TL;DR: Some news companies have signed paid contracts letting AI firms use their journalism. A new study looked at 129.3 million links that AI assistants showed next to their answers in June, and checked which of those links pointed at companies that had signed such a contract. Pages from publishers with an OpenAI contract got cited 48% more often inside ChatGPT. On Google, Perplexity and Gemini, the same publishers did no better than everyone else, and sometimes slightly worse. The advantage is real, and it stops at the edge of the platform that paid for it.
First, what is being counted here
When an AI assistant answers a question, it often reads a handful of web pages first and then lists them underneath. Those links are called citations. Being one of them is how a business gets seen at all now, since a lot of people never scroll down to the ordinary blue links.
A licensing deal is a commercial contract in which an AI company pays a publisher for the right to use its articles. News Corp, Axel Springer, the Financial Times and others have signed them. Nobody outside the two parties has ever known what a deal buys you in terms of actual visibility.
The study
Otterly.ai supplied citation data: 129.3 million citations recorded during June 2026 across 19,812,878 distinct cited URLs, captured on ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Gemini and Claude. Press Ranger supplied the other half, a register of 91 confirmed AI-publisher agreements mapped to 314 domains, with evidence collected through 28 July 2026. The two datasets were joined and announced on 20 August.
The authors are blunt about what this is: a descriptive one-month snapshot showing association, not causation. Licensed publishers are also, on average, large and well-linked, which would earn them citations with or without a contract. Read the numbers as a floor on the gap, not as proof of a mechanism.
The premium is real and it is local
On ChatGPT, a page from a licensed publisher averaged 10.2 citations. An unlicensed one averaged 6.9. That is the 48%. Publishers who signed with OpenAI and nobody else did better still, at 112% above unlicensed pages.
Then it falls apart everywhere else. On Google AI Overviews and AI Mode, Google-licensed publishers came in slightly below comparable unlicensed ones. Perplexity’s licensees landed at parity on Perplexity. Gemini showed no deal advantage either. Copilot looks like an exception at first glance, with an 85% premium, but 99% of Copilot’s citations to Microsoft licensees go to publishers who also hold an OpenAI deal. Publishers licensed by Microsoft alone averaged 2.8 citations per page against 6.3 for unlicensed, which is 56% worse.
Anthropic has disclosed no news licensing deals at all. Claude cites news more heavily than any other engine in the set, at 14.0% of its citations.
The downstream effect is a concentration problem. Licensed publishers now pull 57.9% of their AI citation volume from ChatGPT alone, while unlicensed publishers get 32.4% of theirs there and spread the rest.
The 92.8% underneath
For anyone who is not a newspaper, the more useful number sits further down the report. News and media accounted for 7.2% of all 129.3 million citations. Everything else went to other kinds of pages.
And when you look at what licensed publishers actually got cited for, it was mostly not journalism. Commercial evergreen content made up 58.8% of their cited pages: best-of lists at 38.1%, buying guides and money advice at 15.1%, reviews and tests at 5.6%. Dated news articles were 5.1%. Press releases and wire copy came in under 1%.
So the pages winning citations at licensed publishers are the same pages an ordinary commercial site can write. A ranked list. A buying guide with an actual recommendation in it.
Trade beats mainstream almost everywhere
The study split cited publications into trade press and mainstream press across 16 industries. Trade won 15 of them. Overall, specialist publications took 75.8% of citations against 24.2% for mainstream outlets, and every one of the seven platforms preferred trade, in a range from 69.5% to 82.2%.
Trade media is also mostly unlicensed. Only 17.2% of trade citations came from licensed publishers, against 54.7% within mainstream media. The specialist end of the web is winning citations without any contract at all.
What to do with this
Stop treating AI visibility as one number. A blended “we are cited in AI” figure would have completely hidden the finding above, because the effect exists on one surface and is absent or negative on five others. If you report a single score to a client, you cannot tell them whether a change helped ChatGPT and hurt nothing, or helped nothing anywhere.
Write for the page types that get cited. Ranked lists, comparisons and buying guides are what the engines pulled, even from publishers that were paid for news.
And if you are a niche or B2B site, the trade-media finding is the encouraging part of this study. Depth in one industry appears to beat general authority in the engines’ eyes, and that is buyable with work rather than with a contract.
Splitting measurement per engine is exactly the kind of thing nobody does by hand across a portfolio, because it means running the same prompt set against six different surfaces every week and diffing the results. Preferium measures four AI engines on every plan — ChatGPT, Claude, Perplexity and Gemini — and tracks Google AI Overviews and AI Mode separately, so a gain on one surface and a loss on another do not cancel out into a flat line. More on how the system works, and our earlier note on measuring AI citations covers why the underlying numbers move around so much between runs.
Key takeaways
- Pages from OpenAI-licensed publishers averaged 10.2 citations on ChatGPT versus 6.9 for unlicensed pages, a 48% gap; OpenAI-exclusive signers were 112% ahead.
- The premium did not transfer. Google, Perplexity and Gemini showed licensed publishers at or below parity, and Microsoft-only licensees were 56% behind.
- News is a small slice of AI citations at 7.2% of 129.3 million; 92.8% went to other content types.
- What licensed publishers got cited for was mostly commercial evergreen content: best-of lists 38.1%, buying guides 15.1%, dated news only 5.1%.
- Trade publications took 75.8% of citations and beat mainstream media in 15 of 16 industries, largely without licensing deals.
- The authors describe this as a one-month snapshot showing association, not causation.
