Institutional Guidance on AI Transcription for Research Interviews: A Survey of 130 Universities in the United States, United Kingdom and Canada

Researchers routinely upload interview recordings to cloud AI transcription tools, but it is unclear what their institutions tell them to do. We surveyed the public guidance of 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, in a language-model-assisted workflow from saved page text, and disagreements were resolved against the page text. Ninety institutions (69.2%, 95% CI 60.8-76.5) published something that touches AI tools and recordings or transcription, but only 73 (56.2%) gave transcription-specific guidance; 40 (30.8%) had none we could find. Sixty-one institutions name at least one tool as approved or prohibited. Zoom's AI features are approved at 41 institutions and Microsoft Teams at 20, while stand-alone meeting bots are barred: Read.ai at 17, Fireflies.ai at 11 and Otter.ai at 20, yet Otter.ai is also approved at 5. Only 25.6% of institutions with guidance say researchers must disclose AI transcription in the ethics protocol, and 36.7% publish no date. Inter-coder agreement was high on whether guidance exists (kappa = 0.89) and moderate on stance (kappa = 0.68). Institutions give researchers different answers about the same products, and much of the guidance is drawn from IT pages on meetings rather than from ethics offices. Version 1.0 is a working paper that has not been peer reviewed. The coded dataset is archived separately on Zenodo (10.5281/zenodo.23846482). The author is affiliated with Pepys, which sells an AI transcription service; Pepys is not among the tools coded.