OpenAI, the developer of the GPT models and ChatGPT, acquires training data by licensing content from publishers, platforms and image libraries, by crawling the web with GPTBot, and through a Data Partnerships page whose current terms could not be verified. The deal tracker records 21 licensing agreements with OpenAI from 2022-10-25 to 2026-06-21, and no party has disclosed a price; seven state training, seven cover display only and seven do not say.
Every OpenAI figure here is a dated statement by OpenAI, its counterparty or the party that measured it, checked 2026-11-09. This independent reference on AI training data licensing takes no cut of any sale and covers OpenAI regardless of referral programs.
Each number carries its source and check date, and how DataPacts sources and dates every figure is set out in the methodology.
Where OpenAI's Training Data Comes From: What OpenAI Has Stated
Four routes appear in OpenAI's own documents, its counterparties' releases and vendors' pages: licensed content, its own crawler, data vendors, and what users provide.
- Licensed content from publishers, platforms and image libraries, listed in the table below
- Web crawling by GPTBot, which OpenAI says may train its generative AI foundation models
- Human data and environment vendors that name OpenAI as a customer
- User interactions, whose policy is not covered here
OpenAI told Canada's privacy regulators that less than 1 % of its pre-training datasets came from licensed third-party datasets, in PIPEDA Findings #2026-002 published 2026-05-06; the regulators could not verify it. Licensing is therefore, by OpenAI's own account, a small and documented slice of its pre-training data, while the rest comes from other sources, such as crawling.
The 2019 GPT-2 paper describes the older selection rule for WebText: outbound Reddit links with at least 3 karma, 45 million links, slightly over 8 million documents and 40 GB of text. That is a historical statement about one dataset, not a description of current training.
What ChatGPT users provide is a separate question that this page does not answer. The user-data route is one of the ways in which how AI companies get training data is classified.
OpenAI Company Identity
Court records and counterparties' filings name several OpenAI entities rather than one licensing company: OpenAI OpCo, LLC, OpenAI, L.L.C. and OpenAI Ireland Ltd. The identity table lists each attribute with its source, as of 2026-11-09.
| Attribute | Value | Source |
|---|---|---|
| Entities named in filings | OpenAI OpCo, LLC; OpenAI, L.L.C.; OpenAI Ireland Ltd; The New York Times Company's filing refers to "OpenAI Inc. and various of its corporate affiliates" | NYT Form 10-Q, 2026-09-30; court records in GEMA v. OpenAI (LG München I, 42 O 14139/24) and ANI v. OpenAI (Delhi High Court, CS(COMM) 1028/2024) |
| Registry number | None verified | n/a |
| Website | https://openai.com | openai.com |
| Headquarters country | United States | NYT Form 10-Q, 2026-09-30 |
| Founded | Not recorded | n/a |
| Type | Frontier lab; developer of GPT models and ChatGPT | OpenAI bot documentation |
No registry number and no founding date are stated, because no filing or official history page was checked for them.
OpenAI Data Partnerships: What the Program Page Shows
OpenAI runs a Data Partnerships page, but its current status, the data it asks for and any terms could not be verified on the check date, so DataPacts states none of them. The program terms table shows each term with the value found, as of 2026-11-09.
| Term | What OpenAI states | Checked | Source |
|---|---|---|---|
| Program name and page | OpenAI Data Partnerships; the page at openai.com could not be read on the check date | 2026-11-09 | openai.com/index/data-partnerships |
| Current status | Not verified | 2026-11-09 | n/a |
| Data wanted | Not verified | 2026-11-09 | n/a |
| Application process | Not verified | 2026-11-09 | n/a |
| Payout or price | None published in the sources checked | 2026-11-09 | n/a |
| License scope, term, exclusivity, deletion | Not verified | 2026-11-09 | n/a |
| Referral fee | None recorded | 2026-11-09 | n/a |
A licensor cannot learn from the page whether the program is open, what data it wants, what it pays, what the license covers or whether a referral fee exists. Summaries of the program on other sites are secondary and are not repeated here. Any holder weighing an approach therefore has no published price, scope or timeline to compare, and no OpenAI referral program is recorded either. Whether the page is open to new submissions is likewise not stated by any source checked.
