Consent by Voice: What Smart Speakers Record and Who Agrees to It
The smart speaker is the only computer in the house that other people use without ever agreeing to anything. The account holder tapped through setup months ago; since then, family members, babysitters, and dinner guests have all spoken within range of a device designed to listen. What exactly it keeps, who reviews it, and how long it stays are set in an app most households opened once.
This article separates the mechanics from the myths: what happens after the wake word, what the retention and review settings really control, and the practical configuration that limits the exposure without unplugging the thing.
What the device actually keeps
In standby, the speaker listens locally for its wake word and keeps a rolling few seconds that are continuously overwritten. The change happens on a trigger: from that moment, audio streams to the provider's servers, is transcribed, and typically stored as a recording plus transcript tied to the account. False triggers happen, a similar sounding word, a TV in the background, and those clips get stored the same way. Purchases, messages, and smart home commands ride on that same pipeline. The point for consent purposes: the record being kept is not a summary. It is the audio itself, replayable in the app.
The review setting most people never see
Providers have at times used human reviewers to grade samples of recordings and improve recognition. Modern settings let the account holder opt out of human review and choose shorter retention, including auto deletion after a few months or not saving recordings at all. These options live several layers deep in the companion app, under privacy menus with reassuring names. Changing them does not noticeably degrade the assistant for most households, because the features they feed are aggregate improvements, not your personal service. Set retention to the shortest period you are comfortable with and turn off review participation; it is a five minute task with a real effect.
Guests, kids, and the consent gap
There is no banner for a kitchen. A guest cannot decline to be recorded by someone else's speaker in any meaningful interface sense; their only opt out is not speaking. Households handle this socially when they handle it at all. Children's voices raise the stakes: providers offer kid modes and stricter handling, but they are account settings too, not defaults a visiting parent can invoke. The honest position is that the account holder is making privacy decisions for everyone in earshot, and should make the obvious settings changes with that in mind, the same way they would lock a door that guests rely on.
A practical configuration pass
Open the companion app and do four things. Set recordings to auto delete on the shortest cycle offered, or delete the existing archive entirely. Turn off use of recordings for service improvement and human review. Review the list of enabled third party skills or actions, since each one is another party receiving voice data under its own terms, and remove the ones nobody uses. Finally, place speakers with the same care as cameras: a speaker in a guest room or a home office where calls happen deserves a second thought, and the mute button exists for the evenings when it matters.
Related reading
These earlier guides cover connected questions in more depth: Consent on the Big Screen: Tracking Choices in Connected TV Apps, Consent in Podcast and Audio Apps: Listening Data, Dynamic Ads, and Off-Site Measurement, and Consent in Games and Virtual Worlds: Ads Where There Is No Browser Banner.
Frequently asked questions
Is the speaker recording everything all the time?
No. It listens locally for the wake word using a short rolling buffer. Recording and cloud processing start at a trigger, real or mistaken. The stored archive in your account shows exactly what was kept, and browsing it once is more informative than any article.
Can I delete my voice history?
Yes, in the companion app or the provider's privacy pages, by date range or in full, and you can set automatic deletion going forward. Deleting history can slightly affect personalization like voice recognition tuning, but core functions keep working.
Do voice recordings get used for advertising?
Providers state that voice recordings are not used to target ads, though interactions like shopping requests naturally feed the commerce side of the account. The settings that matter for exposure are retention, human review, and third party skills, which is why the configuration pass focuses there.
What should I do about a speaker where children use it?
Enable the kid profile or child mode if offered, which tightens content and data handling, keep retention short, and remember the recordings are tied to your account for you to review and delete. A speaker in a child's bedroom warrants the shortest retention setting available.