A research agency or insight team can license interview transcripts, focus group recordings and survey answers only where two things line up: the participant's consent wording reaches the new use, and the client contract leaves the agency free to use the material. Syndicated studies the agency paid for are the easiest. Client-commissioned work usually needs the client's agreement. Voice and video identify people, so de-identified transcripts are the practical starting point.
- Read the consent form first. A promise that answers are used "for this study only" blocks reuse.
- Commissioned research usually belongs to the client or is confidential to it.
- A voice is an identifier. Transcripts are easier to license than audio or video.
- Free-text answers need the same review as interviews. Names and employers hide in them.
Can a research agency license old interviews to AI companies?
Sometimes. It depends on what participants were told and who paid for the study. Sort the archive by those two questions before anything else, because each study has its own answer.
The table shows the usual positions. This page is general information, not legal advice. Your consent forms and client contracts decide.
| Material | Usual position | What has to be true |
|---|---|---|
| Syndicated or self-funded studies | The agency owns the work | Consent wording covers sharing with third parties or further use |
| Client-commissioned qualitative work | Confidential to the client, often owned by it | The client agrees in writing, and consent covers it |
| In-house user research at a product company | The company's own records | Participants were told about further use; customer contracts do not forbid it |
| Panel survey data | Governed by the panel terms | The panel terms allow sharing beyond the client |
| Published reports and decks | Copyright of the author or client | You hold the copyright or a licence that allows it |
| Academic or publicly funded interviews | Governed by the ethics approval | The approved consent form allows it. See the consent section |
Does participant consent cover AI training?
Only if the wording reaches it. Consent given for one study does not stretch to a different use by a different organisation. Find the form or script each participant saw and read what it promised about who would see the material and what it would be used for.
Commercial market research is mostly outside the US federal research rules, but those rules are the clearest public model of what good consent wording says about reuse. They ask for a statement either that identifiers might be removed and the information "could be used for future research studies or distributed to another investigator for future research studies without additional informed consent", or that it will not be (45 CFR 46.116). If your form made the second promise, the archive stays where it is.
The US research industry code defines consent as agreement "based upon the person having been provided with clear information about the nature and purpose of the data being collected or used, with whom it will be shared and how it will be used", and it requires researchers to obtain "the research subjects' consent" before transferring personal data to a third party unless a law allows otherwise (Insights Association Code).
In Europe the purpose limit is law: personal data must be "collected for specified, explicit and legitimate purposes and not further processed in a manner that is incompatible with those purposes" (GDPR, Article 5). The guide to GDPR and CCPA in data licensing goes further.
Example: reading four consent wordings
These four wordings are typical of what agencies find in old study files. The readings are ours and are not legal advice. A reviewer goes through every study and puts it in one of these rows.
| What the participant was told | Reading | Decision |
|---|---|---|
| "Your answers will be used for this study only and will be destroyed afterwards" | Reuse is ruled out, and the material should not exist any more | Out |
| "Your answers are confidential and will be reported only in anonymous form" | Silent on reuse. Identifiable material cannot leave. De-identified text is arguable, not certain | Legal review. Do not include audio or video |
| "Recordings may be shared with our client and its partners for research purposes" | Sharing is allowed but tied to the client and to research | Needs the client's agreement and a view on whether model training is research |
| "De-identified responses may be kept and used for future research and product development, including by third parties" | Reaches a licence of de-identified transcripts | In, after de-identification |
Illustrative wordings. For new studies, add the fourth kind of sentence now and say plainly that the material may be used to train or test AI systems.
Does the client own commissioned research?
Check the contract. Copyright does not pass to the client automatically, but most research contracts either assign the deliverables or make everything about the project confidential. Either one stops the agency from licensing the material alone.
Under US copyright law a commissioned work is "made for hire" only in listed categories and only "if the parties expressly agree in a written instrument signed by them" (17 U.S.C. 101). Where a work is made for hire, the commissioning party "is considered the author" (17 U.S.C. 201). So an agency may still own the copyright in a transcript where the contract is silent. The confidentiality clause is usually the real limit, because it covers the client's questions, its product plans and what its customers said.
Advisory firms face the same split between ownership and confidentiality. See the guide for consulting firms.
How do you de-identify interview speech and free text?
Start from transcripts, remove direct identifiers, then read for the indirect ones. People describe their job, employer, town and family in passing, and a combination of small details can point to one person.
Audio and video are harder because the recording itself identifies the speaker. US health privacy rules list "biometric identifiers, including finger and voice prints" and "full face photographic images" among the identifiers to remove (45 CFR 164.514). European law treats data from technical processing of a person's physical or behavioural characteristics that confirms identity as "biometric data" (GDPR, Article 4).
