- Rights come first: staff stories usually belong to the publisher, while freelancers keep copyright unless a signed contract transfers it (17 U.S.C. 201 and the Tasini case).
- Wire stories belong to the wire service, and syndication terms usually let you publish them, not relicense them for AI training.
- Publisher deals we found run from July 2023 (AP and OpenAI) through 2026 (News Corp and Meta in March; AP and Anthropic, reported in November). Most values are undisclosed; the figures that circulate are press reports.
- A local paper with decades of staff reporting holds the scarcest part: local news that never reached the big national archives.
What rights does a publisher need to license a news archive?
Start with who wrote and shot each item, and under what paper. A news archive is a mix of staff stories, freelance pieces, wire copy, syndicated columns, reader letters and photos from staff, stringers and agencies. Each slice has its own owner. A license can only cover the slices you own or have written permission to relicense.
| Rule or source | What it says | What it means for your archive | Date checked |
|---|---|---|---|
| 17 U.S.C. 201(b), works made for hire | For a work made for hire, the employer "is considered the author" | Stories and photos staff made as part of the job usually belong to the publisher | 20 Nov 2026 |
| 17 U.S.C. 201(c), collective works | Copyright in each contribution "vests initially in the author of the contribution" | A freelancer keeps copyright in a piece unless a signed contract says otherwise | 20 Nov 2026 |
| New York Times Co. v. Tasini (US Supreme Court, 2001) | Freelance contracts "in no instance secured" consent to database use, and the publishers lost | Old freelance contracts without electronic or new-media rights may not reach an AI license | 20 Nov 2026 |
| Your wire and syndication agreements | Your own contracts; terms are not public | Wire copy and syndicated columns usually stay out unless the provider agrees in writing | Check your copies |
| Photo agency and stringer terms | Your own contracts; terms are not public | Agency pictures are licensed for publication, so treat them as out until checked | Check your copies |
Sources: 17 U.S.C. 201 and the Tasini opinion, both read on 20 November 2026. This page is general information, not legal advice.
Can we sell our news archive to AI companies?
Yes, though in practice publishers license rather than sell. The archive stays yours, and the AI company gets defined uses for a defined term. When the Associated Press signed with OpenAI in July 2023, AP said the arrangement "sees OpenAI licensing part of AP's text archive". The words that matter are "part of". AP licensed a slice, not everything it had ever carried.
The deals that followed split into two uses. AP News reported that most early publisher deals "gave access to news content that OpenAI uses for training", while the News Corp deal also let the AI company use content to answer users' questions. Training draws on the archive as delivered, though buyers may ask for updates to retrain later models. Display needs fresh stories, summaries and links, often with attribution, for years. A small paper's archive fits the first use better, because the value is in the back catalog, not tomorrow's front page.
The rights check decides how much of the archive qualifies. A paper that relied on staff reporters for forty years can license most of its text. A paper that ran mostly wire copy and syndicated features may have a much thinner licensable core than its page count suggests.
How does news archive licensing for AI work?
A news archive license is a contract with five moving parts: what is in scope, what the buyer may do with it, how long, what is paid and how it ends. Scope is the list of titles, years and content types, with freelance and wire material carved out or cleared. Use is training, evaluation, display in answers, or some mix. Term for training is often the life of the model trained on the data, because a model cannot easily forget what it learned.
Publishers also negotiate things that matter only to news. Corrections and retractions should travel with the data, so a story you corrected in 2009 is not licensed in its uncorrected form. Takedowns matter too: if a court orders a story removed, or you unpublished it for a source's safety, the license should let you pull it from future deliveries. Legal holds, embargoed items and stories under a settlement need the same treatment.
Delivery is usually a structured export from your content system: headline, byline, date, section, body text and photo captions as separate fields, plus a rights flag on every item. The cleaner the rights flag, the faster a buyer can say yes. For the general shape of these contracts, see what goes in a data licensing agreement.
How do you license a news archive for AI training?
We help you license your data. Origin Data Partners is an independent data partner that helps business owners value and license their company records: the company pays Origin nothing, and a buyer may pay Origin when a deal completes. The work on the publisher's side usually runs in six steps.
- Count what you hold: titles, years digitized, number of stories, photos and captions, and which years exist only on microfilm or in print.
- Tag ownership: staff, freelance (with the contract form used that year), wire, syndicated, reader submissions and agency photos.
- Pull the freelance contract templates for each era. Find the year electronic and new-media rights first appear, if they ever do.
- List the items that must stay out: settled stories, legal holds, unpublished items, anything under a court order and stories that name protected sources.
- Decide the use you will offer: training only, evaluation only, or display with attribution. See data licensing pricing models for how each is usually paid.
