Data licensing explained: usage rights, redistribution and AI training

When you buy data, you rarely buy the data itself. You buy a licence: permission to use it in defined ways, for a defined period, by defined people. Two contracts for the same dataset can be worth very different amounts depending on what the licence allows. This guide explains the terms you will meet most often, with particular attention to redistribution and AI use.

This guide is general information, not legal advice. Have a qualified lawyer review any licence before you sign it.

What a data licence actually grants

A data licence is a contract in which the provider grants you rights to use its data, usually without transferring ownership. It typically defines:

• Licensed data: the specific products, fields, coverage and history.
• Permitted use: what you may do with it.
• Restrictions: what you may not do.
• Scope: users, affiliates, territory and term.
• Obligations: confidentiality, security, deletion and audit.
• Risk allocation: warranties, liability caps and indemnities.

Underneath the contract sit the legal rights that make the data licensable in the first place, such as copyright, database rights in some jurisdictions, confidentiality and contractual restrictions from the provider's own sources. If you use data outside the licence, you are usually in breach of contract even where no intellectual property right applies.

Usage rights explained

Internal use and analytics

The most common and least expensive grant. You may analyse the data and use the results inside your organisation. Check the definition of "internal": does it include contractors, consultants and affiliates? Can results be shown to the board, to investors or to clients?

Commercial products and derived data

If the data, or anything built from it, will appear in a product you sell, you need a commercial or "display" licence. The key concept is derived data: scores, aggregates, models or enriched records created by combining the licensed data with your own. Licences differ on three points:

• whether derived data may be shown to customers;
• whether it must be sufficiently transformed that the original data cannot be reconstructed;
• whether you keep it after the licence ends.

Redistribution and sublicensing

Redistribution means passing the data itself, or substantial parts of it, to third parties. Sublicensing means granting others rights to use it. Both are usually prohibited by default and priced separately when allowed. If you are building a platform that exposes data to your users, you need explicit redistribution rights, often with limits on volume, attribution and the users' own reuse.

Marketing and outreach

Contact and audience data used to reach people raises additional legal requirements beyond the licence, such as privacy and electronic marketing rules. The licence should state that outreach is permitted, and the provider should explain the legal basis on which the data was collected and how opt-outs are handled.

AI and machine learning training

Training a model on licensed data is a distinct use, and many older licences neither permit nor clearly forbid it. If you plan to train or fine-tune models, look for an explicit grant that covers:

• training, fine-tuning and evaluation;
• whether model weights trained on the data may be used commercially and kept after termination;
• whether model outputs may reproduce the data, and any restrictions on memorisation;
• whether the provider has the rights to grant this, especially for content created by third parties.

Browse LLM training data to see how providers describe these rights on listings.

Retrieval and grounding (RAG)

Using data to ground an AI assistant at query time is different from training. The data is stored in an index and shown or summarised in answers. A RAG licence should cover storage in a retrieval index, display of excerpts to end users, attribution or linking requirements, freshness obligations and how removed content is purged. See RAG and grounding data for typical offers.

Scope terms: term, territory, users and entities

• Term: one-off licences may be perpetual for the delivered snapshot; subscriptions usually end with the contract. Check renewal notice periods.
• Territory: some licences limit where the data may be used or where the users sit.
• Users or seats: named users, concurrent users or an enterprise-wide grant.
• Entities: which legal entities in your group are covered. Acquisitions can change this mid-contract.

The end of the relationship

What happens at termination is where licences differ most and where buyers are most often surprised:

• Deletion: you may have to delete the licensed data and certify deletion within a set number of days.
• Retention exceptions: legal or regulatory archives, backups that age out and audit trails.
• Derived data survival: whether your models, scores and aggregates can be kept and used.
• Perpetual snapshots: whether data delivered during the term may be used afterwards.

If your business depends on keeping derived data, negotiate this explicitly before signing.

Audit, warranties and indemnities

• Audit rights let the provider verify that you use the data within the licence, usually with notice and limited frequency.
• Warranties are the provider's promises, typically that it has the right to license the data and that it complies with applicable law in collecting it. Accuracy is rarely warranted; data is often supplied "as is".
• Indemnities allocate the cost of third-party claims, for example a claim that the data infringed someone's rights. Ask whether the provider indemnifies you against such claims, and on what cap.
• Liability caps often limit recovery to the fees paid in a period.

A checklist before you sign

1. Does the permitted use match exactly what you described in your specification?
2. Are all teams, contractors and affiliates who will touch the data covered?
3. Is use in customer-facing products, AI training or retrieval explicitly allowed if you need it?
4. Can you keep derived data after termination?
5. Are deletion and certification obligations realistic for your systems?
6. Does the provider warrant its right to license the data?
7. Is there an indemnity for third-party claims, and is the cap meaningful?
8. Are renewal terms and price increase limits clear?

When you contact providers or post a data request on DataStoa, describe your intended use in the first message. Providers can then quote the right licence instead of a default one.

Frequently asked questions

Do I own the data I buy?

Usually not. You receive a licence to use it under defined terms. One-off purchases sometimes include a perpetual licence for the delivered copy, but ownership normally stays with the provider.

Can I use licensed data to train an AI model?

Only if the licence allows it. Many standard licences cover analytics but not training, so ask for an explicit AI training grant and clarify what happens to trained models when the licence ends.

What is the difference between redistribution and display?

Display usually means showing limited data or derived results to your customers inside your product. Redistribution means passing the underlying data on so others can use it. Display is more commonly allowed, often with volume limits.

Is publicly available data free to use without a licence?

Not necessarily. Website terms, database rights, copyright and privacy law can all restrict reuse of public data. If a provider collected it, your use is also governed by your contract with that provider.