Identify the data
List the audio, transcripts and metadata your workflow would send. Mark sensitive material so you can decide whether it should be submitted at all.
Data privacy
Does Deepgram train on my data? What to verify first
Does Deepgram train on my data? A recording can be processed to produce a result without necessarily being used to train a model. Do not infer one from the other: check the current terms for your service, agreement and deployment before submitting sensitive audio.
One-line definition
Training means using data to develop or improve a model; processing means handling data to fulfill a request. The distinction matters because the permissions, retention rules and controls may differ.
-
1
deepgram security
Review the security questions to raise alongside any data-use question.
-
2
is deepgram safe reddit
Separate community impressions from evidence about a specific service agreement.
-
3
Is Deepgram a legitimate company?
Consider company identity separately from contractual privacy protections.
-
4
deepgram reviews
Use customer experiences to frame questions, not to establish binding data terms.
How it works
A reliable answer starts with the documents and settings that apply to your actual use of Deepgram, rather than a general claim about AI services.
- This page cannot confirm your agreement
-
A public explanation does not show which contract, amendment or data-processing terms govern your organization.
WorkaroundAsk the agreement owner for the current signed terms and any applicable data-processing addendum.
- Processing is not proof of training
-
Audio may need to be handled to deliver a speech result. That fact alone does not establish whether it is later used for model development.
WorkaroundLook for explicit language about model improvement, permitted uses and exclusions.
- One service's policy does not cover another
-
Following this page's action takes you to a separate voice product, whose data practices cannot be inferred from Deepgram's terms.
WorkaroundReview that product's own privacy terms before sharing any recording.
Can / cannot do
This review can identify questions to resolve; it cannot replace the terms that govern a particular account or deployment.
Read the applicable terms
Find the current provisions for service delivery, retention, deletion, model improvement and any choices available to your organization.
Get a decision you can keep
Record the document version and the answers to open questions. If the language is unclear, seek written clarification before uploading production recordings.
Who uses it
The same training question has different consequences depending on the recording and the team's obligations.
Security reviewer
An organization is evaluating a speech workflow that could contain confidential calls.
The reviewer checks data-use language alongside access, retention and incident-response questions.
deepgram securityResearch team
Interview recordings include participant information that was collected for a limited purpose.
The team compares any proposed model-improvement use with its consent and research requirements.
is deepgram safe redditProcurement lead
Several people have shared opinions about a provider, but the team needs an approval decision.
The lead distinguishes reviews from the written commitments that apply to the organization.
deepgram reviewsWhat to compare in the written terms
Use these prompts to distinguish an assumption from a provision you can actually verify. The table does not assert what any current Deepgram agreement says.
Assumption to avoid
Documented check
1
Purpose
Assumption to avoid
Audio is processed, so it must train models.
Documented check
Find the permitted purposes for processing and any separate model-improvement clause.
2
Data covered
Assumption to avoid
Only the audio matters.
Documented check
Check whether transcripts, metadata and derived data have distinct treatment.
3
Retention
Assumption to avoid
A completed request means all data is deleted.
Documented check
Look for stated retention periods, exceptions and deletion procedures.
4
Choice
Assumption to avoid
Every customer has the same settings.
Documented check
Confirm which options apply to your agreement and deployment.
5
Third parties
Assumption to avoid
One provider's terms cover every connected tool.
Documented check
Identify each service receiving data and review its applicable terms.
6
Evidence
Assumption to avoid
A forum answer settles the question.
Documented check
Keep the current policy version, signed terms or written clarification.
Review before you share
Take the next step with clear data boundaries
You can explore a separate voice product through the link below. Before sending it audio, review that product's own terms, retention practices and any available controls; Deepgram's policies do not automatically apply there.
- Know what a recording contains
- Check the destination's current terms
- Keep sensitive audio out until questions are resolved
Data privacy FAQ
Do not assume the answer from the fact that Deepgram processes audio. Check the current terms and any agreement governing your use for language about model training or improvement. If the wording does not resolve your situation, request written clarification before sending sensitive data.
No. Transcription is a requested processing task, while training uses data to develop or improve a model. Whether either use is permitted depends on the applicable terms and the data you submit.
Not necessarily. Terms may address recordings, transcripts, metadata and derived information differently. Check how each category is defined before treating a statement about audio as an answer for every output.
Identify the documents that govern your organization's service and ask the agreement owner how they fit together. A general web page cannot interpret your signed terms or confirm which provisions control a particular deployment.