AI licensing deals and SEO: Which publishers feed ChatGPT, Gemini, Copilot, and Meta AI

Since 2023, AI companies have signed dozens of content deals with publishers. OpenAI has one with News Corp, Google with Reddit, and Amazon with The New York Times. Each contract decides what an assistant is allowed to quote, summarize, and link to, and sometimes what it was trained on.

Why does that matter for SEO? These deals are one of the few public clues about which sources each assistant can use freely. If your brand’s visibility in AI answers depends on where you get mentioned, it helps to know which publications have a direct line to which assistant.

In this guide, you’ll find the partnerships we know about as of Q4 2026, what the money buys, and what you can do with that information. Deals change often, so think of the tables below as a current snapshot.

The AI + publisher partnership map

OpenAI started signing publisher deals in 2023 and still has the most partners, with 18 in the table below. It isn’t the only buyer anymore. Google, Microsoft, Meta, and Amazon have each signed their own deals since 2024, and each one works with a different mix of publishers.

Not every deal you’ll see reported is a content license. OpenAI gave $5 million to the American Journalism Project to support local newsrooms. OpenAI and Microsoft together funded a $10 million program at the Lenfest Institute that pays for AI fellows in US newsrooms. Microsoft sponsored an AI-assisted news feed at Semafor. These are all grants and sponsorships; they fund journalism, but they don’t give an assistant the right to use anyone’s articles.

Then there are the publishers who went to court. The New York Times sued OpenAI and Microsoft in December 2023, then signed with Amazon about a year and a half later. News Corp manages both at once: It licenses to OpenAI and Meta while its Dow Jones and New York Post units sue Perplexity.

What the AI + publisher partnership map shows

Eighteen publisher groups have deals with two or more AI companies, and seven of them have three or more.

Eleven more have two deals: News Corp, The Washington Post, Le Monde, Condé Nast, AP, Reuters, Reddit, Time, The Atlantic, Der Spiegel, and the Washington Examiner.

A few observations stand out.

• The deals aren’t exclusive. For example, the Financial Times sells to four buyers. For anyone doing digital PR, that changes the math on a placement: One story in a multi-deal publication is licensed content for several assistants at once.
• There’s a licensed core. Out of thousands of news and magazine brands, a group small enough to fit on one slide accounts for most of the overlap. If you were building a list of publications that matter for AI visibility from scratch, these eighteen would be a reasonable first draft.
• The deal structure keeps changing. The first wave of deals in 2023 and early 2024 was about training data. AP licensed its archive to OpenAI and Reddit sold training access to Google. The second wave was about display, meaning summaries and links inside ChatGPT, Copilot, and Meta AI. The newest deals pay by usage. Microsoft’s marketplace charges per use, Perplexity shares subscription revenue, and Google is piloting a program that pays publishers when their content shapes an AI response.

What the AI + publisher partnership money buys

The most useful distinction here is between training rights and retrieval rights.

Training rights let an AI company feed a publisher’s content into their AI model itself. That shapes what the model “knows” without looking anything up, including what it knows about your brand. This body of knowledge changes slowly over time and you can’t audit it from the outside.

Retrieval rights let an assistant fetch a publisher’s content at the moment someone asks a question, then summarize it and link to it. This is the part you can see: the citations under a ChatGPT answer or the source links in Meta AI. It’s also the part you can influence in weeks instead of years.

Plenty of deals bundle the two. Axel Springer’s agreement with OpenAI covers both: summaries with attribution and links, even for paywalled articles, plus training. Meta’s deal with News Corp reportedly covers real-time retrieval and training as well.

The money moves in three ways:

• Flat fees. News Corp’s deal with OpenAI was reported at close to $250 million over five years. Its deal with Meta is reportedly worth up to $50 million a year for three years. Amazon is said to pay The New York Times $20 million to $25 million a year, and Google pays Reddit about $60 million a year.
• Revenue share. Perplexity’s Comet Plus passes 80% of subscription revenue to publishers. ProRata splits revenue based on which sources an answer drew on.
• Pay-per-use. Microsoft’s publisher marketplace launched with Copilot as its first buyer, paying each time content is used.

