices. Milai stated that this is not a traditional paid search ranking, but an attempt to find a new balance between commercialization and result fairness.

Enable AI customer service of multiple enterprises to respond simultaneously

After the number of answers increases, a new problem arises: how can users quickly make choices among hundreds or thousands of enterprises? The next step for Milai is to connect the customer service capabilities of enterprises to the search. According to the plan, enterprises will submit their own customer service knowledge bases on Milai, including product parameters, frequently asked questions, service policies, quotation ranges and technical materials. When a user puts forward a demand, the system will broadcast the demand to the AI customer service of multiple enterprises to make them respond at the same time.

Subsequently, the system will organize these responses into a comparable table, presenting differences around dimensions such as price, specification, delivery period and service scope. Users can also continue to ask follow-up questions, and after several rounds of communication, narrow down the scope to several enterprises that better meet their needs. It is still the user who makes the final decision, and AI is responsible for compressing the process that originally required calling each enterprise one by one, recording and comparing repeatedly.

The most important part here is not "AI makes recommendations for users", but "AI parallelizes the responses from real customer service of enterprises". Every time a user puts forward a demand, Milai will switch to the persona of the AI customer service of each enterprise, and use the real customer service data of the enterprise to answer the user's question. Even without using Milai's AI, if users call the customer service of each enterprise one by one, the answers they finally get are basically the same. Milai just turns the process that originally takes several hours or days and requires dozens of calls into a single query and simultaneous comparison.

For standardized content such as product parameters, frequently asked questions and service policies, AI customer service can be more stable and complete than a single real person, and will not miss answers due to busy lines, staff replacement or memory deviation; complex problems can still be transferred to human agents. For enterprises, this is an upgrade of customer service capabilities; for users, this is an efficiency upgrade that turns "consulting one by one" into "consulting all at the same time".

This is different from the logic of GEO. GEO usually recommends a small number of merchants to users based on the content crawled by the platform; while Milai allows users to directly ask questions to the customer service knowledge bases of multiple enterprises, the answers come from the enterprises themselves, and the system is only responsible for sorting and comparing, reducing the hallucinations caused by single model generated recommendations.

In terms of business model, Milai is free for individual users to query official websites, products, services and consult AI customer service of multiple merchants in parallel; only when using general search such as DeepSeek and Yuanbao, search points will be charged according to the upstream cost. Merchant keywords start from the free Z0 tier, Z1 is 100 yuan per keyword per year, Z2 is 1000 yuan per keyword per year, Z3 is 10000 yuan per keyword per year, and Z4 is 100000 yuan per keyword per year. AI customer service is tentatively charged to merchants on a per-use or performance-based basis.

Milai is also building a "Traffic Sharing Alliance": alliance sites place a floating window entrance on the homepage to bring the originally lost general traffic into the search pool, and then distribute visitors to more matching merchants through keywords. After completing one of the two options: paying 600 yuan service fee or inviting 6 sites, members can get 10 million M promotion points every year.

According to the data provided by the team, as of November 2026, the Milai website database has included about 1.32 million websites, of which about 465,000 are normally searchable sites; official website search, AI general search, site backend and Traffic Sharing Alliance have all been launched, and the parallel response of AI customer service is still in the product development stage.

In terms of team, Mi Ang is a serial entrepreneur born in 2000. According to his introduction, he once operated the WeChat official account "Artificial Intelligence", founded Mici Academy to provide artificial intelligence training at the age of 18, and received an inquiry from the acquisition department of TAL half a year later, but the acquisition was terminated due to differences in valuation. He joined the State Grid in 2020 to engage in project R&D, developed a Go language quantitative system in 2021, and engaged in basic IoT operation from 2023 to 2025. At present, Milai Search is operated by Zhengzhou Duolan Network Technology Co., Ltd.

It is reported that Milai Search is launching its first round of financing, planning to raise about 10 million RMB and transfer 15% equity. The funds will be mainly used for product R&D, search technology and computing power investment, market promotion and alliance ecosystem construction.

"Milai does not want to select one enterprise for users, but to let users have enough real answers, and then use AI to speed up the comparison and screening process. It is still the user who makes the final decision," said Mi Ang.

(This article is a project contribution. The data in the article is provided by the Milai Search team, and the financing information is subject to the official disclosure of the company.)