Starting September 4, Google begins removing Google Assistant from Android phones, tablets, and Wear OS watches — and replacing it with Gemini. Nobody gets asked. Nobody can switch back. After ten years, "Hey Google" gets a new brain.
The details, briefly: the rollout is gradual over the coming months, it covers Android phones, tablets, smartwatches, and Android Auto, and once a device is upgraded, Gemini is the only assistant on it. Smart speakers, smart displays, and Google TV aren't in this phase — they keep the old Assistant a while longer.
It's easy to file this under "consumer tech news" and move on. I'd argue it's something else: the largest forced onboarding of ordinary people onto generative AI that has ever happened. Every stat you've read about AI adoption so far described people who chose to open ChatGPT or Gemini. This is different. In September, hundreds of millions of people who never made that choice will start talking to an AI model every time they talk to their phone.
Why this matters for your business
The old Assistant was, underneath the voice, mostly a lookup machine. Ask it for a plumber and it matched keywords against business listings and read back names, ratings, and hours — fields from a database.
Gemini doesn't look things up so much as it composes an answer. It reads reviews and notices what they actually say. It reads websites and notices which ones explain, in plain language, what the business does, for whom, at what price. Then it recommends — usually one or two options, with reasons. Everyone else simply isn't in the answer.
Your customers aren't becoming AI users because they're curious. They're becoming AI users because their phone decided for them.
If you've been treating "customers asking AI about my business" as an early-adopter thing you could safely ignore for a year or two, this is the update: the early-adopter phase has an end date, and it's September 4.
Three checks worth doing before then
First, your Google Business Profile. Hours, services, service area, phone number. Boring, yes. But it's still a primary source the model draws from, and a wrong closing time in a confident AI answer is worse than a wrong closing time in a list of links — there's no second result to correct it.
Second, your reviews. A generative model doesn't just count stars; it reads. Ten reviews that say "great service" teach it almost nothing. One review that says "came out Saturday morning and fixed the panel in two hours" teaches it exactly when to recommend you. When you ask happy customers for reviews, ask them to mention what the job was.
Third, your website. If your prices, services, and availability live in an Instagram bio, a PDF, or text baked into images, the model can't reliably use them. A plain page that answers the questions customers actually ask — what you do, where, for how much, how fast — is now sales copy for two audiences: the human and the machine reading on the human's behalf.
The honest caveat
Voice queries are still a minority of how people find businesses, and the rollout will take months, not days. Nothing about your September revenue hinges on this. But the direction only points one way: the default interface in your customers' pockets is becoming a model that reads, reasons, and recommends. The businesses that are easy for it to understand get named. The rest get silence. The fixes above cost an afternoon, and they're the same fixes that help you everywhere else AI answers questions about you.