Some caution is warranted on those numbers: Almost none of them were disclosed by the companies. They come from press reports citing unnamed sources, and the contracts themselves are private, non-exclusive, and time-limited. Use them for a sense of relative scale.

Why AI + publisher partnerships matter for SEO

A license guarantees access, but nobody has shown it guarantees preference

When an AI company licenses a publisher’s content, it pays for permission to use that publisher’s articles. You might expect the assistant to cite those publishers more often as a result. That may be true, but nobody outside the AI companies knows for sure. So treat it as an idea to test, not a fact: Run your own prompts in your category and see which sources come up.

What a license does guarantee is that the assistant can get to the content. Think of the difference between a page being indexed and a page ranking. Being indexed doesn’t mean you’ll rank, but you can’t rank without it. A license works the same way: It doesn’t promise a citation, but it makes sure the publisher’s articles are always available to the assistant.

That matters because assistants can’t always reach other websites. A paywall can block them, and so can a robots.txt rule that turns their crawler away. Licensed publishers avoid both problems. They often send their articles to the AI company directly, through a feed or API, and the assistant has written permission to quote them.

Your own website has no such arrangement. So when a licensed publication writes about your brand, that article may be the most dependable way for an assistant to learn about you and quote what it says, even if it can’t reach your site.

A deal doesn’t mean you’ll be cited correctly

In March 2025, the Tow Center for Digital Journalism tested eight AI search tools on their ability to identify the source of news excerpts. Together they got more than 60% of queries wrong.

A licensing deal didn’t fix it. ChatGPT correctly identified just one of ten excerpts from the San Francisco Chronicle, a Hearst paper covered by an OpenAI partnership. Perplexity Pro pointed to syndicated copies of Texas Tribune articles in three of ten cases, even though the Tribune is a Perplexity partner. Time, which has deals with both OpenAI and Perplexity, was among the most accurately identified publishers, yet no tool got it right every time.

For SEOs, two of these failures are old problems in a new place. The first is broken links. More than half of the responses from Gemini and Grok 3 cited URLs that were made up or led to error pages, so even when the assistant named a source, readers couldn’t reach it. The second is syndication. When Perplexity Pro credited copies of Texas Tribune articles instead of the originals, it made the same mistake a search engine makes when it picks the wrong canonical version of a page.

The tools have changed since that test, so the exact figures will be out of date. The takeaway holds, though. A citation is only worth something if it names the right source and links to a page that loads. A partnership doesn’t guarantee either, so check for yourself:

• Click every citation. Run your brand and product prompts in each assistant and follow the links. Does each one resolve? Does it land on the right page?
• See which version gets the credit. Note whether the assistant cites the original article or a syndicated copy. If you distribute press releases or contributed articles, find out which URL the assistant points to.
• Look for AI traffic hitting 404s. Check your analytics for visits from AI assistants that land on error pages, and redirect the URLs that keep coming up.
• Check coverage in partner publications too. When a licensed outlet writes about you, confirm the assistant attributes the story to that outlet and describes it accurately.
• Repeat every quarter. These products change faster than any published study can track.

There is no single “AI search” to optimize for

Each assistant works from a different set of licensed sources. Le Monde is licensed to OpenAI and Meta. Der Spiegel works with Perplexity and Google. The New York Times licenses to Amazon and took OpenAI to court.

That’s a real change from the search you’re used to. With Google holding most of the market, one strong placement in a respected publication helped you everywhere that mattered. In AI search, the same article can be licensed content for one assistant and just another page on the open web for the next.

