People usually ask this a few minutes before a sensitive call. A client is about to share something personal, the Otter bot is already sitting in the participant list, and a small voice asks whether this is fine.
I don’t think “safe” has a yes or no answer. The more useful question is simpler. Where does the audio go, who can reach it, and did the other person agree to that? Once you can answer those three, you can decide for yourself.
What Otter says it does
It’s only fair to start with Otter’s own words. Its privacy and security page says conversations are private to you and the people you share them with, that two-factor authentication is available, and that staff need your explicit consent before reviewing a recording to troubleshoot something. It also says Otter uses a proprietary method to de-identify user data before training its models, and that it responds to subpoenas and court orders through a documented process.
None of that is sinister. But it does mean the audio and transcript are held on Otter’s systems, rather than only on your computer. Like other information held by a third party, they may be subject to legal process.
The training question needs a closer look at your plan and settings. Otter’s feedback and training controls let you manage permission to access conversations for training and product improvement, including human review. Its Enterprise documentation says Enterprise workspaces are excluded from AI model training by default. The general security page also describes training on de-identified data. I wouldn’t assume the same arrangement applies to every account: check your plan, workspace agreement and current settings, and ask Otter to clarify anything that remains unclear.
Why advisers get nervous
Lawyers have been the most vocal here. An American Bar Association piece from 2025 points out that cloud note-takers bring a third party into a conversation, which can matter for privilege, and that stored transcripts can become discoverable. Employment lawyers at Fisher Phillips have raised similar points about consent, retention and sensitive HR conversations.
You don’t need to be a lawyer for this to apply. If you’re a consultant under NDA, an adviser discussing someone’s finances, or anyone hearing personal stories, your client is trusting you with the conversation. A cloud transcription service quietly widens that circle. Sometimes that’s fine and agreed. Sometimes nobody actually asked.
There’s also the human side. Many people have noticed that a bot joining the call changes the room. Clients ask what it is, a few ask for it to be removed, and the conversation gets a little more careful. For some work that care is healthy. For other work it gets in the way of the honesty you were hoping for.
Questions worth asking any transcription tool
Before your next client call, whatever you use, try answering these.
- Where is the audio processed, on my computer or on someone else’s servers?
- Where are the recording and transcript stored afterwards, and who can open them?
- Is any of it used to train models, and can I switch that off?
- How do I delete a recording, and is it really gone?
- Does a bot join the call, and how will my client feel about that?
- Have I asked for consent in a way my client actually understood?
Those answers help you judge whether a tool fits the conversation. They don’t, by themselves, establish that you’ve met your confidentiality, contractual or legal obligations. The legal examples above are from the US; the rules that apply to your work may differ.
The local option
The alternative is to keep transcription on your own Mac. Apple Silicon is now fast enough to turn speech into text on the device, so the audio doesn’t need to travel to a transcription service at all.
That’s why I built TapScribe. It records your microphone as Me and the call audio playing through your Mac as Them, so nothing joins the meeting. It transcribes on your Mac, plays the audio back alongside the transcript so you can check a name or a tricky phrase, and also handles audio and video files you already have, like a Zoom recording or a WhatsApp voice note. Transcripts export as plain Markdown files that you own.
The honest limits
Local isn’t magic, and I’d rather you hear the limits from me.
- Recording still needs consent. A local tool changes where the audio goes, not your obligations to the person on the call.
- Your files are only as private as the folder they live in. If your Documents folder syncs to iCloud or Dropbox, your recordings sync too. Check that setting before you rely on it.
- Your Mac’s own security matters more now. Use FileVault and a proper password.
- TapScribe works in English only, needs an Apple Silicon Mac on macOS 15 or later, and downloads its speech model once at setup.
- It doesn’t share transcripts with a team or write meeting notes for you. If you need those features, a cloud tool may suit you better, and that’s a fair choice.
A quick checklist before your next client call
- Tell your client you’d like to record, explain where the recording will live, and offer to stop or delete it.
- Decide whether this call needs cloud features at all.
- If not, record locally and check your sync settings.
- Do a ten second test recording so you’re not troubleshooting with the client waiting.
- After the call, keep what you need and delete the rest.
So, is Otter safe for client calls? For plenty of conversations it may be perfectly acceptable, as long as everyone knows and agrees. For the ones where you’d rather the audio wasn’t uploaded for transcription, there’s a practical local alternative. If you’d like to see how it works, the guide to recording a meeting on Mac without a bot walks through it step by step.
Sources
Source wording checked on 30 October 2026. Policies and account controls can change.