It's covered a little farther down in the article.
It's optional, and the user's permission to send a request to ChatGPT is required every time.
> In most cases, LLM-related features built into iOS and macOS use Apple's own models, which live under the Apple Intelligence branding umbrella. But it gives users the choice of referring a prompt to ChatGPT on a case-by-case basis when the prompt is outside the scope of what Apple's models are designed for.
It's optional, and the user's permission to send a request to ChatGPT is required every time.
> In most cases, LLM-related features built into iOS and macOS use Apple's own models, which live under the Apple Intelligence branding umbrella. But it gives users the choice of referring a prompt to ChatGPT on a case-by-case basis when the prompt is outside the scope of what Apple's models are designed for.