Siri runs on Gemini: what it means for brand visibility

Siri now runs on Gemini. That single fact, reported by The Decoder, redraws the citation map for roughly two billion active Apple devices and makes Google's retrieval preferences the de facto standard for a second major consumer AI channel, alongside Google's own products.

Apple's rebuilt "Siri AI" ships with Gemini models at its core, processing queries partly on-device and partly through Apple's Private Cloud Compute infrastructure. Early testers report genuine capability gains: multi-step requests work, screen context is read, workflows chain together without friction. Hallucinations and gaps in personal context remain, but those are table stakes complaints for every deployed LLM. The structural fact is more consequential than the bugs: the model that decides what sources Siri surfaces, which brands it names, and whose content it synthesises is now Google's.

The consolidation nobody voted for

Brand visibility in AI answers has never been a single, unified game. Until now, a communications team could think of ChatGPT (OpenAI's retrieval logic), Gemini (Google's), Perplexity (its own index and summarisation chain), and Siri (Apple's, largely opaque) as four distinct audiences requiring slightly different content strategies. That segmentation is collapsing. With Siri running on Gemini, a brand that does not appear in Google's model responses does not appear to Siri users either. Two surfaces, one model, one set of editorial preferences baked into training and retrieval.

For senior communicators at multilaterals, industrial groups, and financial institutions, the practical consequence is that whatever signals Gemini uses to decide that a source is citable, credible, and relevant now govern an enormous additional slice of queries. If HOLCIM or the World Bank Group or Adecco appears confidently in Gemini responses, that authority propagates through Siri. If it does not appear there, a revamped Siri will not rescue it.

The EU carve-out is a complication, not a reprieve. Regulatory friction has kept Siri AI unavailable in European markets for now, which means EU-headquartered institutions face an asymmetric moment: their counterparts in North America and Asia-Pacific are already being answered by a Gemini-backed Siri, while EU audiences still receive older, weaker responses. That gap will close. The EU's regulatory machinery moves slowly; Apple and Google's commercial incentives do not.

What Gemini's citation logic actually rewards

The working model of how Gemini surfaces sources in generative answers points to three factors: presence in authoritative, crawlable web content; structured data and clear entity signals that allow the model to identify an organisation unambiguously; and earned citation in high-trust third-party publications. None of these are new. What is new is their leverage. A single content and authority strategy now reaches Google Search AI Overviews, Gemini's own assistant, and Siri AI simultaneously. The multiplier effect of getting it right has grown; so has the cost of getting it wrong.

Financial services brands face particular exposure. Compliance constraints often mean thin, carefully hedged web content that is technically accurate but offers Gemini little to synthesise into a confident answer. Institutions in that sector that have not yet invested in substantive, policy-grade thought leadership now have a second, urgent reason to do so: the content that earns Gemini citations is the same content that reaches Siri's 2bn-device footprint.

Philanthropic and policy institutions should note the corollary. Bodies like UNDRR or CGAP, whose authority rests on primary data and technical expertise, are naturally well-positioned if that content is published in crawlable, citable form. The risk is that content locked in PDFs, behind login walls, or distributed only through internal channels is invisible to the model regardless of its quality.

Apple's announcement is a commercial deal between two companies. Its effect on the information environment is considerably larger than that framing suggests. Brands that treat AI visibility as a single-channel problem, solved by optimising for one model, will find that the other channels have quietly adopted the same judge.