Other buyers publish more. Dated payout ranges, referral fees and license scope for the buyers that state them sit in the program terms compared, where OpenAI's row stays marked as unverified.
Who Has Licensed Content to OpenAI: 21 Agreements by Date
The deal tracker records 21 agreements under which publishers, platforms and image libraries license content to OpenAI, from Shutterstock's announcement of 2022-10-25 to Getty Images' display agreement of 2026-06-21. By year the count is 1 in 2022, 2 in 2023, 13 in 2024, 3 in 2025 and 2 in 2026.
The table lists the agreements newest first, as of 2026-11-09, with values as stated by the parties. The use column reads T for training or model improvement, D for display only and n/s for not stated.
| Date | Licensor | What is licensed | Use | Value | Source |
|---|---|---|---|---|---|
| 2026-06-21 | Getty Images Holdings, Inc. | Licensed images for display in ChatGPT search and discovery, multi-year | D | Not disclosed | sec.gov |
| 2026-02 (letter dated 2026-02-12, "recently signed") | Yelp Inc. | Local business reviews and data | n/s | Not disclosed | sec.gov |
| 2025-12-11 | The Walt Disney Company | Three-year license for more than 200 characters plus costumes, props, vehicles and environments for Sora and ChatGPT Images; excludes talent likenesses and voices; subject to definitive agreements | n/s | No license fee stated; Disney states a USD 1 billion equity investment in OpenAI plus warrants (announced 2025-12-11, closing not verified) | thewaltdisneycompany.com |
| 2025-04-22 | The Washington Post | Summaries, quotes and links in ChatGPT | D | Not disclosed | washingtonpost.com |
| 2025-02-14 | Guardian Media Group | Attributed short summaries and extracts in ChatGPT | D | Not disclosed | theguardian.com |
| 2024-10-08 | Hearst | Newspaper and US magazine content (40+ newspapers, 20+ magazine brands) with citations and links | D | Not disclosed | hearst.com |
| 2024-09-24 | GEDI Gruppo Editoriale S.p.A. | News archive (content-sharing agreement) | n/s | Not recorded | garanteprivacy.it |
| 2024-08-20 | Condé Nast | Content of Condé Nast brands in OpenAI products, multi-year | D | Not disclosed | condenast.com |
| 2024-06-27 | TIME USA, LLC | Current content and archives of the last 101 years, displayed with citation and link | D | Not disclosed | time.com |
| 2024-05-29 | The Atlantic Monthly Group LLC | Articles discoverable in ChatGPT with attribution and link | D | Not disclosed | theatlantic.com |
| 2024-05-29 | Vox Media, LLC | Current content and archive, "to be used to train its AI models and be shared within ChatGPT" | T | Not disclosed | vox.com |
| 2024-05-22 | News Corp | Current and archived news from its mastheads, multi-year | n/s | Not disclosed by the parties | newscorp.com |
| 2024-05-16 | Reddit, Inc. | Reddit content via the Reddit Data API for ChatGPT and new products | n/s | Not disclosed | redditinc.com |
| 2024-05-07 | Dotdash Meredith (now People Inc) | Publisher content "used to improve OpenAI's large language models" | T | Not disclosed by the parties | sec.gov |
| 2024-05-06 | Stack Exchange, Inc. (Stack Overflow) | Technical Q&A via OverflowAPI, attribution in ChatGPT | n/s | Not disclosed | stackoverflow.co |
| 2024-04-29 | The Financial Times Limited | News; attributed summaries, quotes and links in ChatGPT | T | Not disclosed | aboutus.ft.com |
| 2024-03-13 | Le Monde | French news; covers model training and answer engines | T | Not disclosed; "une source significative de revenus supplémentaires" (Le Monde) | lemonde.fr |
| 2024-03-13 | Prisa Media | Spanish news (El País, Cinco Días, As, El Huffpost) | T | Not disclosed | elpais.com |
| 2023-12-13 | Axel Springer SE | News (Politico, Business Insider, Bild, Welt); summaries in ChatGPT and training | T | Not disclosed by the parties | axelspringer.com |
| 2023-07-13 | The Associated Press | Part of AP's text archive | n/s | Not disclosed | ap.org |
| 2022-10-25 | Shutterstock, Inc. | Images and metadata (training datasets); expansion announced with the Contributor Fund; start of the original license not dated | T | Not disclosed | shutterstock.com |
Only Le Monde characterizes its revenue, calling the deal a significant source of additional revenue. Disney states an equity investment, which is not a license fee. Every other value is not disclosed, and the full tracker of all buyers sits under AI data licensing deals.
