# AI Data Licensing Deals in 2026: 49 Publisher Deals, and Only 9 Have a Reported Price

AI data licensing deals are agreements in which an AI company pays a publisher or platform to use its content, for model training, for answers that cite live content, or both. They are now common. Press Gazette's tracker, updated October 21, 2026, lists dozens of deals between news publishers and OpenAI, Google, Meta, Microsoft, Amazon, Perplexity, and others ([Press Gazette](https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/)).

But the terms are mostly private. We read every signed deal in that tracker: of 49 licensing deals, only 9 have any reported price, and most of those figures come from press reports, not from the companies. This transparency report covers what is disclosed, what publishers tell AI crawlers in robots.txt, and what EU law now requires AI companies to publish. It focuses on news publishers and platforms, the deals Press Gazette tracks.

> **Key Takeaways**
>
> - Of **49 publisher licensing deals** in Press Gazette's tracker, **9 have a reported price**, **2 state only a revenue-share formula**, and **38 (78%) disclose neither**.
> - The largest reported deals: News Corp and OpenAI, "more than $250m over five years"; Meta and News Corp, "up to $50m per year for at least three years"; Reddit and Google, "around $60m per year."
> - Payment models are shifting from flat fees to **usage-based** terms, such as Microsoft's Publisher Content Marketplace and Amazon's deal with Reach.
> - In our check of 47 news sites, only **3 publish a machine-readable licensing or usage signal** in robots.txt. 13 state licensing terms or AI reservations only in comments.
> - Under the EU AI Act, general-purpose AI providers must publish a training data summary that lists the **top 10% of internet domains** they scraped.

## What are AI data licensing deals?

An AI data licensing deal gives an AI company permission to use content it would otherwise have to crawl, buy, or leave alone. Deals usually cover one or more of three uses:

1. **Training.** Content is used to train or fine-tune a model. Under Meta's deal with News Corp, Meta can "tap article archives for training," according to [The Decoder](https://the-decoder.com/meta-signs-multi-year-ai-deal-with-news-corp-worth-up-to-50-million-a-yearmeta-signs-multi-million-deal-with-news-corp-for-ai-training-and-content).
2. **Retrieval and display.** An assistant fetches current articles to answer a question, often with a citation and link.
3. **Archive access.** A one-time or recurring transfer of historical content, often bundled with the other two.

Our roundup of [AI training data lawsuits](https://www.joinmassive.com/blog/ai-training-data-lawsuits-web-scrapers) covers the other side of this market: what happens when there is no deal.

## Which AI data licensing deals have a reported price?

These are the deals in Press Gazette's tracker with any reported dollar figure. The wording is the tracker's, including who reported the figure.

| Publisher or platform | AI company | Date | Reported value |
|---|---|---|---|
| News Corp | OpenAI | May 2024 | "More than $250m over five years" (WSJ) |
| Reddit | Google | Feb 2024 | "Around $60m per year" (Reuters) |
| News Corp | Meta | Mar 2026 | "Up to $50m per year for at least three years" |
| The New York Times | Amazon | May 2025 | "$20m to $25m per year" (WSJ) |
| Dotdash Meredith | OpenAI | May 2024 | "At least $16m per year" (Adweek) |
| Informa | Microsoft | May 2024 | "An initial fee of $10m+ and then three more recurring annual payments" |
| Bennett, Coleman and Co; Indian Express | OpenAI | Sep 2026 | "Could be worth around $5m per year" and "could be worth $3m per year" (E4m) |
| Several publishers | Snowflake | Jun 2026 | "Six-figure deals" (Digiday) |
| Independent publishers | Google | Feb 2024 | "A five-figure annual sum" (Adweek) |

Source: [Press Gazette deal tracker](https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/), read October 26, 2026. We excluded five grant arrangements, and we can share the full entry-by-entry tally on request.

![Bar chart of disclosure in AI licensing deals and robots.txt files. Of 49 publisher licensing deals, 9 have a reported price, 2 state a revenue-share formula only, and 38 disclose neither. Of 47 news sites' robots.txt files, 3 publish a machine-readable licensing or usage signal and 13 state licensing terms only in comments.](assets/what-gets-disclosed.png "What AI licensing deals and robots.txt files disclose, October 2026.")

Even the reported figures are hard to compare. Some are annual, some cover five years, and some are ceilings ("up to") rather than amounts paid. The two formula-only entries are both Prorata.ai deals with a 50% revenue share; the 2025 one describes it as "50% of revenues driven by those responses." Future, which signed with OpenAI in December 2024, called its deal "not financially material."

