Voice Data Retention: What Apple, Google and Microsoft Keep

Apple and Google do not keep your voice audio by default ^{1 2}, and Microsoft makes voice clips optional ³. Each of the three still keeps something else: a history of your requests, or your chat conversations. Amazon is harder to summarize. Since March 2025 it has processed Alexa recordings in the cloud ⁴, and what it keeps depends on the setting you choose. Business speech APIs work differently again. Azure's real-time service stores nothing at rest ⁵, while its batch service leaves storage and retention to you ⁶.

Three questions decide how much a service keeps about you: whether the audio is saved, whether a text history is saved, and for how long. The sections below answer those questions for each company.

Our overview of speech to text privacy covers where your voice goes in general. This guide is about how long each company keeps what it collects. The table compares each service as of 17 October 2026, using the company's own help and privacy pages. Where a page gives no period, the table says so.

The periods are easier to compare in days. The chart uses each company's stated default or maximum, and the Alexa bar is the time you can review transcripts before automatic deletion begins.

Two of the four figures are maximums, not defaults: Apple's history and Google's reviewed chats. Read the table for the setting that controls each one.

Apple keeps Siri and Dictation history under a random identifier, and audio only if you opt in

Apple separates the audio from the record of your requests. Dictated audio sent to Apple's servers is not stored unless you opt in to Improve Siri and Dictation ¹. The record is different. Your request history, including transcripts, is tied to a random, rotating identifier rather than your Apple Account ⁷. Apple says it may keep that history for up to two years, which is 730 days at most ¹⁹. Apple may also review a subset of your Siri transcripts, and the small subset it has reviewed may be kept beyond two years ²⁰.

Opting in lets Apple store and review a sample of the audio of your Siri and Dictation interactions ²¹. The default has changed before. In 2019 Apple said that by default it would no longer retain audio recordings of Siri interactions ²². The guide to human review of voice recordings covers that history in full.

Deleting works less well than the keep periods suggest. Apple's Mac Help describes deleting Siri and Dictation request history that is less than six months old ²³, and does not say how older history is handled. Apple's two-year period and its six-month deletion window are different lengths, so a request can outlast the time you have to delete it.

On recent versions of macOS, to stop sharing audio recordings, open System Settings, click Privacy & Security, then Analytics & Improvements, and turn off Improve Siri & Dictation ²⁴. Apple's own settings are covered in depth in Is Apple Dictation private? What Apple keeps.

Google saves voice audio to your account only when its setting is on, and Gemini keeps chats on timers

Voice audio is saved to your account only when the setting is on

The Include voice and audio activity setting in your Google Account is off unless you turn it on ². With it off, audio from voice interactions is not saved to your Google Account. The setting does not control audio that other Google services save, such as Google Voice and YouTube ²⁵. Turning it off does not delete what is already saved ²⁶.

To check the setting:

1. Open your Google Account and click Data & privacy.
2. Under History settings, click Web & App Activity.
3. Next to Include voice and audio activity, check or uncheck the box.

Gemini keeps app activity for 18 months by default

Gemini has its own clock. Google says your Gemini Apps activity is auto-deleted after 18 months by default ⁸. That is about 548 days ²⁷.

Deleting your activity does not reach chats that people have reviewed. Google says reviewed chats are not deleted when you delete your activity, and are retained for up to three years ⁹, or 1,095 days at most ²⁸.

For voice, Gemini apps wait in standby until they are activated. Google warns that audio recordings may begin a few seconds before activation ²⁹. If you want to limit what Gemini keeps, check your Gemini Apps Activity setting.

The phone side of this is covered in our guide to Google Assistant voice commands, and what Gemini changes. The shorter answer to whether Google keeps your recordings is in the question on Google and voice recordings.

Microsoft separates online speech recognition, voice clips and Copilot, and each keeps different data

Microsoft runs three separate systems, and each has its own rules.

Windows voice typing uses online speech recognition. Microsoft says it sends voice data only to provide the service and create transcriptions, and does not store, sample or listen to recordings without your permission ¹². Device-based recognition in Windows 11 works differently: it processes your voice locally and sends no voice data to Microsoft ³⁰. The page does not say which features use it, so check the setting before you rely on it for sensitive text.

