Is your transcription tool uploading your audio?

Don't just take the website's word for it. Your browser can show you what the tool does. This test takes five minutes and works on any transcription tool that runs in a browser.

1. Watch the network

1. Open the tool in Chrome or Edge.
2. Press F12 and open the Network tab. Tick Keep log (called Preserve log in older versions).
3. Add a short recording, about a minute long.
4. Watch the list. Each line is one request. Look for new lines that appear as you add the file, or while the page says "Uploading".
5. Click one. Headers, under General, shows where it went (the Request URL). Payload shows what it carried. Your file name, or a large block of data, means that request carried your audio.

Lines that start with blob: or data: are handled inside your browser. They don't leave your computer. Also check the Socket filter (WS in older versions): some tools stream audio over a single long connection.

2. Cut the connection

This is the test that settles it. First transcribe one recording with the connection on: tools that work on your computer often download their speech model the first time. Then open the Network tab's throttling menu (it usually says No throttling) and pick Offline. Or just turn off Wi-Fi. Now add a recording.

• It still transcribes: the work happened on your computer.
• It fails, or waits forever: it needs a server, so your audio goes to one.

Then go back online with the Network tab still open, in case the tool saved the recording to send later.

3. Read the rules the page sets for itself

Click the first request in the list (the page itself) and open Headers. Find Content-Security-Policy and the part that says connect-src. It lists every place the page is allowed to send data. 'self' means only the tool's own address.

This tells you where data can go, not whether it goes. The tool's own server is still a server. Use it with test 2, not instead of it.

4. Read the privacy policy for three words

Search it for upload, servers and train. A tool that never receives your audio has no reason to describe how it stores, keeps or learns from it.

Ratproof passes all four

Ratproof runs the speech model inside your browser tab. When you add a recording, no new line in the Network tab goes to another server. Set it to Offline and it keeps transcribing. Its policy is connect-src 'self'. Its own server sends the page and the model, and, about once a month if you have Pro, renews the licence with only the licence key. Here is exactly what it does with your file.