How to Estimate the Real Numbers Behind App Rankings

Originally published in Chinese on March 18, 2024. Most rankings of AI tools in China rest on data collected through unofficial channels. If you want to collect it yourself, there are several routes. Third-party monitoring The first is web traffic. This is the main method behind today's AI tool rankings. The web is a mature delivery channel, and some third-party sites track page views and use them to produce a ranking. It is a reference value to a degree, but often only a reference. Many large companies resolve their services to one domain and carry traffic on different subdomains, and the traffic of a packaged front end does not follow the same logic as the traffic of the applications beneath it. Baidu's ERNIE Bot and its other services, for example, share one root domain, while different front-end interfaces point to different services. Domain monitoring alone can hardly tell which product or page a visit actually landed on, so I treat web traffic as a reference only. For services with a mobile app, the second route is tracking downloads and installs. Almost every app store shows download counts. You can crawl them and deduplicate by how often the app ships new versions. The deduplicated count is a usable figure for the app's user base. No single channel covers everyone, though: some users only register on the web version and some only download