Voice clips are a separate and optional contribution. You can use voice typing without contributing them ³. The clips you choose to contribute are a different question from the transcription service, and the pages we read do not give a period for keeping them.

Copilot keeps its data on other clocks. Microsoft says it stores Copilot conversation activity for 18 months by default ¹¹. For work or school accounts, Copilot Voice audio is temporarily stored for feedback and deleted after 48 hours ¹⁰.

Training is a separate decision from retention. The guide to whether dictation apps train AI on your voice covers that question. Our answer on Microsoft and voice recordings is shorter.

Amazon has sent Alexa requests to the cloud since March 2025, and your setting decides what is kept

Amazon told owners of certain Echo models in the United States that from March 28, their voice recordings would be sent to and processed in the cloud, and deleted after Alexa processes the requests ⁴.

If you keep recordings, you still choose how long they stay. Amazon's privacy announcement lists three controls:

• Don't save: if you choose not to save your voice recordings, they are deleted automatically after Alexa processes each request ³¹.
• Auto-delete: you can keep saving recordings and have them deleted on an ongoing three- or 18-month basis ¹⁴.
• Transcripts: you can review the transcripts of your Alexa requests for 30 days before Amazon begins deleting them automatically ¹³.

To change these in the Alexa app, tap More, then Settings, then Alexa Privacy, and open Manage Your Alexa Data. The same settings are on Alexa Privacy Settings online. Amazon's announcement does not say which option is the default, so check the setting rather than assume it. The question on whether Alexa stores your recordings gives the short answer.

Business speech APIs keep little by default, and batch jobs keep files in storage you control

Azure's real-time speech to text processes audio only in server memory, and no data is stored at rest ⁵. Real-time processing like this is the setting to look for if you want zero data retention.

Batch transcription is different. You choose where the input audio and the output files are stored, and you control their retention ⁶. To set an expiry for the generated transcript files, use the timeToLive parameter ¹⁵.

Google Cloud's Speech-to-Text does not log customer audio or transcripts by default ¹⁶. Logging is a separate opt-in choice, so a project that never enables it has no logged audio or transcripts under that program.

Amazon Transcribe stores transcription job data and artifacts, and you can delete them with its Delete APIs ¹⁷. An organization can also opt out of having its content used to improve Amazon Transcribe and its other machine-learning services, through an AWS Organizations opt-out policy ¹⁸. Amazon's FAQ does not say how long job data is kept if you never delete it, so do not assume a time limit. For clinical use, the guide to HIPAA-compliant dictation explains what to ask a vendor. The question on whether speech to text APIs store your audio covers the same ground in brief.

Deleting your history is a different control from opting out of improvement programs

Two kinds of control do different jobs. Deleting removes what is already stored. Opting out, or switching a setting off, changes what is saved or used from now on. Vendors often put these on separate screens, and a control for one does not give you the other.

The distinction matters most in the Google case. Turning off voice audio stops future saving, but the audio you already have stays until you delete it ²⁶. On the AWS side, the opt-out covers improvement ¹⁸, while the Delete APIs remove the job data itself ¹⁷.

When you review these settings, read each one as a separate question: is it saving now, is it used for improvement, and is what it already saved still there. A setting that answers only one of the three leaves the other two open.

Three things the pages we read do not state, so check them yourself

• Amazon's default for Alexa recordings: its announcement does not give one. Open Manage Your Alexa Data and read the current setting.
• How long Microsoft keeps contributed voice clips: the pages we read give no period for clips you choose to contribute.
• How long Google Cloud and AWS keep data when you do nothing: Google's logging page and Amazon's FAQ give no period for stored audio or job data.

If the audio itself should never reach a vendor's server, choose on-device recognition, where the speech is processed on your computer. Yap, which we make, is an on-device dictation app for Mac and Windows, and you can see the download options. For the trade-offs between on-device and cloud recognition, read Is voice to text safe? How to dictate privately and on-device speech recognition.