## How are AI licensing payment models changing?

The early deals were mostly flat fees for archive access and display rights. Newer deals tie payment to use:

- **Usage-based marketplaces.** Microsoft's Publisher Content Marketplace "will pay publishers per the use of their content," with Copilot as the first buyer ([Press Gazette](https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/)).
- **Usage-based bilateral deals.** Reach's 2026 agreement with Amazon pays "based on usage," according to the tracker.
- **Revenue shares.** Prorata.ai and Perplexity's publisher programs share revenue from answers that use publishers' content.
- **Pay per crawl.** Infrastructure providers now let sites charge crawlers per request. Our report on [AI crawler blocking and pay per crawl](https://www.joinmassive.com/blog/the-closing-web-ai-crawler-blocking-pay-per-crawl-and-what-it-means-for-agents) covers how that works.

Usage-based terms are fairer in principle, but they make outside scrutiny even harder. A flat fee can be reported once. A usage formula needs usage data that only the buyer holds.

## What do publishers tell AI crawlers about licensing?

A license only helps if crawlers can find the terms. On October 26, 2026, we parsed the robots.txt files of 47 major news sites for our report on [the EDPB's web scraping guidelines](https://www.joinmassive.com/blog/edpb-web-scraping-guidelines). Here is how they express licensing:

| Signal | Sites | Examples |
|---|---|---|
| Really Simple Licensing (RSL) `License:` line, a machine-readable pointer to license terms | 1 | The Guardian |
| `Content-Signal` line with `ai-train=no` | 2 | Corriere della Sera, The Atlantic |
| Licensing contact, or a reservation against text and data mining (TDM) or AI use, in comments only | 13 | The New York Times, Financial Times, Reuters, The Wall Street Journal, BBC, Der Spiegel |
| Named block for at least one AI training crawler | 39 | Most large publishers |

RSL ([rslstandard.org](https://rslstandard.org/)) reached version 1.0 in December 2025, with support from Cloudflare, Akamai, the Associated Press, and others ([The Register](https://www.theregister.com/2025/12/10/really_simple_licensing_spec_takes/)). Adoption on these 47 sites is still one file.

Most publishers block first and put the "how to license us" message in comments, which crawlers skip. Under recent EU rulings, a human-readable reservation may not count as an opt-out at all. In Kneschke v. LAION, Germany's Hamburg Higher Regional Court held in December 2025 that a natural-language reservation "could not be automatically identified or processed" ([Hogan Lovells](https://digital-client-solutions.hlc.com/resources/ai-litigation-case-law-tracker/case/robert-kneschke-vs-laion-e-v/decision/1)). In BoligPortal v. ReData, a Danish appeals court found a reservation in a privacy policy insufficient ([Plesner](https://plesner.com/en/news/high-court-rules-natural-language-opt-out-text-and-data-mining-insufficient)). Our [explainer on whether robots.txt is legally binding](https://www.joinmassive.com/blog/is-robots-txt-legally-binding) covers those cases.

## What does the EU AI Act require AI companies to disclose?

The EU now requires part of the picture to be public. Providers of general-purpose AI models must publish a summary of their training content using a template the European Commission released on July 24, 2025 ([European Commission](https://digital-strategy.ec.europa.eu/en/faqs/template-general-purpose-ai-model-providers-summarise-their-training-content); [Mayer Brown](https://www.mayerbrown.com/en/insights/publications/2025/08/eu-ai-act-news-rules-on-general-purpose-ai-start-applying-guidelines-and-template-for-summary-of-training-data-finalized)). The template sorts training data into six groups:

- Publicly available datasets
- "Private non-publicly available datasets obtained from third parties (licensed datasets)"
- "Data crawled and scraped from online sources"
- User data
- Synthetic data
- Other data

For scraped data, providers must list "the top 10% most relevant internet domains by size of content crawled or scraped." Small and medium-sized companies list the top 5% or 1,000 domains, whichever is lower ([European Commission](https://digital-strategy.ec.europa.eu/en/faqs/template-general-purpose-ai-model-providers-summarise-their-training-content)). The obligations applied from August 2, 2025, for new models. Models already on the market have until August 2, 2027. The Commission can enforce from August 2, 2026. In October 2026, PPC Land, citing Reuters, reported that the Commission had sent information requests to companies that had not yet published their summaries ([PPC Land](https://ppc.land/meta-delays-website-crawling-terms-to-nov-20-after-publisher-complaints/)).

The summary does not require prices. But for the first time, a publisher can see whether its domain is among a model's most-crawled sources, and compare that with whether it ever signed a deal.

## Where does Massive fit?

Massive is a web access network, not a content licensor. Customers use it to reach public pages from real devices in specific locations. Massive does not grant rights to the content they collect, so the licensing questions in this report belong to the customer and the site.