California's test is whether the information "cannot reasonably be used to infer information about, or otherwise be linked to, a particular consumer" (Cal. Civ. Code 1798.140). See anonymization versus pseudonymization for why a participant code alone does not meet it.
- Replace names, employers, products and places with typed placeholders such as [EMPLOYER].
- Keep speaker turns and roles (moderator, participant 3) so the conversation structure survives.
- Remove recruitment screeners that hold contact details, or separate them from the transcripts.
- Review free-text survey answers the same way. Do not assume a closed-question dataset has no free text.
- Hold audio and video back unless consent clearly covers them.
What are interview transcripts and survey responses worth?
No public source gives a price for this kind of record, so this page does not quote one. What raises interest is depth, structure and clean rights, not the number of files.
Survey open-ends are useful when they come with the question asked and the respondent's segment. Closed-question tables alone are thin: many public survey datasets exist already. Published reports are a publishing question more than a data one, and the report's copyright owner is the party who can license it.
| Raises the value | Lowers the value |
|---|---|
| Long moderated conversations with expert or professional participants | Short consumer intercepts |
| Discussion guide, screener criteria and analysis attached to each study | Transcripts with no study context |
| Human-checked transcripts with speaker turns | Raw machine transcripts |
| The analyst's coding frame and the final findings | Raw answers only |
| One consent wording across the archive that covers reuse | A different form for every project, some missing |
| Several languages or markets with the same guide | One-off studies that cannot be compared |
Is audio or video worth more than text?
It can be, because recordings carry tone, pauses and interaction that a transcript loses. It is also much harder to license, because a voice or a face identifies the participant. Most research archives can offer text first.
If audio matters to a buyer, the route is fresh, explicit consent from participants for that use. The guide to call recordings covers recording-consent law, which is a separate question from consent to license.
Common mistakes research teams make
Most come from treating the archive as one asset with one answer.
- Assuming a signed consent form covers any later use.
- Licensing client-commissioned work because the agency still holds the files.
- Treating a participant code as anonymisation while the key still exists.
- Leaving recruitment screeners and incentive payment records in the same folders.
- Offering video of focus groups when only the transcript can be cleared.
- Forgetting that panel providers' terms govern panel responses.
- Keeping material that the consent form promised to destroy.
A research archive as a data asset
Years of moderated interviews, with the guide that shaped each one and the analysis that came out of it, show how people explain their choices and how a skilled researcher gets them to. That is hard to produce on demand. The part that can be licensed is the part where consent and client terms both allow it.
Origin works with buyers and helps a research company sort its studies, describe the part that is clear, find out what it could be worth and decide whether to go further. Nothing is exported at the start. Origin Data Partners is an independent data partner that helps companies get their data bought. A buyer may pay Origin when a deal closes.
Does your company hold records like these? See what your data is worth.
Common questions
Can you sell interview transcripts to AI companies?
Yes where the participant's consent wording covers further use or sharing and no client contract forbids it. De-identify the transcripts first. Studies whose consent form limited use to one project stay out.
Can you sell focus group recordings to AI companies?
Transcripts are possible when consent and client terms allow it. Video and audio identify each participant by face and voice, so they need explicit consent for that use. Most agencies start with de-identified text.
Can you sell user research recordings to AI companies?
A product company owns its user research, but participants were customers or testers who agreed to a specific use. Check the consent script and your customer terms. De-identified notes and transcripts are easier than screen and face recordings.
Can you sell survey responses to AI companies?
Self-funded survey data can be licensed once free-text answers are de-identified. Client-commissioned surveys need the client's agreement, and panel responses are governed by the panel provider's terms.
Who buys market research reports and raw research data?
Reports are bought by publishers, aggregators and corporate subscribers. The raw material behind them, interviews, guides and coded answers, interests AI developers who need examples of real reasoning and expert questioning.
What are interview transcripts worth to an AI buyer?
No public source gives a price. Long expert interviews with the discussion guide and analysis attached, human-checked and covered by one clear consent wording, are worth more than short consumer intercepts or raw machine transcripts.
Sources and further reading
- 45 CFR 46.116: general requirements for informed consent
- 45 CFR 46.102: definitions (research, human subject, identifiable private information)
- 45 CFR 46.104: exempt research (surveys and interviews)
- Insights Association: Code of Standards and Ethics for Market Research and Data Analytics
- Regulation (EU) 2016/679 (GDPR): Articles 4, 5 and 9
- 45 CFR 164.514: de-identification of protected health information (list of identifiers)
- California Civil Code 1798.140: CCPA definitions (deidentified)
- 17 U.S.C. 101: definitions (derivative work, work made for hire)
- 17 U.S.C. 201: ownership of copyright
