
Not deprecating it. Not reducing its features. Removing it. Google emailed users to say Assistant is being discontinued on compatible Android phones and tablets, with Wear OS devices, headphones, and Android Auto following. The rollout runs over several weeks, and once a device switches, there is no going back. Gemini becomes the assistant experience on Android. Assistant launched in 2016; ten years later, the product that most "voice search optimization" advice was written about is being decommissioned while that advice sits unrevised on hundreds of agency blogs. This matters more than a product retirement usually would, because Google Assistant was not incidental to voice search SEO. It was the mechanism. The specific technical recommendations that circulated for the better part of a decade — mark up your content with speakable structured data, add FAQPage schema to win the spoken answer, optimise for the featured snippet Assistant reads aloud — were all recommendations about how one product retrieved and read one answer. Two of those three recommendations no longer apply to anyone. The third applies to a product that is being turned off. So this is a reasonable moment to rebuild the topic from the ground up. What follows is not a refresh of the standard voice search checklist. It is an attempt to establish what is actually true about spoken and conversational search in 2026, what a Canadian business can genuinely act on, what cannot be measured, and where the real constraints sit. Some of those constraints turn out to be in your hosting infrastructure rather than your content, which is not where most of this conversation usually goes. A warning about the rest of this article: it will contradict a great deal of what you have read on this subject. Where it does, the reasoning and the source are given so you can check. What "voice search" actually means in 2026 The phrase has always been sloppy, and the sloppiness is the root of most bad advice about it. Three genuinely different things get bundled under one label, and they behave differently enough that treating them as one topic guarantees you optimise for the wrong one. Path one: voice...













