Licensing research interviews and field recordings to AI

Read the consent you gave

Begin with the original consent, not the recording. What use did the participant agree to. Was confidentiality or anonymity promised. Was the data said to be destroyed after the study, or kept. Was any secondary use mentioned, and if so, how widely.

If the consent was narrow, the recording inherits that limit. A promise of confidentiality made in an interview does not evaporate because a buyer is interested later. Where the original scope does not reach AI training, the honest options are re-consent or leaving that material out.

Ethics and IRB alignment

Interviews gathered under an ethics board or institutional review carry conditions beyond the consent form. Secondary use, data sharing and retention are often governed by the approved protocol. In the United States, human-subjects research is generally governed by the Common Rule, and many institutions apply its standards by policy even where not strictly required.

Before listing, align with your ethics office or IRB. Ask whether the protocol permits sharing, whether a new review or amendment is needed, and whether re-consent is required for a use the participants did not foresee. This protects the participants and the institution, and it is exactly what a serious buyer expects to see documented.

De-identification, and the limit of it

De-identification removes the obvious identifiers: names, places, employers, anything that points to a person. For transcripts and notes, this is routine. For audio it is harder, because the voice itself can identify the speaker even after the words are cleaned.

That limit matters. Where a promise of anonymity was made, releasing identifiable audio may break it even with names removed. Options include using transcripts rather than audio, obtaining consent that covers the voice, or restricting listings to material where identification was never promised.

Who actually owns the recording

Ownership in research is rarely simple. The recording may sit with the researcher, the institution, or a funder, and grant terms can assign rights to one of them. Participants usually do not own the recording, but they hold rights in their own voice and personal data that survive the ownership question.

Sort both before listing. Confirm which body may grant a licence, and confirm that participant consent reaches the use. Both need to be true. Owning the file is not enough if the people in it never agreed to this.

Why field recordings carry real value

This material is often exactly what buyers struggle to find. Low-resource and endangered languages. Regional dialects. Oral histories. Speech recorded in real settings rather than a studio. Domain-specific talk from clinics, courts, fields and workshops.

Scarcity and authenticity are the drivers. A crawler cannot collect a language that was never written down, or a conversation that only exists on your recorder. Prepared carefully, a modest collection of rare, consented recordings can matter more than a large common one.

Age adds to this, not against it. A recording of a dialect as it was spoken decades ago cannot be made again. Time turns an ordinary tape into a scarce one.

A careful path to listing

Move in order. Read the consent. Clear the ethics position. De-identify what can be de-identified. Confirm ownership and the right to grant. Keep the provenance, because a buyer will examine it in diligence.

Where you cannot satisfy all of that, re-consent or set the material aside. Careful and documented is the only version of this that lasts.

One re-consent request, worded plainly, often recovers more material than owners expect. Participants who valued the work are frequently willing to see it reused, when asked directly.