So before you decide which publications to target, find out which assistants your audience uses. A brand whose buyers ask ChatGPT for recommendations needs a different media list from one whose customers spend their day in Instagram and WhatsApp, where Meta AI lives. Here’s how to work it out:

• Check your referral data. Look for visits from chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com. The mix shows you where your audience already asks questions.
• Match each assistant to its partners. If ChatGPT leads, OpenAI’s list is your starting point. If you sell on Amazon, look at the Rufus partners Condé Nast and Hearst. If your customers are on Meta’s apps, look at CNN, USA Today, and People Inc.
• Factor in your markets. Licensing is uneven by country and language. France has Le Monde, Le Figaro, and AFP. Germany has Axel Springer, Der Spiegel, and Süddeutsche Zeitung. Spain has Prisa, and the Nordics have Schibsted. If you do international SEO, build a list for each market.
• Report by assistant. A single blended “AI visibility” number hides these differences. Track citations and mentions separately for each assistant you care about.
• Use overlap to save effort. Publications with several deals, like the Financial Times or The Guardian, cover more than one assistant with a single placement.

Reddit has two of the biggest deals

Reddit licenses its content to both Google and OpenAI. The OpenAI deal gives ChatGPT real-time, structured access through Reddit’s Data API.

That makes what people say about your brand in relevant subreddits a direct input to two of the largest assistants. It’s a strong argument for taking community presence seriously, and a strong argument against trying to fake it.

Review and lifestyle publishers now feed shopping answers

Amazon licensed Condé Nast and Hearst content for Rufus, its shopping assistant. Hearst’s side of that deal includes Good Housekeeping, Cosmopolitan, and Harper’s Bazaar. People Inc., whose brands include Allrecipes, Food & Wine, and Verywell Health, describes itself as Meta’s first lifestyle content partner.

If you sell consumer products, the roundups and reviews in those titles are now licensed inputs to assistants that recommend things to buy. You may want to consider working with them as worth the investment to get more eyeballs on your brand. 

Publishers are signing because search traffic is falling

People Inc. says Google search went from about 54% of its traffic to 24% in two years. That’s the backdrop to every deal on this page.

It also changes what earned coverage is worth. A feature in one of these publications may send you fewer clicks than it did in 2023, while reaching more people through the assistants that license it — and therefore netting you more qualified traffic.

What to do with these AI + publisher partnerships

Rework your digital PR target list

Add two columns to whatever list you use to prioritize outreach: which assistants license each publication and how many. Weigh that alongside relevance and authority, not in place of them. A trade title your buyers read every week still beats a passing mention in a licensed giant that has nothing to do with your market.

Where the two overlap, move those outlets up. A relevant placement in a publication with three or four deals is about as efficient as earned media gets right now.

Set crawler policy one company at a time

The controls differ by company, and the differences matter.

OpenAI runs separate crawlers, and each setting is independent. GPTBot collects content that may be used for training. OAI-SearchBot decides whether you can appear in ChatGPT’s search answers. You can block the first and allow the second. Block OAI-SearchBot and you drop out of ChatGPT search answers.

Google has changed the most. Until this year, the only way out of AI Overviews was to limit your snippets across Search. Google now has a Search Console toggle that removes a site from AI Overviews and AI Mode while keeping it in regular results. It started as a UK test in June 2026 and reached all sites worldwide on August 31.

Three details to know before you touch it:

• Google says the setting won’t be used as a ranking signal for regular results.
• Opted-out sites get no impressions or traffic from Google’s AI features.
• It doesn’t cover the Gemini app. Google-Extended remains the control for training and grounding in Google’s other systems.

Should you opt out? For most brands, probably not. You aren’t being paid either way, and being present in the answer is the whole point. For publishers whose business depends on the click, it’s a harder call, and there isn’t enough data yet to make it with confidence.

Watch the usage-based payment programs

Google’s contribution pilot has been offered to dozens of publishers and extends beyond news organizations. Microsoft’s marketplace is pay-per-use by design.

