Google’s new Gemini assistant is here, but it still confuses 5 AM with 5 PM

Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.

Google’s shiny new Gemini voice assistant has officially landed on a handful of smart speakers, and according to Google, it’s going great! 

In a celebratory blog post marking two weeks since rollout, the company even quoted one very enthusiastic user: “My Google Home just got upgraded with Gemini. It’s actually useful now beyond asking for today’s weather.” 

But a quick scroll down to the FAQ paints a less magical picture. 

Among the top complaints: Gemini sometimes misunderstands basic commands, like confusing “set an alarm for 5PM” with “wake me up at 5AM,” and struggles to control smart home devices. 

Which is awkward, considering that “understand what I say” and “turn on the lights” are kind of the two main jobs for a voice assistant.

Gemini is the flashy centerpiece of Google’s “Gemini for Home” overhaul, which infuses AI across the Google Home ecosystem. 

The Home app now includes an “Ask Home” chatbot that lets you type commands in plain English, and Nest cameras can generate daily summaries of what they’ve seen (because who doesn’t love an AI surveillance diary?).

Getting Gemini, however, isn’t simple. 

You can’t just update your speaker. You have to join an “Early Access” program that’s different from Google’s “Public Preview” program, which is different from nobody really knows. 

Even The Verge doesn’t have it yet, and Reddit threads suggest most users are still waiting in line. 

Google says the full rollout won’t happen until next spring, possibly alongside a new Google Home speaker.

When it does arrive, Gemini promises to be more conversational and context-aware. 

In theory, you’ll be able to say things like, “Hey Google, lock the door, turn off the lights except in the kitchen, and make it warmer,” and it’ll just work. 

In reality? Early testers say chaining commands together is hit-or-miss, or just miss.

As Google Home’s product lead admitted, large language models are creative but not great at repeating the same tasks reliably. 

Gemini might tell you a bedtime story, but don’t count on it to set your alarm correctly.