Training, Display or Not Stated: What Each OpenAI License Covers
Seven of the 21 agreements state that content may train or improve OpenAI's models, seven cover display in ChatGPT or OpenAI products only, and seven do not say. The summary table groups the licensors by the use each party states, as of 2026-11-09.
| Use stated | Count | Licensors | Source |
|---|---|---|---|
| Training or model improvement | 7 | Axel Springer, Dotdash Meredith, Financial Times, Le Monde, Prisa Media, Shutterstock, Vox Media | Each party's release |
| Display only | 7 | The Atlantic, Condé Nast, Getty Images, Guardian Media Group, Hearst, TIME, The Washington Post | Each party's release |
| Not stated | 7 | The Associated Press, Disney, GEDI, News Corp, Reddit, Stack Overflow, Yelp | Each party's release |
Counterparty events matter to a licensor. The Italian data protection authority issued a formal warning on the GEDI archive deal in provvedimento n. 741 of 2024-11-27, under Article 58(2)(a) GDPR, stating that the processing could breach GDPR Articles 9, 10, 13 and 14 and Chapter III. Getty Images filed a prepackaged chapter 11 case in the District of Delaware on 2026-10-21, and its 8-K of that day names ChatGPT among the AI search tools its images power. Shutterstock names OpenAI among its AI data customers (2026-03-19), and its Contributor Fund has issued earnings from the OpenAI datasets to contributors since Q4 2022, with no total published.
The seven training agreements and the seven display agreements differ in what they allow, so the license text decides, not the announcement. Disney's license, for instance, excludes talent likenesses and voices and is subject to definitive agreements, and Yelp's letter says only that it recently signed. A holder should write the permitted use, term and deletion into the AI data licensing agreement itself, since only seven releases mention training at all.
Keeping Content Out of OpenAI Training: Crawler Opt-outs
OpenAI states that disallowing GPTBot in robots.txt signals that a site's content should not be used to train its generative AI foundation models (checked 2026-11-09). It documents four user agents, shown with its stated purpose.
- GPTBot: crawls content that may be used to train generative AI foundation models
- OAI-SearchBot: surfaces sites in ChatGPT search results
- ChatGPT-User: visits triggered by a user action, for which "robots.txt rules may not apply"
- OAI-AdsBot: checks ad landing pages and is not used to train generative AI foundation models
The crawls overlap. If a site allows both OAI-SearchBot and GPTBot, OpenAI may use the results of one crawl for both purposes, so allowing search can mean allowing training. How to allow, block or charge AI crawlers is covered per bot in the crawler pages.
Cloudflare measured OpenAI's crawl-to-refer ratio at 887:1 for 2025-08-01 to 2025-08-07, meaning many crawls per visitor referred back, and put GPTBot at about 7.5 % of Verified Bot traffic in 2025, second after Googlebot. These are Cloudflare's measurements, not OpenAI's.
A robots.txt block applies to future crawls only; it does not remove content already collected, and it does not change any license a site has signed, which is why the agreements above and the opt-out are separate decisions.
OpenAI is among the 22 signatories listed on the EU General-Purpose AI Code of Practice page, last updated 2026-10-07 and checked 2026-11-09. An OpenAI training content summary under Article 53 was not located. The code asks for the copyright policy that EU AI Act Article 53 copyright policy requires.
