Pressing a microphone button in an app is a small act of trust. You should know where the recording goes. Here is the whole path, from the button to the text in your brief.
The short version
- Postfjord never stores your recording.
- The audio is sent to OpenAI once, to be turned into text, and OpenAI keeps none of it.
- Neither OpenAI nor Anthropic trains its models on it.
- Only the text you keep ends up in Postfjord.
The path, step by step
1. Your browser records. When you press Dictate, your browser asks for the microphone the first time. The recording lives in your browser's memory while you talk, for up to three minutes.
2. The audio goes to our server in the EU. When you press stop, the audio is sent over an encrypted connection to a Postfjord function in Google Cloud in Belgium. The function checks that you are signed in and a member of the workspace. It holds the audio in memory only. Nothing is written to our database or file storage.
3. OpenAI turns it into text. The function sends the audio to OpenAI's transcription service and gets the text back. For this service OpenAI's own documentation lists no retention for abuse monitoring and no stored application state: the audio is processed and not kept. OpenAI has not used data sent to its API for training since March 2023.
4. The text comes back to you. The text lands in the brief, the idea box or the email brief. The audio is gone from our side the moment the function finishes. What you save from there is ordinary Postfjord content, like any text you type.
5. Claude writes the post, if you ask. When you press Generate draft, the brief goes to Anthropic to write the draft, under terms that exclude training. That step is the same whether you typed the brief or said it.
Who can hear it
Nobody. There is no copy of the audio to play back, not for you, not for us, not for anyone in your workspace.
What it costs
A minute of dictation costs us about three tenths of a cent at OpenAI's list price. It comes out of your plan's monthly AI budget, the same budget as your drafts. At that price the budget runs out on drafts long before it runs out on dictation.
What we tell you in writing
OpenAI is listed as a provider in our privacy policy, next to the other services that process data for Postfjord. If a workspace has connected its own ChatGPT key under Settings, AI, dictation runs on that key instead, under the workspace's own agreement with OpenAI.
If anything here changes, the privacy policy changes first, and this post gets an update at the top.