你刚刷到这条消息，本来准备顺手划走，但又怕自己错过了真正会影响下一步判断的那一点。
最容易做错的，是Connect more of your apps to Search；代价往往是如果只盯表面热闹，你很容易在错误方向上花掉时间、预算和注意力。；我先给一个保守判断：Search命中率先吃metadata，不吃模型玄学。。
My conservative read is this: Search hit rate seems to respond to metadata first, not model mystique. By metadata here, I mean the labels and descriptions you give the app.
What the original message is actually saying is narrower than the headline. Across two Apps SDK guides, the same mechanism shows up: ChatGPT decides whether to call your app from the metadata you provide, not from some hidden ranking magic. In plain English, the first decision is often whether the app sounds like the right fit.
That is why three fields carry more weight than most teams give them: app name, description, and parameter docs. The optimize-metadata guide says those fields improve recall and reduce false triggers. The connect-chatgpt guide says the description is used during discovery, which is simply the stage where ChatGPT decides which app to try.
The second useful signal is process. OpenAI recommends a golden prompt set, meaning a fixed list of test prompts you can rerun. That turns connector selection from vibe-checking into something you can regression test.
A product update is worth your time not because it lists more features, but because it changes your next decision. For this one, the decision change is practical: if you are adding apps now, review metadata with the same seriousness as tool behavior.
Boundary: this is a docs-based read from the Apps SDK connect-chatgpt and optimize-metadata guides as of July 2026, not a production benchmark. It does not prove metadata outweighs tool boundaries or data quality. It does suggest that "Connect more of your apps to Search" should push more teams to audit naming, descriptions, and parameter docs before spending another week tuning the model.
Share this with anyone wiring apps into ChatGPT and still treating metadata as copy polish instead of routing logic.
真正该讨论的是：Connect more of your apps to Search