Is OpenAI a Reliable Licensee? Disclosures, Counterparties and Disputes
OpenAI's licensing record is documented mainly by its counterparties: 21 agreements rest on the licensor's release, a securities filing or a regulator's decision, and OpenAI publishes no license terms of its own. Reddit's Q2 2024 shareholder letter names new data licensing partnerships "including with OpenAI and Sprinklr", and News Corp's chief executive Robert Thomson said on 2026-11-04: "We are pleased to have partners in OpenAI and Meta."
Several points stay open. The Data Partnerships page could not be verified, the share of licensed data rests on OpenAI's statement that the regulators could not verify, and no Article 53 summary was located. Values are undisclosed, so a licensor cannot benchmark an offer against the record. Reputation here means documented conduct only: the record shows signed agreements and the disputes routed below, and it does not rank OpenAI against other buyers.
Litigation is outside this profile. OpenAI is a defendant in In re OpenAI, Inc. Copyright Infringement Litigation, MDL No. 3143, in the Southern District of New York, and the lawsuits over training data are tracked case by case.
Benefits, Risks, Downsides and Alternatives to Licensing Data to OpenAI
Licensing to OpenAI puts content in front of a large AI developer that has signed agreements with major publishers, but OpenAI publishes no terms, and every recorded value is undisclosed.
Benefits
- Display agreements carry attribution and links in ChatGPT, as the releases of the Atlantic, Hearst, TIME and the Washington Post state
- A track record of 21 recorded agreements across news, platforms, images and characters
- A licensee that counterparties name in their own filings and shareholder letters, such as Reddit's Q2 2024 letter and News Corp's results release of 2026-11-04
Risks
- No published license terms, so scope, term and deletion depend on negotiation
- Regulatory exposure for personal data in archives, as the GEDI warning shows
- Counterparty events, as Getty Images' chapter 11 filing of 2026-10-21 shows
Downsides
- Values are never public, so no price can be benchmarked
- One crawl may serve search and training, so an opt-in to one use can carry the other
Alternatives
- Keep the content and block GPTBot in robots.txt
- Sell to a different type of buyer, such as a data vendor or marketplace
- Not license at all, which is a valid outcome
The alternatives to selling data include keeping it, and the choice between a lab and other routes is compared on the buyer-choice page.
OpenAI Among Frontier Labs That Acquire Data
OpenAI is one of 20 AI developers profiled as data buyers on DataPacts, and the only one with 21 agreements in the deal tracker. It is compared with other frontier AI labs as data buyers, and Anthropic's profile covers its book purchases and licenses, while Google is the other licensee in Reddit's deals and Microsoft is OpenAI's co-defendant in news suits. Every buyer is compared under who buys AI training data, and the dated tracker of all agreements is kept under the deals page. Example rules are given under robots.txt for AI crawlers, with detail on GPTBot and OpenAI's other crawlers. For choosing a buyer for your data, a lab is one of four routes. Most licensors are news publishers licensing content to AI companies.
More About OpenAI as a Data Buyer
OpenAI and Human Data Vendors
Human data vendors and RL-environment builders name OpenAI among their customers on their own pages, which is a second way OpenAI buys training data. Andon Labs lists OpenAI under "Trusted by" on its home page (checked 2026-11-09), and Gray Swan named OpenAI among users on 2026-05-28; both are the vendors' claims. Expert contributor work is how labs buy human data, including from vendors that build RL environments.
Frequently Asked Questions About OpenAI and Data Licensing
Does OpenAI pay for training data?
OpenAI licenses content from publishers, platforms and image libraries, but no party has disclosed a price. OpenAI also told Canada's privacy regulators that licensed datasets were under 1 % of its pre-training data, which the regulators could not verify.
Can a company apply to OpenAI Data Partnerships?
OpenAI has a Data Partnerships page, but its status, eligibility and terms could not be verified on 2026-11-09. No application process is described here for that reason.
How much has OpenAI paid publishers?
No party to the 21 recorded agreements has disclosed an amount. Le Monde calls its deal a significant source of additional revenue, and Disney states an equity investment rather than a fee.
Which publishers and platforms have licensed content to OpenAI?
The deal tracker records 21 licensors. Examples are news publishers such as News Corp and Axel Springer, platforms such as Reddit and Stack Overflow, image libraries such as Shutterstock and Getty Images, and Disney for characters.