Google’s pilot is by invitation for now. But if your site publishes original data, research, or reference content that assistants lean on, payment is no longer a far-fetched idea. It’s worth a calendar reminder to check each quarter whether eligibility has widened.

Be careful with sponsored content

If you buy an advertorial on a licensed site, that page may flow into assistants like any other article. It isn’t clear how assistants treat labeled sponsored content once it’s in the feed.

Some marketers will read that as a loophole. It’s more likely a liability. Assistants that repeat paid claims as editorial fact are a problem the platforms will want to fix, much as Google eventually did with site reputation abuse. Label sponsored content clearly and don’t build a strategy on it.

Keep your brand description consistent

An assistant that summarizes your company from five licensed sources will reproduce whatever those sources disagree on. If one calls you a payments startup and another calls you a banking platform, expect a muddled answer.

Use the same one-line description, the same category, and the same key facts in your press materials, your site, and your executive bios. It’s unglamorous work, and it’s among the few things here that are fully in your control.

How to measure your AI visibility

None of this is worth acting on if you can’t tell whether it’s working. Four measurements cover most of what you need.

Citation share by assistant

Build a fixed set of prompts for your category, somewhere between 50 and 100, and run it through each assistant on a schedule. Record which domains get cited and where your brand is mentioned. Tools will do this for you at scale, but a spreadsheet and an afternoon a month is enough to start.

Search Console’s AI reports

Alongside the opt-out toggle, Google added reporting on AI features: impressions, which of your pages appear in AI responses, and in which countries. First-party data on AI visibility has been scarce, so use it.

Server logs

Look for OpenAI’s user agents. OAI-SearchBot tells you your pages are being indexed for ChatGPT search. ChatGPT-User is more interesting, because it fires when a person’s question sends ChatGPT to your page.

The partner test

Take the cited domains from your prompt set and tag each one as licensed or unlicensed for that assistant. Then compare against the regular search results for the same queries. If partners show up far more in the AI answers than in the organic results, you have evidence that licensing shapes citations in your vertical. If they don’t, you’ve saved yourself from chasing a pattern that isn’t there.

Risks and open questions for AI partnerships

A few things could make parts of this guide wrong within a year.

The courts continue to hear copyright suits

Copyright suits such as The New York Times’s case against OpenAI and Microsoft bear directly on what AI companies can do without a license. A ruling that makes unlicensed use harder makes every deal on this page more valuable. A ruling that makes it easier does the reverse.

Some of these deals expire soon

The terms we know of run three to five years, and the earliest deals date from 2023. A target list built on today’s map will go stale sooner than you’d like, which is why the re-test step below matters.

Google’s pilot may not mean what it seems to

Google describes its December 2025 deals as payment for extended display rights and content delivery, not as licensing deals, and they build on the existing Google News Showcase program. The features being tested live in Google News. Don’t assume those partners get any preference in AI Overviews.

Nobody knows the long-run cost of opting out

The worldwide toggle has existed since August 31, 2026. Any confident claim about what leaving AI Overviews does to total traffic is running ahead of the data.

Payment without clicks

Usage-based programs reward content that shapes an answer, whether or not anyone visits the source. That’s good for publishers’ revenue and uncertain for everyone who relies on those publishers sending readers their way.

A five-step framework

If you take one process away from this guide, make it this one.

1. Audit. Run your prompt set and record which sources each assistant cites in your category.
2. Map. Compare those sources against the partnership table above. Note which are licensed, and to whom.
3. Prioritize. Shift PR, community, and content effort toward the sources that are both relevant to your buyers and present in the answers.
4. Set policy. Decide, company by company, which crawlers and AI features your site allows. Write the decision down, along with the reason.
5. Re-test every quarter. Deals get signed, expire, and change shape. So will the citations.

The partnership map won’t tell you what an assistant will say about your brand. It tells you where each assistant has paid to look, and that’s a better starting point than guessing where your PR might result in citations or mentions.

Keep reading about AI SEO and how you can maximize your visibility in this new search landscape.