Summary
- Otter.ai is one of the most well-known meeting transcription tools in the world. It’s used heavily in corporate environments, online classrooms, and remote teams.
- But when you look at Otter’s data and training policies, a much more complicated picture emerges.
- Otter does train its AI models on user recordings and transcripts. They say they use “de-identified audio” and may use conversation data to improve their AI systems.
- They also ask for explicit consent for manual review of certain recordings which means some human staff can access your content if you check a box.
- There’s also ongoing litigation alleging that Otter used recordings to train models without properly notifying all participants.
- At Good Tape, we believe your private conversations shouldn’t be used to train anything ever. We never train on your data, never take unnecessary rights, and never leave you guessing.
- If privacy matters to you, Otter’s approach is not in your favor.
Otter.ai gained popularity as a transcription and meeting-assistant platform. It can record and transcribe live conversations, integrate with Zoom and Google Meet, and automatically generate notes and summaries. The idea is appealing, especially for busy teams who want an all-in-one assistant to capture meeting minutes automatically. However their terms are less than ideal for most teams.
Otter’s data policies deserve attention
Otter has done an impressive job building an AI-first meeting assistant. I respect innovation in our industry, and normally I prefer to focus on positives but we’re seeing an alarming trend amongst our competitors: training on user data. Sometimes transparency matters more than politeness.
Otter trains on your recordings and transcripts
Otter’s policies clearly state that:
“We train our proprietary artificial intelligence technology on de-identified audio recordings. We also train our technology on transcriptions.”
This means:
- Your audio
- Your transcripts
- Your conversations
- Your meeting content
…may be used to improve Otter’s AI systems.
“De-identified” sounds comforting, but in practice, audio can often reveal identity even after metadata is stripped. The voice is an identifier. And combining text data with patterns, topics, or unique phrasing can make re-identification possible.
Otter’s approach is a standard Silicon Valley method: collect everything possible, anonymize it, then feed it into the model. But for journalists, researchers, therapists, lawyers, and confidential teams, “de-identified” is not enough.
They allow manual human review
Otter says they only access your recordings manually if you explicitly check a box for example when rating transcript quality. But most users don’t realize what that box enables:
- Human staff listening to your recording
- Human staff reviewing your transcript
- Manual audit of your content
To be clear: that’s not inherently wrong. Some companies use human reviewers.
But it needs to be truly clear, and most people clicking “rate transcript” don’t expect that they’re enabling human access to their conversations.
There is active litigation around their data practices
A class-action lawsuit in the United States alleges that Otter.ai:
- used recordings to train machine learning models
- without adequately informing all participants in the recordings
This doesn’t mean Otter is guilty, lawsuits take time. But it does mean their data practices have been questioned publicly, legally, and at scale. If you handle anything sensitive, this should matter.
Their data distinctions are confusing for normal users
Otter says that:
- “Imported customer data” (like Google Drive documents) isn’t used for training
- But recordings/transcripts inside Otter can be used
This is a distinction most users will never understand. A simple question becomes complicated:
“What exactly is being used to train the AI?”
Good Tape’s answer is much simpler: Nothing you upload is ever used to train anything. No legal footnotes. No exceptions. No “de-identified models”. No box you accidentally check. Just: your data stays your data. Check out our CEO statement to learn more.
We believe in a different way of doing things
Good Tape exists for people who want simplicity, clarity, and real privacy. If you trust us with your voice, we handle it with the respect it deserves.
That means:
- We don’t train AI models on your files
- We don’t reuse your audio or transcripts
- We don’t take perpetual licenses
- We don’t access your content unless absolutely necessary for support
- We don’t hide anything in fine print
- We don’t confuse users with complicated distinctions
We want you to feel safe uploading your most important work, without needing to read a legal textbook. See our terms and see the difference.
What we recommend when privacy matters
If your recordings contain:
- sensitive internal meetings
- confidential strategy discussions
- legal or HR conversations
- research interviews
- patient or client information
- unannounced projects
- private moments
- or simply anything you don’t want used to train an AI model
…then Otter’s policies are not aligned with your work. There are better options that offer transcription without data extraction. Your recordings are your intellectual property. Your conversations are part of your life and work. They should remain yours and in your full control.
Trusted by professionals and teams at
Want to read more?
Check out these related resources
Transcription you can actually trust
Good Tape doesn’t train on your data. EVER.