What Massive can be transparent about is its own supply and controls. Every device on the network opted in through the Massive SDK; our [explainer on the opt-in network](https://www.joinmassive.com/blog/how-massives-opt-in-network-works) shows that flow. Customers can block up to 1,000 domains on their own traffic from the dashboard, on top of the categories Massive blocks for every account ([Massive Docs](https://docs.joinmassive.com/residential/domain-blocking); our [domain blocking report](https://www.joinmassive.com/blog/domain-blocking-transparency)). Our [Q3 network integrity update](https://www.joinmassive.com/blog/q3-2026-network-integrity-update) links each compliance claim to a page you can check.

For data teams, the practical step is a source register: for each domain you collect from, record the date, its robots.txt, any license or signal you found, and the use you planned. That record lines up with the categories in the EU template, and it is what a publisher, auditor, or regulator will ask for first.

## Frequently Asked Questions

### What is an AI data licensing deal?

It is a contract in which an AI company pays a publisher, platform, or data owner to use its content for model training, for live answers that cite the content, or for archive access. Terms range from flat annual fees to usage-based payments and revenue shares.

### How much do AI companies pay for licensed content?

Reported figures range from five-figure annual sums to "more than $250m over five years" for News Corp's deal with OpenAI. Most deals don't disclose a price. In Press Gazette's tracker, 38 of 49 licensing deals have no reported price or formula.

### Which AI companies have signed the most licensing deals?

In Press Gazette's tracker, OpenAI is the buyer in 18 of 49 licensing deals, followed by Google (6), Microsoft and Perplexity (5 each), Prorata.ai (4), and Meta and Amazon (3 each). Other trackers count differently, because many agreements are announced without terms or not announced at all.

### Do AI companies have to disclose their training data?

In the EU, providers of general-purpose AI models must publish a summary of their training content, including the top 10% of internet domains they scraped. The summary does not have to include prices or license terms.

### How can a publisher signal licensing terms to AI crawlers?

Publishers can block named AI crawlers in robots.txt, add a machine-readable usage signal such as Cloudflare's `Content-Signal`, or point to license terms with an RSL `License:` line. A licensing email in a robots.txt comment is readable by people, but crawlers skip it.

## The bottom line

- AI data licensing deals are now routine, but 78% of the 49 deals we counted disclose no price or formula.
- Payment is moving toward usage-based models, which are harder for outsiders to check.
- Publishers mostly block AI crawlers and leave licensing terms in comments. Watch the EU training data summaries: with enforcement possible since August 2, 2026, they are the first required public record of what was scraped.

## Sources

- Press Gazette, ["Who's suing AI and who's signing"](https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/), updated October 21, 2026
- The Decoder, ["Meta signs multi-year AI deal with News Corp worth up to $50 million a year"](https://the-decoder.com/meta-signs-multi-year-ai-deal-with-news-corp-worth-up-to-50-million-a-yearmeta-signs-multi-million-deal-with-news-corp-for-ai-training-and-content)
- The Register, ["Really Simple Licensing spec takes"](https://www.theregister.com/2025/12/10/really_simple_licensing_spec_takes/), December 10, 2025
- RSL Collective, [Really Simple Licensing standard](https://rslstandard.org/)
- European Commission, ["Template for general-purpose AI model providers to summarise their training content"](https://digital-strategy.ec.europa.eu/en/faqs/template-general-purpose-ai-model-providers-summarise-their-training-content)
- Hogan Lovells, [AI litigation case law tracker: Robert Kneschke v. LAION e.V.](https://digital-client-solutions.hlc.com/resources/ai-litigation-case-law-tracker/case/robert-kneschke-vs-laion-e-v/decision/1)
- Plesner, ["High Court rules natural language opt-out of text and data mining insufficient"](https://plesner.com/en/news/high-court-rules-natural-language-opt-out-text-and-data-mining-insufficient), May 12, 2026
- Mayer Brown, ["EU AI Act News: Rules on General-Purpose AI Start Applying, Guidelines and Template for Summary of Training Data Finalized"](https://www.mayerbrown.com/en/insights/publications/2025/08/eu-ai-act-news-rules-on-general-purpose-ai-start-applying-guidelines-and-template-for-summary-of-training-data-finalized), August 1, 2025
- PPC Land, ["Meta delays website-crawling terms to Nov 20 after publisher complaints"](https://ppc.land/meta-delays-website-crawling-terms-to-nov-20-after-publisher-complaints/), October 23, 2026
- Massive, tally of signed deals in Press Gazette's tracker, October 26, 2026 (first-party analysis)
- Massive, robots.txt survey of 47 news sites, October 26, 2026 (first-party test)