Updated
A straight answer, with sources
Sell Data to OpenAI
Can you sell data to OpenAI? A small number of organizations can, either through the Data Partnerships program OpenAI announced in November 2023 or through large licensing deals like the ones it has signed with major news publishers. For most established businesses, the realistic way to get paid for old records is a license to the specialist companies that collect training data for AI labs.
DillisData is not affiliated with OpenAI. We have no agreement, partnership or special access with OpenAI or with any other AI lab, and we will never suggest otherwise. We are a data agency that points business owners to the AI data programs offering the best pay for their kind of old records, and the owner signs up with the program directly. Below is what OpenAI has said in public, what it means for you, and how to see a real price range for your company.
- No ties to any lab
- Every claim sourced
- Costs owners nothing
The short answer
Does OpenAI Buy Data From Companies?
Sometimes, and mostly at very large scale. On November 9, 2023, OpenAI said it wanted to work with outside organizations to build new training datasets. It asked for large datasets that are not already easy to find on the open web. It said it cared most about material that shows what people mean and want, using the phrase "expresses human intention", and it gave long writing and conversations as examples.
The program had two tracks. One was a public, open-source collection that anyone could use to train models. The other was a private track, where an organization's data would help train OpenAI's own models and stay confidential. The first partners OpenAI named were the government of Iceland, working with the Icelandic language company Miðeind, and the Free Law Project, a nonprofit that collects court documents.
Two details matter for a business owner. Reports on the program said OpenAI did not want datasets with sensitive or personal information, or data owned by a third party. And TechCrunch pointed out that the announcement said nothing about paying the organizations that contributed. As of October 2026 we could not confirm whether the program still takes new partners, so check OpenAI's own site before you count on it.
Who gets paid
OpenAI Buying Data: Who Has Actually Been Paid
The OpenAI deals with reported prices involve very large content owners. News Corp, which owns The Wall Street Journal, the New York Post and The Times of London, signed a multiyear license that the Journal reported could be worth more than $250 million over five years. Dotdash Meredith, the publisher behind Investopedia and Better Homes & Gardens, gets at least $16 million a year from OpenAI, according to financial documents reviewed by Adweek.
These companies own millions of articles and decades of archives. A deal of that kind is out of reach for a 60-person firm, and the reason is size, not quality. A lab has limited time for negotiation, and it makes sense that it spends that time on the largest sources first.
Records from ordinary businesses still reach AI labs, but by a different road. As a rule, labs source company records from specialist data suppliers, not from mid-size firms one by one. A supplier gathers email, chat, tickets and documents from many businesses, removes names and personal details, and packages the result for the labs it serves. That supplier is the buyer a mid-size company actually signs with, and it is the kind of program DillisData matches you with. Our page on how frontier labs source company records explains that chain in more detail.
Matching the request
What OpenAI Asked For, and the Records You Already Keep
The kinds of data OpenAI described in 2023 line up closely with ordinary company files. Here is where each kind usually sits inside a normal business. For more on each record type, read our main page on how to sell data to AI companies.
| What labs have said they want | The matching business records | Where they usually live |
|---|---|---|
| Long-form writing | Proposals, reports, manuals, standard operating procedures | SharePoint, Google Drive, OneDrive, Box |
| Conversations | Email threads between staff, vendors and clients | Microsoft 365, Gmail, Exchange archives |
| Back and forth that solves problems | Internal chat about orders, delays and fixes | Slack, Microsoft Teams |
| Data that is hard to find online | Support tickets with the question and the answer | Zendesk, Freshdesk, Intercom |
| Step by step work | Work orders, job notes, project tasks | ERP systems, Jira, Asana, field service apps |
| Audio and speech | Sales and support calls, recorded meetings | Zoom, RingCentral, Dialpad |
| Expert knowledge | How-to guides, wikis, code | Confluence, Notion, GitHub |
What stays out
What OpenAI and Other Buyers Don't Want in a Dataset
OpenAI said in 2023 it did not want sensitive personal data or data that belongs to someone else, and a careful seller should hold to the same limits. A law firm that studies these deals, Frankfurt Kurnit Klein & Selz, lists the issues below. This is general information, not legal advice. Have your lawyer review any license.
Other companies' secrets
Old threads can contain details a client or vendor shared with you under an NDA. Those threads, or those whole accounts, should be listed as exclusions in the contract.
Customer personal data
Customer lists, payment details and account records are usually left out completely. Spirit Airlines' auction left out passenger bookings and payments for the same reason.
Old privacy promises
Your privacy policy or staff handbook may never have mentioned AI training. According to the firm, the FTC treats a quiet rewrite of earlier privacy terms as a possible unfair or deceptive practice.
California employee data
California's CCPA may treat a paid transfer of personal details about California staff as a sale, with extra duties, unless the data is properly de-identified first.
