It always turns out to be a feed

The fastest way to understand a company's real business is to watch what it builds after the exciting idea dies.

OpenAI spent a year talking about conversational shopping: buy things inside chat, skip the website, check out instantly without ever leaving the conversation. It was framed as a clean break from Google and traditional e-commerce.

That version didn't last. Instant checkout was shut down in March.

What's showing up instead is more familiar. Product data is now being pulled into an ads system. There's a structure where each item carries its own listing, performance metrics, and advertising signals. Feeds provide the raw input, and that input gets turned into ads.

If that feels familiar, it's because it is. It looks a lot like Google Merchant Center.

So the company that was supposed to reinvent search is, in practice, rebuilding the infrastructure that powers shopping ads on Google. Not the ranking algorithms or the interface layer, but the less visible plumbing that actually makes commerce work: structured product feeds that map items to queries.

That's not necessarily a contradiction. It's a pattern. Every platform that tries to monetize shopping ends up in the same place. Interfaces can change dramatically, but the underlying requirement doesn't: you still need clean, structured product data before anything else can happen.

The interface battles get the attention, ChatGPT versus Gemini versus Rufus, but under all that, the same system keeps reappearing. Once you introduce ads or transactions, you need something to sell, something structured, something measurable. That “something” is always a feed.

So the real signal isn't in the announcement about how people will shop next. It's in what gets built quietly afterward. And more often than not, it turns out to be a feed.