AI Content Deals Beyond OpenAI: Amazon, Meta, Microsoft, Anthropic and Apple

Explore with AI

Open in ChatGPTOpen in ClaudeOpen in PerplexityOpen in Google AI Mode

Last updated: August 10, 2026

When people talk about AI content licensing, the conversation usually starts and ends with OpenAI. That is understandable: OpenAI has signed the most deals and the largest one on record (News Corp, reportedly worth more than $250 million). But OpenAI is not the whole map. Amazon, Meta, Microsoft, Anthropic, Mistral and Apple each pay for (or pointedly refuse to pay for) training and grounding data in very different ways, and those choices shape which sources their assistants can cite.

This post covers everyone except the three players who already have their own dedicated breakdowns. For their deals, see OpenAI’s publisher deals, Google and Gemini’s publisher deals, and the Perplexity Publishers’ Program. For the full picture across all vendors, the AI content licensing hub ties it together, and the AI copyright lawsuits tracker covers the litigation side. Everything below is dated to July 2026 and labeled “reported” or “undisclosed” wherever the numbers are not officially confirmed.

Why these deals matter for visibility

A licensing deal does two things. First, it can feed a model’s training data. Second, and increasingly more important, it can wire a publisher into the assistant’s live “grounding” layer, the set of sources an assistant retrieves and cites when it answers a real-time question. A brand that is licensed into Meta AI or Amazon’s Rufus is far more likely to be surfaced, quoted and linked than one that is not. That is why the map of who signed with whom is not trivia: it is a map of which sources each assistant is structurally biased to trust.

The deals at a glance (signed)

These agreements have been confirmed by the companies or reported as signed. Values are marked “reported” when they come from press reports rather than official disclosure, and “undisclosed” when neither side has put a number on them.

Amazon: the quiet aggregator

Amazon was slow to sign public content deals, then moved fast in 2025. Its first was with The New York Times, announced on May 29, 2025. The multi-year agreement covers Times news articles, NYT Cooking recipes and The Athletic’s sports coverage, and lets Amazon surface that material across its AI products, including Alexa, with direct links back to Times properties. It was the Times’ first generative-AI licensing arrangement, notable because the paper is simultaneously suing OpenAI and Microsoft over training. Amazon and the Times did not disclose terms; press reports (first surfacing in July 2025) put the value at roughly $20 million to $25 million per year. Treat that figure as reported, not confirmed.

About six weeks later, Amazon signed Condé Nast and Hearst, this time to feed Rufus, its LLM-powered shopping assistant. Those deals bring Vogue, GQ and The New Yorker (Condé Nast) plus Cosmopolitan, Good Housekeeping and Harper’s Bazaar (Hearst) into Rufus’s product and recommendation answers. Terms are undisclosed. Amazon is also a longtime customer of Shutterstock’s AI image-licensing program, alongside most of the other majors.

One thing to watch: in February 2026, multiple outlets reported that Amazon is exploring a marketplace where media sites could sell their content to AI companies. As of July 2026 that is reported and exploratory, not a launched product, so it does not belong in the “signed” column yet.

Microsoft: paying through the product, then the marketplace

Microsoft’s posture is split. It is a co-defendant with OpenAI in the Times lawsuit, yet it has also become one of the more structured buyers of licensed content, largely because it wants clean data flowing into Copilot.

Its clearest training deal is with HarperCollins, announced in November 2024: a three-year agreement to train an AI model on select previously published nonfiction titles. The economics are unusually transparent for this space. HarperCollins offered $5,000 per book, split 50/50 between the publisher and the author, authors had to opt in, and the terms capped usage at “no more than 200 consecutive words and/or five percent of a book’s text.” Fiction was excluded. HarperCollins was the first Big Five publisher to strike an AI training deal.

On the news side, Microsoft launched Copilot Daily in October 2024 and pays a roster of publishers (the Financial Times, Reuters, Axel Springer, Hearst and the USA Today Network) to feed its licensed news summaries. Terms are undisclosed. In February 2026 Microsoft went a step further and launched its Publisher Content Marketplace (first previewed at a Monaco publisher summit in September 2025), a “click-to-sign,” pay-per-use platform where publishers set their own terms. Early participants include the Associated Press, Condé Nast, Hearst, Business Insider, USA Today and Vox Media. It is the clearest sign yet that grounding data, not just training data, is becoming a metered commodity.

Meta: one deal, and a lot of litigation

Meta is the clearest example of the “sue, don’t license” school. It has exactly one notable news licensing deal: Reuters, announced October 25, 2024. Under it, Meta AI can draw on Reuters reporting to answer news and current-events questions, with responses linking back to the underlying Reuters story. Terms are undisclosed and the deal is described as multi-year. It was Meta’s first news publisher deal in years. Meta is also a Shutterstock image-licensing customer.

Beyond that, Meta’s story is legal, not commercial. Court filings in the authors’ case allege that Meta briefly considered licensing deals with major publishers in 2023 and then deliberately abandoned them, with one internal note reasoning that licensing even a single book would undercut its “fair use” defense. Meta trained its Llama models partly on the LibGen collection of pirated books. In Kadrey v. Meta, Judge Vince Chhabria ruled on June 25, 2025 that Meta’s training use was fair use, but he framed it narrowly: the plaintiffs, he wrote, “made the wrong arguments” and a better record of market harm “very well might” have gone the other way. In May 2026, a fresh class action (led by Scott Turow, with Hachette, Macmillan, McGraw Hill and others) was filed against Meta over the same underlying conduct. For the deeper legal thread, see the AI copyright lawsuits tracker.

