What is Data Provenance?
The documented history of where data came from, who consented to it and how it was handled, which is what a buyer's legal review asks for before anything can be purchased.
Provenance is the answer to a simple question: where did this come from, and can you show it? For AI training data the record usually has to establish the origin of the material, the rights held over it, the consent obtained from anyone in it, and the chain of custody between collection and delivery.
It is not a formality. Provenance is the most common reason a supplier is turned away in this market, ahead of quality and ahead of price. A buyer who cannot show where their training data came from is carrying the risk themselves, and no serious legal review accepts a dataset whose origin is undocumented, however good the content is.
What counts as documented varies with the material. For recordings of real people it means a signed release from every speaker, specific enough about AI training use to survive review, plus a speaker manifest tying each file to a person and a consent record. For licensed content it means the agreement that granted the rights. For web-derived material it means the license or permission under which it was taken.
Provenance is also what makes data resellable. A collection with a complete rights chain can be licensed repeatedly; the same collection without one is unsellable at any price, and no intermediary can repair the gap after the fact.
What a provenance record covers
A provenance record answers four questions: where the material came from, what rights are held over it, who consented to it, and how it was handled between collection and delivery. The answers look different by source. Commissioned recordings need a signed release from every speaker. Licensed content needs the agreement that granted the rights. Web-derived material needs the licence or permission under which it was taken.
The record also has to cover what happened after collection. Every transformation is a step in the chain (transcription, anonymization, conversion into a training format, the creation of synthetic derivatives), and each step should be documented. A set assembled from several sources needs the chain preserved per source, not flattened into one summary.
The test does not change with the material: could a reviewer, given only the record, trace any file in the delivery back to a person or an agreement? If not, that gap is where the risk sits, and it is the part of the record a buyer should probe first.
How provenance is checked before a deal
Provenance is what a buyer's legal review asks for first, and it is the most common reason a supplier is turned away in this market, ahead of quality and ahead of price. The practical checklist is a speaker manifest tying each file to a person and a consent record, the release forms themselves, and the licence for any content the supplier did not record.
Ask for it before signing, not at delivery. A supplier who holds the chain can produce it in days; one who does not will produce excuses, and the excuses do not improve with time. Once the contract is signed, a gap in the chain becomes the buyer's problem, because no intermediary can repair it after the fact.
There is a commercial reason to care beyond legal risk. A collection with a complete rights chain can be licensed repeatedly, to different buyers, in different territories. The same collection with a hole in its history is unsellable at any price. Provenance is not only a cost of doing business; it is what makes the asset tradeable.
Related terms
-
Consent
The speaker knows how their recording will be used, and has signed to say so.
-
Data Compliance
Where the data came from, whether it may be used that way, and whether there is paper to prove it. That whole chain is compliance.
-
Data Licensing
Granting the right to use data you control, under terms that say what the other party may do with it, as distinct from selling it outright.
Keep reading
-
Full glossary
Every term we explain, in one list.
-
How to buy training data
Where these terms actually show up, and which ones change a quote.
-
Help center
How a project runs, from specification to delivery.
Submit a sourcing request
Tell us the language, the hours, and what the data needs to look like. You will get a real number and a real timeline, not a range. If we cannot source it well, we will tell you that instead.
- Pilot batch before the full run, so problems surface early.
- Consent documentation delivered with the data.
- No medical or clinical data. No recorded telephone calls.
- You send the spec Language, hours, recording conditions, and the deadline.
- We reply within two business days A real number and a real timeline, not a range.
- You approve a pilot batch A small run first, so problems surface before the full order.
- Full run, then delivery with documents Audio, transcripts, and the consent record for every speaker.