Does Deepgram train on your data?
Deepgram's terms say it may use your audio and other content to train and test its models. The way out is per request, a parameter on each API call, or a setting in the Saga app.
A self-serve opt-out exists
"We may also use Your Content to improve our Services, to develop other products and services, and for other business purposes, including training and testing our Models."
- 10 Sept 2026
- 5 Oct 2026
- All accounts
- global
- Fetched from the company
- D42562CE…42AE
How we read it: Resolved from UNCLEAR on 10 September 2026 on the terms, last updated 6 August 2026. Training is the default and the opt-out is granular: for the APIs it is set request by request, with the Model Improvement Partnership Program page naming the parameter, and for the Saga app it is a setting; after opting out "any new content you submit won't be used", which leaves what was submitted before. C rather than D because the terms state that training is what happens unless you opt out. The program page calls participation voluntary, which sits oddly beside terms where it is the default.
What each plan can refuse
The same questions, asked of every plan we could identify. Every cell comes from a quoted clause or a documented search - see the evidence above.
- Self-serve toggle
- Not stated
- 36 / 100
- Not stated
- Unknown
How to opt out of Deepgram AI training
Plan by plan, from what Deepgram's own policy says about refusing. Where it publishes the steps, they're listed here with the limits they carry. Where there's nothing to switch off, or nothing graded, this says so instead of inventing a menu path.
All accounts
Yes. Deepgram has a self-serve setting.
- Add mip_opt_out=true as a query parameter to every request you want excluded.
- The opt-out applies to that request only; requests without it are used.
- Open your settings in Saga.
- Use the opt-out choice there; new content after that is not used.
What this policy has done since we started watching
The record starts here. Deepgram's 3 documents were first captured on 14 Aug 2026, and there's no earlier copy to compare them against, so this verdict has no history behind it yet. From now on every re-read is diffed against those copies, and anything that moves appears here.
Vendor response
We haven't yet written to Deepgram about this verdict, and they haven't disputed, clarified or announced a change to any clause quoted above. When either happens it's published here, word for word.
Report a wrong clause