What we looked at
Researchers routinely upload interview recordings to cloud AI transcription tools, and it is not obvious what their own institution has told them to do. So we read the public pages. The sample was 130 institutions: 100 US R1 universities drawn at random from 187, 15 of the 24 UK Russell Group members, and all 15 Canadian U15 members.
Each institution was coded twice, independently, from saved page text, and disagreements were settled against the page text. The pages were collected on 30 September 2026. The full method, the counts and the underlying data are in the paper, which is on SSRN and Zenodo with the dataset under a CC BY 4.0 licence.
Most universities say something, but not always about transcription
Ninety institutions (69.2%) published something that touches AI tools and recordings or transcription. Only 73 (56.2%) gave guidance specific to transcription. Forty (30.8%) had nothing we could find.
Much of what exists comes from IT pages about meetings rather than from ethics offices. A researcher reading the IT page may learn which meeting platform is licensed and miss that no one has written down what an ethics board expects of an interview file.
The same tool gets different answers
Sixty-one institutions name at least one tool as approved or prohibited. The pattern is consistent in outline: platforms the university already licenses are the tools most often approved, and stand-alone meeting bots are the tools most often barred.
Underneath that outline, institutions disagree. A tool one university prohibits is approved at another, and a few tools sit on both lists across the sample. The counts are institutions, not mentions. If your colleague at another university was told yes, that tells you little about what your own ethics office will say.
Disclosure is the rarer rule
Among institutions with guidance, 25.6% say researchers must disclose AI transcription in the ethics protocol. The rest are silent or unclear, so that figure is a lower bound. And 36.7% publish no date at all, which makes it hard to tell whether a page describes this year's rules.
In practice, many ethics boards will want to know which tool touches the audio and where the file goes, whether or not a page says so. Writing both into the protocol costs a paragraph. Being asked about it after fieldwork costs more.
What to do with a missing rule
A missing rule is a question to ask, not a permission. Before you upload research recordings, check your ethics office or IRB, and ask what they expect for cloud processing of audio. Our qualitative research transcription guide and confidential transcription guide cover the practical steps, and the researcher workflow shows how that looks end to end.
The other half of the question is what the vendor's own documents say. We looked at that separately: do transcription services train on your recordings, and how long they keep your audio. The pre-upload checklist puts both together.
Limits of the survey
The sample covers large research universities in three countries, and pages change. Counts are institutions, not tools mentioned, and a tool can be approved at one institution and prohibited at another. The study did not test any vendor's product. Pepys sells a cloud AI transcription service, no institution in the sample names Pepys, and the paper states that interest.

