1 June 2026
What Happens to Your Audio When You Use a Transcription Tool
Most transcription tools upload your audio to third-party servers without telling you who sees it, how long it stays, or whether it gets deleted.
A lawyer friend called me last year. She had recorded a client intake interview and wanted a transcript. She uploaded the file to a free tool, got her text back in minutes, and felt good about it. Two weeks later, she read the terms of service. The recording sat on their servers. No deletion timeline. No opt-out.
That conversation stuck with me. Audio files are not like other data. They carry tone, identity, and context that text alone cannot. When the audio is sensitive, the stakes are real.
What happens to your audio
Most transcription tools work like this: you upload the file, it goes to a third-party speech-to-text API, and the provider stores it for some period. Some keep it for months. Some use it to train models. Most are vague about both.
AssemblyAI's privacy policy is specific about retention windows and offers a delete API endpoint. That is the exception. Most tools outsource transcription to a provider without telling you which one.
The summary layer adds another hop. If the tool generates AI summaries, your transcript travels to a second service: OpenAI, Anthropic, or another LLM provider. Each hop is a new data-handling agreement you did not review.
Questions to ask before you upload
- Who processes the audio? The tool you see and the service that transcribes it may be different companies.
- How long is audio retained? Vague language about "operational purposes" is not a retention policy.
- Is the audio deleted after transcription? Ask for a deletion endpoint, not a promise.
- Does the LLM provider store transcripts? Some offer Zero Data Retention contracts. Most do not by default.
- Where is data stored? Geography matters if you are subject to GDPR.
What zero data retention means
Zero Data Retention is a contract term. It means the provider agrees not to store your data beyond the API call lifetime. OpenAI's enterprise privacy program covers this for qualifying customers. Anthropic offers similar provisions. These guarantees apply only if the tool you use has that contract in place with the provider.
Ask the tool, not just the provider.
Practical steps
- Read the terms before uploading anything you would not want public.
- Look for tools that name their sub-processors and link to their deletion policies.
- For privileged or regulated data, treat any cloud transcription as a risk that needs sign-off from your organization.
- Test with a non-sensitive clip first. Understand what the tool does before you commit.
Transcribe-It deletes your audio from storage after handing it off to AssemblyAI, calls AssemblyAI's delete endpoint when transcription completes, and routes summaries through a gateway with Zero Data Retention in place for both OpenAI and Anthropic, so you get a transcript, AI summary, and action points in your inbox with minimal data exposure.