Exclusive or non-exclusive: can you sell your data twice?
Yes, usually. Most AI data deals are non-exclusive, so you keep the right to license the same data to other buyers; Reddit licensed its data to Google and then to OpenAI. Grant exclusivity only for a short window, for a narrow use, and for a premium that covers the buyers you would turn away.
What is the difference between an exclusive and a non-exclusive data license?
A non-exclusive license lets one buyer use your data while you keep the right to license the same data to others. An exclusive license promises that nobody else gets it, for the period and use the contract sets. In licensing law generally, a non-exclusive licensor "retains the right to license the same right to others", and "exclusive licenses must be in writing", the Copyright Alliance explains.
| Non-exclusive | Exclusive | |
|---|---|---|
| Other buyers | You can license the same data to them | Locked out for the term and use agreed |
| Price from one buyer | Lower | Higher, if you negotiate it |
| Total income | Can grow with each new buyer | Capped at what one buyer pays |
| Your own use | Yours to keep, if the contract says so | Can be restricted too, so read it closely |
| Fits | Most companies, most of the time | One buyer pays more than several would, for a short window |
Can you sell the same data to more than one AI company?
Yes, and the biggest sellers do. "Most deals let publishers keep their options open. They don't lock into just one AI partner," a May 2025 review of AI content licensing deals found. It lists the deals of Shutterstock, Reddit, Axel Springer, the Associated Press, Informa and The Atlantic as non-exclusive.
Reddit is the clearest example. Its February 2024 stock market filing showed contracts to license its data to customers including Google, and in May 2024 it signed a second deal with OpenAI for the same kind of content, TechCrunch reported.
The same data can't be sold outright twice. If you sell it, it's the buyer's. Licensing is what lets you go back to the market.
When is an exclusive data license worth it?
Rarely, and only on your terms. It can make sense when:
- One buyer offers more than you'd expect from several buyers together over the same period.
- The data is valuable only because it's rare, and every extra copy would lower what each buyer pays.
- You want one buyer to handle a hard job, such as de-identifying a large archive, in return for a head start.
What does the offer look like?
A buyer wants the data, no exclusivity asked
- Keep it non-exclusive.
- Confirm in writing that you can license it to others and use it yourself.
- Limit their use to AI training, with no resale.
A higher price if nobody else gets it
- Ask how long and for what use: both should be narrow.
- Price it against what other buyers would pay you in that time.
- Offer a short window, such as 12 months, rather than the whole term.
Exclusivity asked for, with no premium
- Decline it, or trade it for something concrete: a higher fee or a shorter window.
- A buyer asking for exclusivity at no extra cost is asking you to give up other sales for free.
How do you limit exclusivity if you agree to it?
Narrow it on every axis you can. UK law firm Bristows notes that "exclusivity, territorial scope and sub-licensing should also be addressed" in any license to train AI.
- Time: a window shorter than the license, such as the first 12 months.
- Use: AI training only. Wiley's one-time deal covered content "for specific use in training LLM models".
- Scope: leave some data out. News Corp's deal with OpenAI excluded its Factiva database and its HarperCollins book arm.
- Region or language: exclusive for one language or market, open everywhere else.
- Your own use: keep the right to use the data yourself. A Malaysian law firm advises that the licensor keep the right to train "its own AI that it may want to deploy in the future".
What should a non-exclusive license still protect?
- A reset clause: the Associated Press reportedly can "reset" its OpenAI deal if someone else gets better terms. Ask for the same, so a later buyer's higher price can lift yours.
- No resale or sub-licensing by the buyer: you decide who else gets the data, not them.
- A clear scope: which systems, which years, what's left out.
- Deletion of the raw data at the end, and an answer on what happens to trained models. See perpetual vs term data licenses.
The full checklist is in data licensing agreement terms.
How much more should exclusivity cost?
There is no published benchmark: none of the public AI data deals we found disclose an exclusivity premium. So price it from your own side: the premium should cover what you'd expect other buyers to pay you during the exclusive window.
| Non-exclusive | Exclusive for 2 years | |
|---|---|---|
| Buyer A | $80,000 | $120,000 |
| Buyer B, same period | $80,000 | Not allowed |
| Total over 2 years | $160,000 | $120,000 |
Here the exclusive offer looks bigger but leaves you $40,000 worse off. Ask for more, shorten the window, or stay non-exclusive. Not sure anyone would buy your data at all? See is selling company data to AI worth it, or see if your data can sell.
Questions
Can I sell the same data to multiple AI companies?
Yes, if each license is non-exclusive, which is the norm. Reddit licensed its data to Google and then to OpenAI. You can't if you sold the data outright or granted exclusivity.
What is a non-exclusive data license?
A license that lets one buyer use your data while you keep the right to license the same data to others and, if the contract says so, to use it yourself.
Are AI data licensing deals usually exclusive?
No. A May 2025 review of AI content licensing deals found most let sellers keep their options open, including deals by Shutterstock, Reddit, Axel Springer and the Associated Press.
Should I give an AI company an exclusive data license?
Only for a premium that covers the other buyers you'd turn away, for a short window and a narrow use such as AI training. Otherwise keep it non-exclusive.
Does an exclusive license need to be in writing?
Yes. In US copyright law exclusive licenses must be in writing; non-exclusive ones need not be, but always get any data license in writing.
Sources
- The 7 deal points of AI content licensing agreements (9 May 2025) Rob Kelly, Media & the Machine
- OpenAI inks deal to train AI on Reddit data (16 May 2024) TechCrunch
- Reddit says it's made $203M so far licensing its data (22 February 2024) TechCrunch
- Exclusive vs. nonexclusive licenses Copyright Alliance
- Licensing content to train AI: essential points to consider (13 November 2024) Bristows LLP, law firm
- Licensing of data for AI model training: things to take note of (27 February 2024) HHQ, law firm