Why Alexa devices listen for a wake word and how to improve privacy
Alexa has to monitor nearby sound to recognise its wake word, but Amazon provides physical and account controls for limiting what is recorded.
Amazon Alexa devices are always listening in one narrow sense: they monitor nearby sound so they can recognise the wake word that starts a request. That differs from recording every conversation and sending it all to Amazon, although the distinction is easy to miss when a microphone is in your home. An Echo checks nearby audio for acoustic patterns matching its wake word. Until Alexa believes it has heard that trigger, Amazon says it is not meant to send the surrounding conversation to the cloud as an Alexa request.
The quickest privacy control
On most Echo devices, pressing the microphone button disables the microphones and turns on a red indicator. Amazon also provides account controls through the Alexa app: open the menu, select Alexa Privacy, then Manage Your Alexa Data and Voice Recordings and Transcripts. Depending on the device and account, recordings can be retained for three months, 18 months, until deletion or not at all. Users can review and delete voice recordings, smart-home history and detected-sound history. Some deletion-by-voice commands are unavailable on Alexa+ and Amazon Kids devices, while third-party products may retain their own histories.
The trade-off
There is no setting that lets an Echo hear its wake word without listening for sound. The meaningful choice is between convenience and control: leave the microphone active and manage recordings, or mute it whenever privacy matters more than instant access. Amazon's controls cannot make a connected microphone risk-free, but they give owners more visibility over stored audio and cloud access.
Our opinion
Always listening is technically accurate but incomplete. It describes the microphone's readiness, not proof that Alexa stores every word spoken in the room. That distinction should not erase privacy concerns: a system waiting for a trigger can still misinterpret speech, and users deserve obvious controls. The sensible middle ground is to use the microphone button as a real privacy switch, review recordings periodically and disable retention if personalisation is not worth the trade-off.