Staff talk about pay and conditions
Federal labor law (the NLRA) may protect messages where workers compare wages, shifts or conditions. Many owners leave those channels out.
No take-backs
Once records train a model, they can't practically be pulled back out. That is why choosing what goes in happens before signing, and why every exclusion should be in writing.
See What a License to Your Records Could Bring
Eight questions about your company give you a cautious range in about two minutes. You don't share any files or logins.
The supplier route
How to Sell Data to OpenAI's Suppliers and Other AI Buyers
You can't control which lab ends up training on a licensed copy. You can control the price, the terms and what gets included. This is how the process runs with us.
Answer eight questions
Tell us your staff count, your age as a company, your main software and your team's country. That gives an early range with no uploads.
2 minutesWe find your best fits
We check how each paying program handles personal details, what it buys and what it pays, then rank the top 3 for your company.
SecondsYou pick and sign up
Each match flags any exclusive or permanent terms. You write your exclusions into the program's contract and sign up directly. The choice is yours.
One clickThe buyer copies and pays
The buyer pulls its copy directly from your systems, so your files never pass through DillisData. It strips personal details and then pays on its published schedule, somewhere between 7 and 60 days.
7 to 60 daysOn the record
What Large AI Companies Have Paid for Data
These are reported figures from public sources. They show real demand. They are not a forecast of your offer.
Reported value of News Corp's five-year content license with OpenAI, covering current and archived articles.
Variety, citing the WSJ, May 2024The fixed yearly minimum OpenAI pays Dotdash Meredith, with a variable amount on top.
Engadget, citing Adweek, Nov 2024Reported yearly value of Reddit's deal letting Google train AI on its content.
Reuters via Nieman Lab, Feb 2024Google's winning auction bid for Spirit Airlines' internal email, chat and files, with customer data excluded. Still pending court approval as of October 2026.
Axios, Aug 2026, The Next Web, Oct 2026Timing
Why Owners Are Moving Now
Today's prices for company records have no guaranteed end date, and no guaranteed future either. Here is why we think delay has a cost.
The public supply is thinning
A 2024 study from Epoch AI projected that models may exhaust the stock of public human writing between 2026 and 2032. OpenAI co-founder Ilya Sutskever told researchers in December 2024 that AI had hit "peak data".
First records in a field earn more
A buyer pays most for the kind of records it doesn't have yet. Once several companies in your line of work have licensed theirs, later offers tend to shrink.
Labs keep finding other ways
As labs fill their gaps and improve other training methods, they will need less of this. Nobody knows when, so it makes sense to learn your number while demand is high.
Lab priorities shift
What labs ask suppliers for changes as their models improve. The record types in demand today may not be on next year's list.
Questions
Questions owners ask
Does OpenAI buy data from small and mid-size businesses?
Not directly, in most cases. OpenAI's announced program asked for very large datasets, and its reported paid deals are with big publishers. A mid-size company's records usually reach labs through specialist suppliers, and that supplier is who pays you.
How do I apply to OpenAI's Data Partnerships program?
In 2023 OpenAI invited interested organizations to contact it through its website. We could not confirm whether it still accepts new partners as of October 2026, so check OpenAI's site. Keep in mind the announcement did not mention payment, and it focused on very large datasets.
Is DillisData connected to OpenAI?
No. We have no relationship with OpenAI, and we don't claim one with any lab. We match you, the owner, with paying programs. The program pays our referral fee separately, so it never reduces the amount you receive.
Can I sell data to Anthropic?
Anthropic is another frontier lab. As with OpenAI, a mid-size business is unlikely to sign with it directly. Business records generally reach labs like Anthropic through the suppliers that serve them, so the practical step is the same: license to a supplier on terms you control.
Can I sell data to Google?
Google has paid for data in public. Reuters reported its Reddit content deal at about $60 million a year, and Google won the August 2026 auction for Spirit Airlines' internal records with a $10 million bid, still pending court approval as of October 2026. Both were large and unusual. A typical company gets paid through suppliers instead.
Will my records end up training ChatGPT?
No one can promise that either way. A supplier may serve more than one lab. If it matters to you which labs receive a copy, ask the program before you sign up and get the answer written into the license.
What does OpenAI pay per dataset?
OpenAI publishes no price list. The only figures in public are reported totals from large deals, like News Corp's and Dotdash Meredith's. For a business, the useful number is what suppliers offer for your headcount and years of records, which is what our estimate approximates.
If I license records, does the buyer own them?
No. The buyer gets permission to use a copy under set terms, and the originals stay yours, along with the choice of what is excluded. Our guide to a data licensing agreement shows the clauses to read closely.
Next step
Get a Real Range for Your Company's Records
Skip the cold emails to AI labs. Answer eight questions and see what buyers may offer a business like yours, before you decide anything.
Free. About two minutes. Ranked matches with reasons.