Smart speaker privacy: what they're actually listening for

In one line: The microphone isn't the scary part. Retention, human review and a quiet 2025 policy change are — and all three have settings you should check today.

Every smart speaker conversation eventually hits the same question: is it always listening? The honest answer has two parts — one reassuring, one that deserves your attention. The microphone is designed to wait locally for a wake word ("Alexa", "Hey Google") and only starts sending audio anywhere after it hears one. That part is the standard design and it's the reassuring part. The attention-deserving part is what happens to the audio after the wake word: where it goes, how long it stays, and who else can hear it.

The part people didn't know about

In 2019 it emerged that all the major assistants — Amazon, Google, Apple — had human reviewers listening to some fraction of recorded voice commands, to grade how well the software handled them. The BBC reported Amazon's response at the time: reviewers annotated what Amazon called "an extremely small sample of Alexa voice recordings", with safeguards and a "zero tolerance policy" for abuse. The scandal wasn't only that humans listened — it was that the terms never clearly said so, and opting out wasn't obvious. Every major vendor then shipped real controls: opt-outs from human grading, deletion tools, auto-deletion timers. Apple went furthest and made human grading opt-in with iOS 13.2 later that year.

What quietly changed in 2025

Here's the development most owners missed. Until spring 2025, Alexa offered a setting called "Do Not Send Voice Recordings" — device-side processing, your audio never leaving the speaker. Amazon emailed customers in March 2025 that this option would stop working on March 28, 2025. Amazon's stated reason, quoted by Malwarebytes: "As we continue to expand Alexa's capabilities with generative AI features that rely on the processing power of Amazon's secure cloud, we have decided to no longer support this feature." The change applies to US customers on certain newer devices (the fourth-generation Echo Dot, Echo Show 10 and Echo Show 15, set to English); Amazon says voice recordings are now deleted after processing and that you can still stop recordings from being saved. The practical takeaway stands regardless of device: cloud processing of your voice is now the default path on newer hardware, and the choice left to you is about storage and improvement programs, not transmission.

The settings that actually matter

On Alexa (in the app: Settings, then Alexa Privacy, then Manage Your Alexa Data): turn off the option to help improve Amazon services using your voice recordings; set recordings to auto-delete after three or eighteen months — or choose "don't save recordings"; and enable voice deletion so you can say "Alexa, delete what I just said" or "delete everything I said today".

On Google (myactivity.google.com, under Web & App Activity): the key switch is "include voice and audio recordings" — unchecking it means your audio isn't retained for review at all. Set auto-delete to three or eighteen months while you're there, and review what's already stored.

On Apple (Settings, Privacy & Security, Analytics & Improvements): "Improve Siri & Dictation" is the human-grading toggle — off means your recordings aren't used for grading. Since it's opt-in by default, this one mostly matters to confirm it stayed off.

And the most reliable control ever shipped: the physical mute button on the device itself. When company is over, or the conversation isn't the speaker's business, a hardware mute beats every software promise.

What the settings don't fix

Third-party skills and apps are governed by their own developers, not by your assistant's privacy dashboard — a setting that limits Amazon's retention says nothing about what a quiz skill does with your answers. Treat skills like phone apps: check what permissions they actually need, and delete the ones you don't use.

Shared spaces are the under-discussed risk. A speaker in a kitchen hears everyone who visits, including people who never agreed to Amazon's terms. Guest mode (on Google) and hardware mute are the polite-host answers; the bedroom is a defensible no-speaker zone.

Your voice data is data. The same logic that applies to free apps applies here — the data a service collects is part of the price, and it's worth deciding consciously what you're paying.

None of this makes smart speakers indefensible. It makes them a trade: convenience for data, with the terms set by the vendor and adjustable — within limits — by you. Five minutes in the privacy settings is the difference between accepting that trade knowingly and accepting it by default.

Sources

All of Smart home All of BRYME Tech