What Is AI Data Licensing? A Plain-English Guide for Companies

AI data licensing is an agreement in which a company gives an AI developer permission to use a defined dataset, most often to train AI models, under written terms, while the company keeps ownership of the data. This guide explains what that means for a small or mid-size company holding ordinary business data. It covers what a license typically includes, how licensing differs from selling outright, and the work that has to happen before anything is signed.

What AI data licensing means

AI data licensing means granting permission to use data, not handing over ownership. The company that holds the data (the licensor) lets an AI developer (the licensee) use a specific dataset for agreed purposes, for an agreed period, under agreed conditions.

A few terms you’ll see throughout:

• License: written permission to use something you own, on conditions you set, without transferring ownership.
• Training data: the examples an AI model learns from.
• AI lab: a company that builds AI models; in these deals, it is the buyer.

For a small or mid-size company, the data in question is ordinary operating material: internal messages, code, documents, records and support tickets.

Why AI labs license data

Public text is finite. Epoch AI estimates the total effective stock of human-generated public text at about 300 trillion tokens, with a 90% confidence interval of 100 trillion to 1,000 trillion. A token is a short chunk of text, often part of a word. Epoch projects, with 80% confidence, that this stock will be fully used at some point between 2026 and 2032 (Epoch AI).

The Associated Press reported that this limited supply could push AI companies toward data now considered private, such as emails or text messages (AP via Courthouse News). That makes it important that any business data offered for licensing is cleared for rights and has identifying details removed first, as described below.

Do AI data licensing deals really exist?

Yes. Licensing deals between content owners and AI developers have been publicly reported. The examples below all involve large online platforms or news publishers.

Date reported	Who	What was reported
Feb 21, 2024	Reddit (a large online forum) and Google	Reuters reported a partnership to make Reddit content available for training Google’s AI models, under a contract estimated at about $60 million per year (Lutzker & Lutzker).
IPO disclosures	Reddit	Disclosures showed $203 million in aggregate data licensing contracts (Yahoo Finance).
Mar 11, 2024	Thomson Reuters (a news and information company) and unnamed AI companies	The CEO told the Financial Times the company had struck ‘a number’ of deals to use its news content for training. The counterparties and amounts were not disclosed (Press Gazette).
May 6, 2024	Stack Overflow (a large developer Q&A site) and OpenAI	The two announced an API partnership (OpenAI), which SiliconANGLE described as a content licensing partnership to provide coding knowledge to OpenAI’s models (SiliconANGLE).
May 22, 2024	News Corp (a news publisher) and OpenAI	The deal lets OpenAI’s models use content from many of News Corp’s major news brands in the UK, US and Australia. The Wall Street Journal valued it at more than $250m over five years (Press Gazette), with payment as cash plus credits for using AI technology (Deadline).
2024	OpenAI and news organisations	OpenAI was reported to be offering news organisations between $1m and $5m per year to license copyrighted content for training (Press Gazette).

The Reddit–Google deal shows that a license can run both ways. Google gets access to the Reddit Data API, which is a structured way for one software system to request data from another. Google says it will use that access ‘to improve its products and services,’ including ‘more efficient ways to train models.’ In return, Reddit gets access to Google’s Vertex AI service to improve its internal search results (Tom’s Guide).

What these deals tell a smaller company

• Licensing happens. Labs have signed paid agreements for content they train on.
• Terms can stay private. The Thomson Reuters example shows that counterparties and amounts are not always disclosed.
• Payment isn’t always pure cash. Credits and access to services have been part of the consideration (what each side gives in exchange).
• The headline numbers come from very large platforms. Treat them as evidence that licensing happens, not as a benchmark for what any other dataset is worth.

Licensing vs. selling your data outright

Licensing lets you keep ownership and set conditions on use. Selling outright transfers ownership to the buyer. In intellectual property terms, an outright transfer is called an assignment. WIPO (the World Intellectual Property Organization) notes that a license, unlike an assignment, generally does not give the licensee the right to authorize others to carry out the covered acts (WIPO).

Two definitions, from WIPO’s trademark licensing guide (WIPO):

• Exclusive license: granted to only one licensee. It stops the owner from using the asset or licensing anyone else.
• Non-exclusive license: does neither, so the owner can keep using the asset and license it to others.