- Prepare a sample of a few hundred cleared staff stories with full metadata, so a buyer can judge quality before anyone sees the full archive.
When you know roughly what you hold, check what your archive may be worth. The estimate starts from a description, not from your files. Not the owner? Advisers can introduce the company.
How much is a news archive worth to AI companies?
Most publisher deal values are undisclosed. When AP News reported the OpenAI and News Corp deal, it said the companies "would not talk about the length or value of the deal". The figure that circulates, "more than $250 million over five years", was a Wall Street Journal estimate quoted by AP, not a number either company confirmed. The New York Times Company's release on its Amazon license, dated 29 May 2025, gave no price at all.
No buyer has published a price for a local or regional newspaper archive. Any per-story or per-year figure you see for a small paper is an estimate or an opening position, not a market rate. The national deals also price things a local paper cannot offer, such as daily new content, brand names shown in answers and global reach.
What moves value for a smaller archive: how many years are digitized as clean text rather than page images, how much of it is staff reporting you can clear, how local and specific the coverage is, and whether metadata such as dates, sections and bylines survives. Decades of city council, court and school board coverage that never reached the national databases is the kind of material that is hard to find anywhere else.
Who are the top news content buyers for AI?
The table lists publisher deals the parties announced themselves, or that a major outlet reported, with the date, what was said about value and the status as far as we could find on 20 November 2026. The list is not exhaustive. A name here is a dated market fact, not a recommendation.
| Date | Publisher and AI company | What was licensed, as stated | Value disclosed? | Status, 20 Nov 2026 |
|---|---|---|---|---|
| 13 July 2023 | Associated Press and OpenAI | OpenAI licenses "part of AP's text archive" | No | Announced; no report of it ending found |
| 13 December 2023 | Axel Springer and OpenAI | Summaries of Axel Springer content in ChatGPT, and content "for advancing the training" of OpenAI's models | No | Announced; no report of it ending found |
| 29 April 2024 | Financial Times and OpenAI | A "strategic partnership and licensing agreement" | No | Announced; no report of it ending found |
| 2024 (AP News report) | News Corp and OpenAI | Content for training and for answering users' questions | No; WSJ estimate of more than $250 million over five years | Announced; no report of it ending found |
| 29 May 2025 | The New York Times and Amazon | Summaries and excerpts in Amazon products, and training Amazon's foundation models | No | Announced; no report of it ending found |
| 30 July 2025 | Gannett (USA TODAY Network) and Perplexity | Content from USA TODAY and over 200 local publications | No | Announced; no report of it ending found |
| 5 December 2025 | Meta and publishers including CNN, Fox News, Le Monde Group, People Inc. and USA TODAY Co. | News in Meta AI answers; USA TODAY Co. says new and archival content from more than 200 local titles | No | Announced; no report of it ending found |
| 3 March 2026 | News Corp and Meta | Content for Meta AI responses and for training | No; WSJ reported up to $50 million a year for three years | Confirmed by News Corp to Engadget; terms not given |
| November 2026 | Associated Press and Anthropic | Election data and content shown in Claude, with attribution | No | Reported by Adweek from two people familiar; neither company confirmed on the record |
Sources: AP, OpenAI on Axel Springer, the FT, AP News on News Corp, The New York Times Company, USA TODAY Co. on Perplexity, TechCrunch on Meta, USA TODAY Co. on Meta, Engadget on News Corp and Meta and Adweek on AP and Anthropic, each read on 20 November 2026. "No report of it ending found" means we searched and found none, not that the deal is confirmed live. Origin's AI data licensing deals tracker lists more deals across all data types.
News publisher AI deals: who is suing and who is signing?
Both happen, sometimes at the same company. AP News reported that The New York Times "sued OpenAI and Microsoft, accusing the companies of effectively stealing the work of its journalists for use in training chatbots". Later the Times signed its own license with Amazon. Suing one AI company and licensing to another is a normal position for a publisher, not a contradiction.
The lawsuits matter to a small publisher in one practical way. They are why AI companies now ask for clean rights. A buyer that has watched publishers sue over unlicensed training wants a seller who can show which stories it owns, which freelancers signed what, and which items are excluded. That paperwork is the product as much as the text.
How we checked this
We read every source on this page on 20 November 2026. The rights rules come from primary law: the copyright statute (17 U.S.C. 201) and the Supreme Court opinion in Tasini. Deal facts come from the companies' own press releases (AP, the FT, OpenAI, The New York Times Company and USA TODAY Co.) and from reporting by AP News, TechCrunch, Engadget and Adweek. For each deal we searched for later news of it ending or changing. Market context comes from the US Copyright Office's report on generative AI training.
