I used a voice assistant for most things for about three years — timers, messages, music, home controls, questions. I turned it off eighteen months ago and have not gone back.
The capability had improved over that period. That was not the issue.
What it was genuinely good at
Being fair first, because these are real and I do miss some of them.
Timers and alarms, which are the single most used function for most people and which voice handles better than any interface.
Hands-free operation while cooking, driving or carrying things.
Simple home controls, where saying it is faster than finding an application.
And accessibility, which is the strongest case for the whole category and is frequently treated as a footnote.
Where it consistently failed
Anything requiring more than one step. Compound requests were understood inconsistently, and the failure mode was to do part of it, which is worse than doing nothing.
Anything with names in it. Contacts, place names, band names and anything not in common vocabulary were transcribed wrongly often enough that I stopped trying.
Anything ambiguous, where it would guess rather than ask, and guess wrongly.
And anything where the answer mattered, because I could not tell whether a confident answer was correct.
The reason I actually stopped
Not the failures, which I had adapted to.
It was the accumulation of things it did that I had not asked for. Suggestions, prompts, notifications about what it could do, offers to help with things I was already doing.
Each was individually minor. Together they produced a device that was interjecting rather than responding, and I noticed I had started phrasing things carefully to avoid triggering it.
That is an odd position to be in with a tool. When I found myself editing my behaviour around a piece of software, it seemed reasonable to remove the software.
The listening question
Worth addressing plainly because it generates a lot of speculation.
Wake word detection runs locally on a small model that listens for a specific phrase and does not transmit anything until it triggers. That is the design and it is technically verifiable through network monitoring.
What does get transmitted is audio after a trigger, including false triggers, which happen regularly. Recordings following accidental activation have been retained and, in several documented cases, reviewed by humans for quality purposes.
Those practices came to light through reporting and led to policy changes at several companies, including opt-in requirements and clearer disclosure.
The reasonable position, as I understand it, is that continuous transmission is not happening, and accidental capture is real and more frequent than most people assume.
What I replaced it with
Less than I expected to need.
Timers on a physical device in the kitchen, which is faster than voice and does not require a phone.
Home controls through an application, which is slower and completely predictable.
And for the hands-free cases, mostly just doing it later, which turns out to be fine.
The one I genuinely miss is asking a factual question while doing something else, which is a real convenience with no good substitute.
Where I would still recommend one
I do not think my conclusion generalises, and there are cases where it is clearly the right tool.
Anybody with a visual impairment or limited mobility, for whom this is not a convenience but an interface.
Households where a shared device genuinely coordinates things.
Driving, where the alternative is worse in every respect.
And anybody who simply likes it, which is a sufficient reason and does not need defending.
The wider point
The interesting thing about the last eighteen months is how little I have missed.
Which suggests the assistant was solving problems it had partly created — the friction of getting a phone out to do trivial things, on a device designed to be picked up frequently.
Removing both the assistant and some of the reasons to pick up the phone addressed more than removing either would have.
That is a personal conclusion from a sample of one, and I mention it because I have heard several other people describe the same thing without connecting the two.
The children question
One area I have thought about since and have no settled view on.
Voice assistants are used heavily by children, who talk to them naturally and without the self-consciousness adults have.
Which raises questions about what is recorded, what is retained, and what the interaction teaches about how to ask for things.
Most platforms now offer child-specific modes with tighter retention and content controls, and they are off by default.
Worth finding and configuring, whichever conclusion you reach about the broader question.
Turning it off properly
Harder than it should be, which is itself informative.
Disabling the wake word is usually straightforward. Removing the assistant from the lock screen, from a dedicated hardware button, and from the shortcuts it has inserted elsewhere generally requires visiting several separate settings.
On some devices the hardware button cannot be reassigned at all.
Budget twenty minutes rather than two, and expect to find one or two entry points weeks later that you missed.