Trace your commute

Voice assistants

How voice assistants handle audio and transcripts

Wake detection, on-device processing, retained transcripts and optional improvement programs are separate questions. Read the current service and setting instead of treating every voice interaction alike.

Latest source check .

Conceptual illustration

Activation and cloud recording are different stages

Amazon describes wake-word or button activation before ordinary Alexa audio streams to its cloud, with false activations possible. Follow Up Mode can allow further requests without repeating the wake word. Its indicators and history help explain what a device processed; this is the company's description, not proof of a particular room's activity. Amazon: Alexa data and privacy FAQ

Audio processing and transcript retention can differ

Apple's Siri and Dictation notice says settings indicate when audio is processed on the device. Otherwise audio can be sent to Apple. Without the improvement opt-in, Apple says it does not store that audio, while request transcripts and related data have a separate retention policy. Do not infer an absence of transcripts from an audio-storage statement. Apple: Siri and Dictation default processing

Read the exact optional improvement program

Apple's 09-14-2026 Siri and Dictation notice describes optional audio storage and review, request data retained for up to two years, and a reviewed subset that may remain longer. It distinguishes audio review by Apple employees from review of other data. Apple: optional Siri and Dictation improvement

The separate Improve Siri and Apple Intelligence notice covers opting in with Siri AI, including text and audio interactions used to improve foundation models. It states a two-year period with a longer-lived reviewed subset and describes steps to remove personal details. Its scope should not be substituted for the other setting's disclosure. Apple: separate Siri and Apple Intelligence option

Gemini separates activity history from audio improvement

Google's 10-23-2026 Privacy Hub distinguishes Keep Activity from a separate audio and Gemini Live recording-improvement choice, which is off by default. Transcripts and uploads can fall under Keep Activity even when that audio choice is off. With Keep Activity off, chats still remain with the account for 72 hours; feedback can have additional use and retention consequences. Previously reviewed, account-disconnected material can remain for up to three years. Google: Gemini Apps privacy and retention

Deleting voice history has a defined scope

Amazon says its no-saved-recordings choice deletes audio after request processing, but chat transcripts and typed requests remain for 30 days after the chat's last interaction. Deleting Alexa history does not delete a skill developer's retained data. Read the specific data category and consequence before choosing a deletion or retention option. Amazon: Alexa data and privacy FAQ

Questions and answers

Does no saved audio mean no request transcript exists?

Apple describes transcripts and related request data separately from whether audio is stored. Apple: Siri and Dictation default processing

Does turning off Gemini Keep Activity mean no retention?

Google describes 72-hour account retention and separate feedback and reviewed-data qualifications. Google: Gemini Apps privacy and retention

Can changing Alexa retention affect voice ID?

Amazon says choosing not to save recordings removes associated voice IDs. Amazon: Alexa data and privacy FAQ

Sources and scope

Primary disclosures checked 10-23-2026. Apple improvement notices dated 09-14-2026 and support instructions dated 09-16-2026 were read. Google's Privacy Hub displayed an overall 10-23-2026 update and a distinct embedded-notice date of 10-13-2026. Amazon's official help body was readable on its Amazon help host after the ordinary storefront help URL returned a service error. These are attributed product disclosures, not an independent recording audit or an inference about a person's conversations.

Keep exploring