	License	Outright sale (assignment)
Ownership	You keep it	Moves to the buyer
Reuse by you	Possible under a non-exclusive license	Generally not, because you no longer own it
Buyer authorizing others	Generally not allowed under a license	Up to the new owner
Exclusivity	Negotiable: exclusive, non-exclusive or time-limited	Not applicable; the buyer owns the asset
Term	Set in the agreement	Ownership has moved; any ongoing terms depend on the contract
Withdrawing	Depends on the license’s term and deletion clauses	Generally no ongoing control once ownership moves, subject to the contract

Why non-exclusive can mean licensing more than once. Under a non-exclusive agreement, the licensor can grant rights to several licensees at once while keeping ownership (WIPO). After its OpenAI deal, News Corp was reported to be looking at deals with other AI developers. One trade publication said this would fit the non-exclusive nature of most AI licensing agreements, though that is press commentary, not a measured statistic (Storyboard18).

The trade-off is price. WIPO says exclusivity should cost more, so licensors can expect higher up-front lump sums and royalties for exclusive deals (WIPO). Royalties are ongoing payments tied to use. Exclusivity doesn’t have to be permanent either. A licensee is sometimes offered a limited exclusive ‘head start’ period, such as six months or a year (Cambridge University Press).

This is general information about how licenses and assignments work. How a particular agreement behaves depends on its wording, so have counsel review it.

What’s in an AI data license: the key terms

An AI data license spells out what data is covered, how it’s delivered, who else can get it and for how long, what the buyer may do with it, what happens when it ends, and how you get paid. WIPO’s guidance for small businesses suggests defining scope by asking what is being licensed, to whom and for what purpose. It also covers whether the license is exclusive, the territory, and possibly a field of use, meaning a limit on the industry or application (WIPO).

Term	What it settles	Questions to ask
Scope	Which tools, date ranges and data types are covered	Which systems and years are in? Is anything excluded?
Format	How the data is delivered	What file format and structure? How does delivery happen?
Exclusivity	Whether anyone else can license the same data	Exclusive, non-exclusive, or a time-limited head start?
Term	How long the license lasts	When does it end? Can it renew?
Permitted use	Training, evaluation (testing how a model performs), or both	Are there limits on field of use, territory or onward sharing?
Deletion and unlisting	What happens to the data when the license ends or you withdraw	Must the buyer delete copies? Can you stop further sales?
Payment	Amount, timing and form	Cash, credits or services? When is it paid?

Two of these terms deserve a closer look:

• Permitted use can be broad. In the Reddit–Google deal, Google described its use as improving ‘its products and services,’ including model training (Tom’s Guide). If you want narrower limits, they need to be written down.
• Payment can take more than one form. The News Corp–OpenAI deal was reported to pay in cash plus credits for using AI technology (Deadline).

This section is general information, not legal advice. Contract terms interact in ways that depend on the full agreement and the law that applies, so have counsel review any license before you sign.

What has to happen before you can license data

Two things come first: confirming you have the right to license the data, and removing details that identify people. Both should be finished before any buyer sees anything.

Rights check

A rights check asks whether anything you’ve already agreed to limits what you can do with the data. Four areas are worth reviewing:

• Customer contracts: some may restrict how data from or about customers can be used or shared.
• Privacy policy: what you’ve told people about how their information is used.
• Employee communications: what staff were told about internal messages and documents.
• Third-party content: material in your systems that belongs to someone else, such as files a vendor or partner sent you.

For how one exchange runs this check, see rights check.

On privacy policies specifically, the US Federal Trade Commission published guidance on February 13, 2024 (Crowell & Moring). It says it may be unfair or deceptive for a company to adopt more permissive data practices, such as sharing consumers’ data with third parties or using it for AI training, and to tell consumers only through a surreptitious, retroactive change to its terms of service or privacy policy (FTC). Whether your data is cleared depends on your specific contracts and policies. Treat this as general information and ask counsel to review them.

De-identification

De-identification means removing or replacing details that identify people, such as names, email addresses, phone numbers and account IDs. The goal is a dataset that keeps its business value without revealing who was involved. How far you need to go depends on the data and the law that applies, which the next section covers.

