Is my dictation used to train AI models?

Most dictation apps answer this with a policy. They receive your audio or your transcript on their servers, then promise not to use it to train their models, or offer a “privacy” setting that turns that promise on. It is a commitment about data they have already collected, and you have to trust that the setting stays on and the promise holds.

Dicta answers it with architecture instead of a promise. Speech recognition and the AI rewrite both run on your Mac, using on-device models, so your voice and your text never reach a server we or anyone else controls. There is no dataset of your dictation to train on, because none is ever collected. Run Dicta behind a block-all firewall and it still works. See the Privacy page for the short list of non-content network calls the app does make (model download, update check, license refresh, and a daily usage-statistics upload linked to your license).