Mistral: Europe’s one big deal

Mistral, the Paris-based OpenAI challenger, has been thin on content deals, with one significant exception: Agence France-Presse (AFP), announced January 16, 2025. The multi-year agreement lets Mistral’s assistant, Le Chat, tap AFP’s entire text archive back to 1983, roughly 2,300 stories a day across six languages. Photos and videos are excluded, and the value was not disclosed. It was the first deal of its kind for both AFP and Mistral, and it landed one day after Google announced its own agreement with the Associated Press. It is genuinely the standout in Mistral’s portfolio: outside AFP, the company’s licensing footprint is light.

Anthropic: almost no licensing, one proposed settlement

Here is where honesty matters more than padding. Anthropic, the maker of Claude, has done very few public content-licensing deals. It is not on the news-publisher deal sheets that OpenAI, Microsoft, Amazon and Google dominate. What defines Anthropic instead is litigation.

In Bartz v. Anthropic, authors alleged the company trained Claude on millions of pirated books from the LibGen and PiLiMi collections. The parties proposed a $1.5 billion settlement, and Judge William Alsup granted preliminary approval on September 25, 2025. Final approval remained pending after the May 2026 fairness hearing. If approved, the settlement would cover roughly 500,000 works, require Anthropic to destroy the pirated datasets, resolve past conduct, and create no ongoing licensing scheme. It is therefore a proposed legal settlement, not a commercial content license.

Apple: mostly talks, one signed image deal

Apple’s news-content strategy is, as of July 2026, mostly reported talks rather than signed agreements. In late 2024, press reports (including The Information and The New York Times) said Apple had approached major publishers, among them Condé Nast, NBC News and IAC, with multi-year offers “worth at least $50 million” to license news archives for AI training. Reception was mixed, with publishers wary of legal exposure and unclear on how Apple would use the content. As far as public reporting shows, those did not convert into announced deals.

The one signed Apple agreement in this space is with Shutterstock, reported in April 2024, for millions of images for AI training, at a reported $25 million to $50 million. So the accurate framing is: Apple has a signed image deal and a set of reported news talks. Do not treat the Condé Nast / NBC / IAC discussions as done deals.

Signed vs. only in talks

• Signed: Amazon-NYT, Amazon-Condé Nast, Amazon-Hearst, Meta-Reuters, Microsoft-HarperCollins, Microsoft’s Copilot Daily roster, Microsoft’s Publisher Content Marketplace, Mistral-AFP, Apple-Shutterstock, Reddit-Google, Reddit-OpenAI, Stack Overflow-Google/OpenAI.
• Reported talks or exploratory, not signed: Apple’s news-publisher discussions (Condé Nast, NBC News, IAC); Amazon’s exploratory AI content marketplace.

The data and platform deals underneath

Some of the most consequential agreements are not with newspapers at all. They are with the platforms that sit on huge pools of user-generated or stock content.

Reddit is the clearest example. In February 2024, on the same day it filed to go public, Reddit signed a data-licensing deal with Google reported at $60 million a year, giving Google access to real-time Reddit content. A few months later it struck a similar deal with OpenAI, reported at roughly $70 million a year. By September 2025, Reddit was reportedly pushing both partners toward usage-based pricing that scales as its content becomes more central to AI answers. Reddit’s specific arrangements are covered in more depth in the OpenAI and Google deal breakdowns.

Shutterstock quietly built a whole business here, licensing image and media training data to Meta, Google, Amazon, Apple and OpenAI, with initial per-customer deals reported in the $25 million to $50 million range. Its AI licensing line generated about $104 million in 2023 and it has guided toward far larger figures since. Stack Overflow licensed its developer Q&A corpus (through its OverflowAPI) to both Google and OpenAI in spring 2024 on a non-exclusive basis, with terms undisclosed.

What this means for your AI visibility

The through-line across all of these companies is simple: the assistants people use every day are being wired, deal by deal, to prefer specific sources. When Amazon licenses The New York Times and Condé Nast, Rufus and Alexa become more likely to cite them. When Microsoft’s marketplace meters access per use, the publishers inside it get surfaced and the ones outside it fade. Meta’s answers lean on Reuters; Mistral’s lean on AFP. If your brand or your category is not in the pipeline that feeds a given assistant, you can be invisible in that assistant’s answers no matter how strong your website is.

That is exactly the layer LLM Pulse measures. It tracks which sources each assistant actually cites, whether your brand is mentioned, and how you rank against competitors across ChatGPT, Perplexity, Gemini, Google AI Mode and Google AI Overviews. Sentiment analysis is included, while Claude, Copilot, Grok, DeepSeek and Alexa for Shopping (formerly Amazon Rufus) are optional paid add-ons on every paid plan. When a new licensing deal lands and a competitor suddenly starts getting cited, you see the citation and share-of-voice shift instead of guessing. If you want the practical playbook, our guides on how to rank in ChatGPT and AI search optimization pair well with this map.

The bottom line

Outside the big three, the strategies could not be more different. Amazon has become one of the most active buyers, aimed squarely at Alexa and Rufus. Microsoft is building metered infrastructure for grounding content. Meta signed once and litigated the rest. Mistral has one flagship deal (AFP). Anthropic has almost no licensing and a proposed settlement still awaiting final approval. Apple is still mostly talking. As these deals harden, they will keep redrawing which sources each assistant can see, which is why watching the map, and your own place in it, matters.