Does Microsoft keep your voice recordings?

Microsoft runs three separate systems that touch your voice, and each keeps different things. Windows voice typing sends your speech to Microsoft to create the text, and Microsoft says it does not keep recordings without your permission. Voice clips are a separate, optional contribution that can be kept for up to two years. Copilot keeps your conversations on a longer clock. The sections below take each one in turn.

Windows voice typing sends speech to Microsoft, but keeps recordings only with your permission

Voice typing on Windows uses online speech recognition. Microsoft says it sends voice data to Microsoft only to provide the service and to create transcriptions, and that it does not store, sample or listen to recordings without your permission ¹. That is the vendor's own description of its handling, not an independent audit.

The other mode works differently. Microsoft says device-based speech recognition in Windows 11 processes your voice locally and sends no voice data to Microsoft ². The same page does not say which features use device-based recognition, and it describes voice typing as online. Check which mode your voice typing uses before you dictate anything sensitive.

Voice clips are optional, and Microsoft keeps the ones you contribute for up to two years

Voice clips are separate from voice typing. You can use voice typing without contributing voice clips, and contributing them is optional ³.

If you do contribute, Microsoft says it keeps those clips to be sampled and listened to for up to two years ⁴. That is the longest period Microsoft states for voice clips, so it is the figure to plan around. The page is undated and describes a policy change from October 2020, so check the current version before you rely on it.

Copilot keeps conversations for 18 months, and voice audio for 48 hours at work

Copilot has its own rules, and they are not the same as voice typing. Microsoft says it stores Copilot conversation activity for 18 months by default, and you can delete individual conversations or the whole history at any time ⁵. That is about 548 days.

For Copilot Voice on work or school accounts, Microsoft says voice audio is temporarily stored for feedback and deleted after 48 hours ⁶. The 48-hour figure is stated for work or school accounts. Read the statement for your own account type.

Training is a further choice. Microsoft says it uses data from Copilot, including voice and conversation activity, for AI training, except for certain categories of users and users who have opted out ⁷. The Copilot app for Windows has a Training on voice conversations setting, separate from the one for text ⁸. Retention and training are separate decisions, and our guide to whether dictation apps train AI on your voice covers the second one.

It depends on the feature, your account and your settings

The table shows why the answer is hard to give in one sentence: the periods come from different pages, and some pages give no period at all.

What to do if you do not want Microsoft to keep your voice

1. Do not contribute voice clips. Voice typing works without them ³.
2. Delete Copilot conversations you no longer need. You can delete one at a time or the whole history ⁵.
3. Turn off Training on voice conversations in the Copilot app if you do not want voice used for training ⁸.
4. For sensitive text, dictate with a tool that processes speech on your computer, and confirm that it does before you rely on it.

If the content should not leave your computer at all, on-device dictation is the simplest way to keep it there. Yap, which we make, is an on-device dictation app for Mac and Windows, and you can see the download options.

Microsoft's rules for Windows voice typing are not the rules for every Microsoft product, and the rules for Azure's speech service are different again. For that, read the question on cloud speech-to-text APIs. For the other platforms, read the question on Google and the question on Apple. The full comparison is in our guide to voice data retention, part of our overview of speech to text privacy.