The GPT Store Has No Distribution Barrier, So the Fight Moves to User Data

Jiang Wu · 2024-02-21

*Originally published in Chinese on February 21, 2024.*

OpenAI opened the GPT Store, its marketplace for custom GPTs, on January 10, 2024. A month and a half on, the question is how a business around it would work.

Decentralized, but platform-centralized

A few basic points need to be settled first. For now the store is decentralized, not centralized, and it is not a closed system.

Part of its traffic comes from front-end traffic, and front-end traffic is not necessarily controllable. So the centralization turns into centralization at the platform level, and that platform is not entirely in OpenAI's hands. In the end the key question is where the users are, because that is where the product will appear. It might be a WeChat mini program, or the mini programs of various other platforms, which are another form of light app.

What decentralization means here

It is decentralized because it has no distribution barrier of its own. It can be a web page, a mini program, or anything that opens in a browser. With no barrier to use, all you need is a front-end interface, and that interface can sit anywhere users can reach: a chat box, a mini program, a WeChat link, a WeChat official account, anything. That is lighter than the old app store. An app store is tied to a device to some degree, and distributing through it still makes demands on the app's own attributes and structure. If no app structure is needed and an interaction interface alone can distribute the service, that is a big convenience for the user. For the platform it creates a chance that the entry point gets split.

The front-end entry and where traffic comes from

Once decentralization is defined, the next question is where that decentralized front-end entry comes from. That would be a new logic of traffic distribution. Traffic goes where the users are, and whoever can get traffic can get users. It is a chicken-and-egg problem. Underneath, it comes down to how an app meets the user's needs or scenario and pulls the user onto its own platform. On this point Chinese companies have some advantage, since they are skilled at playing traffic in many ways. They may simply not have thought that the traffic could be turned around, with a GPT Store behind it taking part in operations or sharing revenue.

What the competition turns on

The logic of redistributing traffic has not changed in essence. Whoever has users at the front end, and the ability to get traffic, can then complete service delivery at the back end. OpenAI has no absolute advantage here. Or rather its absolute advantage does not come from the GPT Store itself but from holding the user data at the front-end entry.

For example, if a user puts their information into OpenAI's account system and the services that follow are built on that user data, accuracy for the user and the value of front-end input will both be higher.

The reverse also works. Whoever holds enough user data today can attach a similar service at the back end, and users get the same quality of service without moving their data. That logic holds for now.

The core barrier in future is who keeps more user data on their own platform. Say a user stores past travel information and past writing on platform A. Services built on that information keep creating value on A, and the user stays and keeps using your service.

Stages ahead

For this kind of competition, the short-term question is whose cost of acquiring traffic and reaching users is lower. In the medium term it is whose products are good enough, varied enough, and closer to what users need. In the long term it is who can store more front-end user data. If the cost of switching is high, users stay on the platform.

Work down from that logic and you can roughly see where the key competitive factors and value points sit for OpenAI, for Poe, and for everyone building their own app store on a model system. What each should do next follows a path you can trace.

Home · All articles · Topics · 简体中文