What does Google's speech to text privacy policy say about your voice?

The policy to read is Google's general privacy policy. It says that, depending on your settings, Google can save audio recordings of your voice interactions with Search, Assistant, Maps and Gboard ¹. The policy does not say how long that audio is kept. What is actually kept depends on the Google product you use, and each product has its own setting or document.

The policy saves voice audio only through your settings

The policy's wording, "depending on your settings", matters. The setting for voice and audio activity in your Google Account is Include voice and audio activity, and it is off unless you turn it on ². With it off, audio from voice interactions is not saved to your Google Account by default.

The policy lists Gboard alongside those services. So the same settings-based rule applies to audio from Gboard that you agree to share, which is covered in the next section.

Gboard keeps speech only if you donate it, and for up to 18 months

Donating audio to Gboard is a choice. Google says data donation is voluntary, and you may be prompted to share audio snippets when you use dictation ³.

If you agree, Gboard's conventional learning sends snippets of your speech to Google. Each snippet includes a maximum of 15 seconds of your speech ⁴, though a snippet can run longer when there is silence or unrecognized audio. Google says it won't keep these snippets beyond 18 months ⁵.

The 18 months is a ceiling, not a schedule. Google's page does not say that every snippet is kept for the full period, so the safer reading is that 18 months is the longest it will keep a donated snippet.

Cloud Speech-to-Text keeps nothing unless you opt in to logging

For developers who use Google Cloud Speech-to-Text, the default is no logging. Cloud Speech-to-Text does not log customer audio or transcripts unless the customer opts into data logging ⁶.

If you do opt in, Google says that only a limited set of authorized Google employees and contractors can access the logged data ⁷. Using the API does not switch logging on, because it is an opt-in program.

The policy gives no retention period for voice audio

The privacy policy describes what Google collects and why. It does not say how long voice audio is kept ¹. The periods come from the product pages: 18 months for donated Gboard snippets ⁵. For audio saved to your Google Account, check the account setting, since the policy gives no period. If a product's page gives no period, do not assume one.

Your phone and keyboard decide where speech goes

The product matters more than the policy text. On an iPhone, Google's App Store listing for Gboard says that if you use Gboard's microphone, voice input is sent to Google for transcription ⁸. So on an iPhone, Gboard's microphone sends your speech to Google to be transcribed.

Offline voice typing is a different path, because it needs no connection to send your speech anywhere. For how it compares with Apple's Mac dictation, see which is more private, Apple dictation or Google voice typing.

What to do

1. Name the product. Decide whether you mean Assistant or Gemini, Gboard, or Cloud Speech-to-Text. Each one has different rules.
2. Check your Google Account. Open your Google Account, click Data & privacy, then under History settings, open Web & App Activity and check Include voice and audio activity.
3. Turn off Gboard audio donation. At the top of the keyboard, tap Settings, then Privacy, and under Voice, turn Audio donations off.
4. Keep data logging off for Cloud Speech-to-Text unless you have a reason to turn it on, and read the terms first. If your speech includes health or other regulated information, check with your compliance team before you send any audio.

For the question of whether Google keeps your recordings at all, read whether Google keeps voice recordings. If you use the browser's speech service, whether the Web Speech API sends audio to Google explains that path. Our guide to how long each company keeps voice data compares Google with Apple, Microsoft and Amazon, and our overview of speech to text privacy covers where your voice goes in general. The term federated learning is explained in the glossary.