What we could not verify: the price of any publisher deal, because none of the parties published one; the terms of any wire or syndication contract, which are not public; whether each deal in the table is still in force, beyond finding no report that it ended; and whether any independent local paper has licensed its archive alone, because we found no public record of one. Local titles have been licensed inside chain deals.
Which content licensing companies handle AI deals for publishers?
Content licensing means granting someone the right to use material you own, for a set use and term, in return for payment. For AI, publishers have three routes: deal directly with an AI company, as the publishers above did; license through a collective licensing body; or use an intermediary that bundles content from many rights holders. The US Copyright Office's report on AI training quotes CCC saying it "already offers market-based, global non-exclusive voluntary licenses to support AI", and cites Created by Humans describing itself as an "AI licensing platform for creators".
The same report quotes an industry analysis calling news "the most active area for individual deals right now by far". For a small publisher, a direct deal is rare. Collective and intermediary routes exist, but read the terms: some take exclusivity, some set the price for you, and some cover only display, not training.
For the wider picture of AI licensing across content types, see AI licensing explained. Book, journal and textbook publishers have their own guide to licensing books, journals and textbooks to AI.
For a newsroom, live pages and the archive can be priced apart: TollBit charges AI bots for current pages at the publisher's rate, and Redpine describes per-use licensing of peer-reviewed research rather than news.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
What is content licensing for a news publisher?
Content licensing is a contract that lets another company use your stories, photos or data for a set purpose and time, in return for payment. For AI, the purpose is usually model training, display of summaries in answers, or both. The publisher keeps ownership. A clear license names the titles and years covered, excludes material the publisher does not own, and says what happens to corrections and takedowns.
Can a local newspaper license its archive to AI companies?
Yes, if it owns the material. Staff reporting is usually the paper's own under work-for-hire rules, while freelance pieces, wire stories and agency photos need checking. Local titles have been licensed inside chain deals, such as the USA TODAY Network's 200+ local papers, but we found no record of an independent local paper licensing alone, and no published price. Its strength is local coverage that national archives never carried.
Do we need freelancers' permission to license old articles to AI?
Often, yes. Under 17 U.S.C. 201(c), a freelancer keeps copyright in a contribution unless a signed contract transfers it. In the Tasini case the Supreme Court ruled against publishers who put freelance articles into databases without that consent. Check the contract form used in each era. If it has no electronic or new-media rights, either ask the writers or leave their pieces out.
Can we include wire stories and syndicated columns in an AI license?
Usually not. Wire copy belongs to the wire service, and a syndicated column belongs to its author or syndicate. Your agreement with them typically lets you publish the material, not relicense it to a third party for AI training. Strip wire and syndicated items from the export by byline, slug or source field, and include them only if the provider agrees in writing.
Are archive photos licensed along with the text?
Only the photos you own. Staff photographers' work made on the job is usually yours. Freelance and stringer photos follow the freelancer's contract, and agency photos are licensed for publication, not for relicensing. Captions are part of your text, but the image itself may not be. Many publishers license text first and treat photos as a separate, later deal once rights are sorted.
What happens to corrections and takedown requests after an AI license is signed?
That depends on the contract, so negotiate it before signing. Ask that corrected versions replace originals in every future delivery, and that you can pull items removed by court order, settlement or for a source's safety. A model already trained on an item cannot easily forget it, so the right to remove usually covers future deliveries and training runs, not models already built.
Are AI content licensing deals with publishers exclusive?
The public announcements we read do not say. AP, the FT, Axel Springer, News Corp and The New York Times did not publish exclusivity terms, and The New York Times sued one AI company while licensing to another. For a smaller publisher, a non-exclusive license keeps the option of licensing the same archive again, which matters when one deal alone may be modest.
Sources and further reading
- 17 U.S.C. 201: Ownership of copyright (LII)
- New York Times Co. v. Tasini, 533 U.S. 483 (2001) (LII)
- AP: AP, OpenAI agree to share select news content and technology (13 July 2023)
- OpenAI: Partnership with Axel Springer (13 December 2023)
- AP News: OpenAI and News Corp content deal
- Financial Times: strategic partnership with OpenAI (29 April 2024)
- The New York Times Company and Amazon announce licensing agreement (29 May 2025)
- USA TODAY Co.: Gannett and Perplexity announce AI content licensing agreement (30 July 2025)
- TechCrunch: Meta signs commercial AI data agreements with publishers (5 December 2025)
- USA TODAY Co.: USA TODAY Co. and Meta announce multi-year AI licensing partnership (5 December 2025)
- Engadget: Meta signs a multimillion dollar AI licensing deal with News Corp (3 March 2026)
- Adweek: Anthropic strikes its first licensing deal, with the AP, for election coverage (November 2026)
- US Copyright Office: Copyright and AI, Part 3, Generative AI Training (pre-publication, May 2025)