What ‘de-identified’ means legally (general information)

‘De-identified’ has no single legal definition. In the UK, the ICO distinguishes anonymous information, which falls outside data protection law, from pseudonymous information, which is still personal data, and it discourages the term ‘de-identified’ altogether.

The UK regulator, the Information Commissioner’s Office (ICO), draws this line for any dataset (ICO):

• Anonymous information does not relate to an identified or identifiable person and is not subject to data protection law.
• Pseudonymous information is data where identifiers have been replaced but people could still be re-identified. It is still personal data, and the law applies to it.

The ICO adds that if anyone could use reasonably available means to re-identify the people in a dataset, it has not been effectively anonymised, only pseudonymised (ICO). It discourages the term ‘de-identified data’ because the term isn’t defined in UK law and can mean different things in different contexts (Taylor Wessing).

This section describes the UK ICO’s position only. Whether a particular dataset meets the legal standard where you operate depends on the data, the context and the law that applies, so ask counsel. For what our exchange removes before anything is listed, see de-identification.

How AI data licenses are priced

There is no standard public rate card. Reported deal values vary widely, many terms are never disclosed, and licensing guidance from WIPO says exclusive deals should cost more (WIPO).

Reported figures mix yearly and multi-year amounts, so they don’t line up as a single range. OpenAI was reported to be offering news organisations $1m to $5m per year (Press Gazette). Google’s contract for Reddit content was estimated at about $60 million per year (Lutzker & Lutzker), and Reddit’s IPO disclosures showed $203 million in aggregate data licensing contracts (Yahoo Finance). The Wall Street Journal valued News Corp’s deal with OpenAI at more than $250m over five years (Press Gazette). All of these involve large platforms or publishers, and some deals, like those Thomson Reuters described, came with no disclosed amounts at all.

At SellMyData, price depends on data type, volume, years of history and current lab demand, and there is no published price list. Its homepage calculator asks for:

• employees today and at peak
• years of history
• data types
• how much history is retained
• industry
• whether you want an exclusive or non-exclusive deal

It then shows an indicative payout range calibrated to reported deals and client quotes as of October 2026. That range is not an offer (how the estimate works). Sellers pay nothing up front. A commission is taken only when a sale closes, and the target is payment within 7 days of closing. On the exchange, a “sale” means licensing a de-identified dataset to an AI lab, on terms you approve.

Should your company license its data?

Licensing may be worth exploring if you hold years of real business records, can clear the rights to them, and are comfortable with an AI lab training on a de-identified version. A quick checklist:

• Do you have years of real operating data? To size what you hold, see the guides for Slack, Google Drive, Microsoft Teams and OneDrive and SharePoint.
• Can you clear the rights? Review customer contracts, your privacy policy, employee communications and third-party content with counsel.
• Are you comfortable with de-identified use? A lab would train on your records with identifying details removed, not on the raw originals.
• Do you want control at each step? Look for a process where you approve the data, the buyer and the terms.

What a seller-controlled process looks like

At SellMyData, the work is currently done by hand with each company, and the company stays in control at every step:

1. You export data from the tools and date ranges you choose. The exchange never writes to your systems.
2. The exchange cleans and prepares the export. People’s names, contact details and identifiers are removed and replaced with consistent placeholders before anything is listed (what is removed).
3. You approve the cleaned result.
4. A rights check covering customer contracts, privacy policy, employee communications and third-party content is done with you before listing (rights check).
5. Buyers see only a redacted sample before purchase. After a sale, the licensed lab receives the de-identified dataset, never raw data.
6. Nothing is licensed without your sign-off on the buyer, the terms and the price. Scope, format, exclusivity and term are agreed per dataset, and you can unlist and delete your data at any time.

The exchange does not buy consumer personal data, health records tied to people, or payment card data. Exports are kept in a restricted folder accessible only to the founder and deleted within 30 days of the de-identified dataset being approved (security).

If you’d like a sense of value before sharing anything, answer a few questions to get an estimated price range. There’s no minimum headcount and no sales call. Every request is reviewed by a person, who replies by email. For the full step-by-step process, see How to Sell Your Company’s Data to AI Labs.