The surfaces multiply, the models consolidate: why "Apple visibility" now means "Gemini visibility"
Every week brings a new AI-search surface. A browser learns to shop. An assistant gets a new voice. A search box grows an answer. It is easy to treat each one as a separate place you need to "show up." But the more useful way to read this week's news is the opposite: **the number of places answers appear keeps going up, while the number of models actually composing those answers keeps going down.** For anyone trying to be recommended by AI, that consolidation is the story worth acting on.
The quiet fact under the loud headlines
The loudest headlines this week were OpenAI's: the company shipped its GPT-5.6 family three tiers, from a heavyweight "Sol" model down to a fast, cheap "Luna" - and launched ChatGPT Work, which pulls context from your connected apps and files to draft documents, spreadsheets and decks. That continues ChatGPT's steady drift from answer engine toward decision-and-work engine, and it matters. But it wasn't the most consequential change for how buyers actually encounter recommendations.
That title goes to Apple. Apple's next-generation Siri and Apple Intelligence are built on Apple's own foundation models plus a "deep collaboration" with Google: the companies confirmed that a custom Gemini model runs the heavier tier of Apple's AI. As Apple's new architecture makes clear, on-device tasks eg dictation, on-screen awareness, quick personal lookups etc. stay on Apple silicon, while world-knowledge and complex-reasoning questions route to a custom ~1.2-trillion-parameter Gemini model inside Apple's Private Cloud Compute. The deal, first reported in January, is worth roughly a billion dollars a year.
Why this is bigger than a licensing deal
Siri is not an app people choose to install. It's switched on by default on roughly a billion iPhones. It is, in practical terms, the only mass-market AI answer surface that requires no opt-in at all. And for exactly the class of query where a buyer asks "who's good at X?" or "what should I use for Y?", that default-on surface now leans on Gemini.
Layer that on top of a change Apple made earlier this year: as of June, Applebot-crawled content may be used to train Apple's models and generate Siri and Search answers unless a site explicitly opts out via Applebot-Extended and the nosnippet tag. So ordinary web content can already flow into Apple's answers with no action by the publisher, and the model reasoning over much of that content is Gemini.
The practical consequence is clean: for world-knowledge questions, "getting recommended on Apple" and "getting recommended on Gemini" are no longer independent problems. They overlap. A business that has done the work to be understood and recommended by Gemini has, for a meaningful slice of queries, also done part of the work for Siri, and vice versa. Treating them as two disconnected line items on a checklist overstates the work and misses the shared root.
The pattern this fits
This is not a one-off. Perplexity's browser agent runs on Claude and its "Model Council" blends several vendors' models into a single answer. Google's own AI Overviews and AI Mode run on Gemini across every query. Grok is the outlier that proves the rule. It's structurally different precisely because it grounds answers in live X posts that no other engine weights. Underneath the proliferating logos, a handful of frontier models are doing the actual retrieving and reasoning.
For measurement, that's freeing. You do not need a separate, panicked strategy for every new box with a chat cursor in it. You need to understand which model sits behind each surface, and whether that model understands and recommends you. The surface tells you where an answer appears; the model tells you why you're in it, or why you're not.
What to actually do
Two moves follow. First, map surfaces to models before you map budgets. Before treating Siri, Gemini, and AI Overviews as three campaigns, recognise how much retrieval they share, and don't pay three times to solve one problem. Second, the point we make every week because it keeps being the point, measure recommendation, not just citation.A dashboard that counts how often you're mentioned across ten surfaces can look busy while telling you nothing about whether a real buyer, asking a real question, was actually steered toward you. Put your buyers' questions to the models that matter, see whether you appear in the recommendation, and diagnose what's missing when you don't.
The surfaces will keep multiplying. The models behind them are consolidating. Optimise for the models, and you optimise for every surface they power, including the one that's already on, by default, in a billion pockets.
---
Clarity Digital Advisory helps companies get recommended, not just mentioned, by the AI platforms buyers now use to shortlist vendors.