OpenAI's Apple breakup is a distribution lesson for everyone else

OpenAI took what one executive called a "leap of faith" on Apple and now Ars Technica reports the company is "actively working with an outside legal firm" about it. The complaint isn't that the integration doesn't exist. It's that Apple made users "specifically invoke the word 'ChatGPT'" to trigger it, and wrapped the outputs in small windows users can ignore. Distribution without invocation is not distribution.

That's the part worth sitting with if you're trying to be cited inside AI answers.

Invocation is the whole game

The OpenAI-Apple story is structurally identical to the brand-in-ChatGPT story. You can be technically present and effectively invisible. ChatGPT can "know" your company exists, have your docs in its training data, even retrieve your site during a search query, and still surface a competitor because the user's phrasing didn't pull your name into the answer. Apple required users to say "ChatGPT" out loud. ChatGPT requires users to ask a question your content actually answers, in language the model maps to your brand.

Most brands optimize for the first half (existing in the index) and ignore the second half (being the obvious retrieval target for a specific phrasing). That's the same mistake Apple made for OpenAI.

What this looks like in practice

Pull the five queries you most want to win in your category. Not branded queries. The unbranded ones, the ones a buyer would actually type before they've heard of you. For each, ask ChatGPT cold, in a fresh session. If your brand doesn't come up in the first answer, look at what does, and read the language in those cited sources. The cited pages almost always contain the exact phrasing of the question somewhere in the body or a heading. Your pages probably contain your own phrasing of the question, which is not the same thing.

This week: rewrite the H1 and first paragraph of your top three category pages so they contain the literal phrasing a buyer would use, not the phrasing your product marketing team prefers. "Best [category] for [use case]" beats "The [Brand] Platform for Modern Teams" every time inside a retrieval pipeline. It also sounds worse on a billboard, which is why most teams won't do it.

The part I'm less sure about

I think the invocation problem gets worse, not better, as ChatGPT moves toward agent-style answers where the model picks one source and synthesizes, but I could see it going the other way if OpenAI ships a verified brand layer that lets companies claim category authority directly (the Bloomberg sourcing hints at OpenAI building its own device, which suggests they're done waiting for distribution partners to do the work). Either path makes the current state of "hope ChatGPT picks you" temporary. Neither makes it less important right now.

The footnote

So the same buyer who can't find you in ChatGPT will soon be running the same query through Claude and Gemini too, with different retrieval logic in each.