Is the answer obtained when AI searches for your brand correct? Jianwei GEO develops AI search monitoring workbench.

When users ask AI about a certain type of product or service, brands need to know whether their names appear in the responses, whether the introductions are accurate, and which web pages the responses cite. Jianwei GEO takes these issues as the product entry point, and develops an AI search monitoring workstation for merchants and brand operation personnel, which integrates product materials, content review and response records in one place for operation personnel to check and retest.

GEO usually refers to Generative Engine Optimization. Jianwei GEO currently focuses on the material sorting and effect observation links: first clarify the facts publicly provided by the brand, then save the response evidence under different questions, to help users judge which part of the materials needs to be modified.

A single brand mention hardly explains the effect

In actual inspection, "the AI mentioned the brand" is not enough to prove that the brand has obtained an effective recommendation. The same response may only be a neutral introduction, or it may misstate the product price or applicable conditions. For operation personnel in charge of content and customer acquisition, in addition to paying attention to whether the brand appears, they also need to check the response content and citation sources.

Manual inspection can start from a spreadsheet. As the number of questions, platforms and detection time increases, the team needs to repeatedly sort out complete responses, label sources, and check the versions before and after modification. If only a few screenshots are saved, it is easy to miss the question conditions and samples that do not mention the brand, and it is difficult to compare the results before and after. Jianwei GEO hopes to incorporate this part of record work into the daily operation process.

The product is targeted at merchants and small and medium-sized brand operation teams that need to maintain product materials and continuously observe AI responses, and also provides a working space for service personnel who assist customers in sorting materials and retesting. Its specific focus task is to allow users to find problems in the responses and go back to the original records for inspection, rather than only getting an unexplainable score.

Associate product materials with response evidence

Users can manually enter or import product information in batches, supplement names, specifications, prices and public URLs, and record the sources for specific fields. The system generates product descriptions, frequently asked questions and structured web data based on these materials, and the content can only be exported after manual confirmation; after the product facts change, re-review is required. In this way, when checking AI responses later, there is a clear version of materials for comparison.

In the monitoring link, users build a question bank around the questions that customers may raise. The system collects responses through the configured official network search interface, saves the questions, platforms, models, detection time, original responses and citation links, and also supports manual entry of response evidence for consumer-end products. The two types of records are labeled separately to avoid directly taking API sampling as the results users see in the App.

Detection records are stored in association with the corresponding product material version. Operation personnel can look back at which pages a certain response cited, and then check against the price and specifications at that time; after the materials are modified, historical records can still be used for retest comparison. Interface adaptation for multiple model services has been incorporated into the product, and the actual available scope is subject to the permissions, model configuration and call acceptance of each account.

For example, if operation personnel want to check whether the AI misreports the price of a certain product, they can first save the product information with material sources, and then collect complete responses under related questions. The system gives rule prompts for some differences in prices and specifications, which are verified manually; operation personnel check the public materials accordingly, record the modified content, and then use the same questions for retesting. This is the usage process of the product, and does not mean that a certain customer growth result has been achieved.

Statistically, brand mentions, official website citations and manually judged recommendations are recorded separately. Sampling can be repeated under the same conditions, and the results show both the number of samples and the statistical interval; the recommendation rate is based on manually reviewed responses. The product cannot determine whether external platforms crawl, cite or recommend a certain brand, and the change in results before and after cannot alone prove that a certain content modification has produced an effect.

Starting from own brand application

Wang Shuai, founder of Jianwei GEO, once worked at ByteDance and Didi. At present, the project has been launched and put into operation, taking Jianwei GEO's own brand as the first application object to carry out material sorting, content release and search performance inspection.

In the practice of its own brand, the project has launched an official website, and published content on Zhihu and WeChat Official Accounts to introduce the AI search inspection method and specific problems in website deployment. These works provide public materials and inspection records for subsequent tracking; the inclusion, citation and recommendation after content release need to be verified separately.

The billing design of Jianwei GEO includes two methods: customers bringing their own model accounts and entrusted operation. The former means customers bear the cost of the model platform, and the project party charges software or service fees; the latter means the project party configures the interface, records service fees and detection usage separately, and sets the upper limit of call cost. The product has designed billing functions for account opening configuration fee, monthly service fee and detection fee, and the specific charging standard still needs to be determined in combination with the delivery scope and actual call cost.