Black Friday offers.

See offers
Pepys

Blog

Do transcription services train on your recordings? What 53 sets of terms say

The question users, ethics offices and procurement teams ask first is the one the documents answer least clearly.

By Pepys · Published

Mr Pepys, thinking: "does it just listen learn from my recordings, too?" (with "listen" crossed out)

The short version

Only 14 of 53 services (26.4%) have documents that say customer content is not used to train AI models. Another 15 (28.3%) have at least one document that permits training by default, including 7 whose own documents contradict each other, and for 18 (34.0%) the documents are silent or unclear. Of the 15, 8 describe no way to opt out.

What we read

People upload interviews, meetings and calls to AI transcription and meeting-notes services, and the service's public documents are the main thing a user can consult about what happens to the audio. We analyzed the privacy policies, terms of service, data processing agreements, sub-processor lists and AI-data pages of 53 services: 35 consumer and prosumer transcription tools, 10 meeting-notes bots, 5 quote-based services and 3 meeting platforms.

All 247 documents were taken from Internet Archive captures, not live pages, and 233 (94.3%) were retrieved. Two coders worked independently from the saved text under a codebook fixed before collection, every quote was checked as an exact substring of the saved document, and disagreements were adjudicated against the text. The method, the counts and the dataset are in the paper.

The clear answers are the minority

Fourteen services (26.4%, 95% CI 16.4 to 39.6) have documents that say customer content is not used to train AI models, with no contradicting document. Five (9.4%) describe training only with the user's affirmative choice.

Fifteen (28.3%) have at least one document that permits training by default: 4 without qualification, 4 limited to de-identified or aggregated data, and 7 whose documents contradict each other. For 18 (34.0%) the documents are silent (11) or unclear (7). Taken together, for 25 services (47.2%) the documents do not give a clear answer.

When a company's documents disagree with each other

In each of the seven conflicts, at least one document permits training and another from the same company says content is not used. A user who reads only the privacy policy can come away with a different answer from one who reads the terms or the help center.

Some contradictions are probably pages that were not updated when a policy changed. Others are different terms for consumer and business plans. The paper codes what the documents say and does not decide which one governs. For an ethics office, that matters: approving a tool on the strength of a help-center page may mean relying on the document that does not control.

Opting out is harder than it should be

Where training is the default, opt-outs are scarce. Of the 15 services whose documents permit default training, 8 describe no way to opt out, 4 describe a request to the company, 3 describe a self-serve setting, and 1 limits opt-out to enterprise customers.

That last pattern is wider than one service. Fifteen services (28.3%) state different training or retention terms for business or enterprise customers than for self-serve users. Individuals and small teams can end up with weaker terms than large buyers.

Third-party AI providers are a separate layer. Documents of 19 services (35.8%) say those providers may not train on customer content, 2 permit it, and 28 (52.8%) say nothing.

What to do before you upload

Read more than one document. Put the privacy policy, the terms and the AI or security page side by side, and note any place they disagree. Look for a plain statement about training, who it applies to, and whether you can switch it off yourself. Save the date you read it, because terms move: compared with captures from October 2024, 23 of 41 comparable services are coded differently on training, and in 16 of those the 2024 policy had said nothing about it.

The pre-upload checklist turns this into eight questions. If your recordings are research data, what universities allow is the other half of the picture, how long services keep your audio covers retention, and our guide to confidential transcription covers handling sensitive audio.

Limits and disclosure

The study reads documents, not systems. A service may do more or less than its documents say, and archive captures can lag the live page. The paper counts what the documents say; it does not rank or recommend services. Pepys sells a cloud AI transcription service and is one of the 53, coded by the same procedure as the rest from captures taken on 3 October 2026. How Pepys Transcribe handles a recording is documented in its privacy and data handling reference.

Share this article

Questions, answered

Do AI transcription services train on your audio?

It depends on the service, and often the documents do not say clearly. Of 53 services, 14 (26.4%) state that customer content is not used to train AI models, 15 (28.3%) permit training by default in at least one document, and 18 (34.0%) are silent or unclear.

Can you opt out of AI training on a transcription service?

Sometimes. Among the 15 services whose documents permit default training, 8 describe no way to opt out, 4 describe a request to the company, 3 describe a self-serve setting, and 1 limits opt-out to enterprise customers.

Why do a company's privacy policy and terms disagree about training?

In the study, seven services published documents that contradict each other. Some are probably pages that were not updated after a policy change, and others reflect different terms for different plans. The paper codes what each document says without deciding which one governs.

References

  1. 1.Shrestha (2026), Who Keeps the Recording? (SSRN) – SSRN
  2. 2.Training, retention and access terms of 53 services: coded dataset (Zenodo) – Zenodo

Keep reading

Keep Pepys in your Google results

If our transcription guides are useful, add Pepys as a Preferred Source in Google.

Add Pepys as a preferred source on Google

Don't just take our word for it.

Ask ChatGPT, Claude, or Perplexity what Pepys is and who it's for. One click, and your favorite AI does the homework.

Built for the four complaints above

Pay as you go, real exports, no training on your audio. Start free with 60 minutes, no card.