Three different things get folded into the claim that a licensing deal makes a publisher more visible to AI: how often an assistant cites you, how often a user then clicks through to your site, and whether any money changes hands. Until August 2026 nobody had measured the first one at scale. Now somebody has, and the three turn out to move independently. Citation share tilts toward the platform you signed with. Clickthrough has collapsed for licensed and unlicensed publishers alike. Payment remains available only to the few hundred domains that were offered a contract in the first place. A publisher deciding whether to chase a deal for visibility reasons is buying one of those three things, not all of them.
What the study actually measured
The evidence comes from a joint analysis published on 20 August 2026 by Press Ranger, a press release distribution platform, and OtterlyAI, an AI search monitoring tool. It joins two first-party datasets that had not previously been put together.
OtterlyAI supplied the citation side: 129.3 million citations across more than 20 million cited URLs, captured during June 2026 on ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Gemini and Claude. Press Ranger supplied the deal side: a database of 91 confirmed licensing agreements between AI companies and publishers, mapped to 314 publisher domains, with every record backed by a public source through 28 July 2026.
The critical methodological choice is that the headline figure is normalised. It counts citations per cited page rather than total citations, which stops the result from simply restating that large publishers publish more pages. That is a meaningful control, and it is the reason the study is worth reading at all.
The 48 per cent premium, and where it does not appear
On ChatGPT, pages from publishers holding an OpenAI licensing deal collected an average of 10.2 citations each. Pages from publishers without one collected 6.9. That is the 48 per cent premium.
Across all seven platforms combined the OpenAI cohort's advantage was 46 per cent, at 10.7 citations per cited page against 7.3. Isolating publishers that had signed with OpenAI and nobody else sharpens the effect considerably: those outlets earned 112 per cent more citations per page on ChatGPT than unlicensed publishers.
The interesting part is what happens on the other platforms. OpenAI was the only licensor whose deals showed a clear home-platform advantage. Google-licensed publishers were cited on Google AI Overviews at a slightly lower rate than comparable unlicensed publishers. Perplexity's licensed publishers landed at parity on Perplexity. Whatever is happening on ChatGPT is not a general property of licensing.
It is also concentrating. Publishers who signed with OpenAI now draw 57.9 per cent of their entire AI citation volume from ChatGPT alone, while unlicensed publishers keep a more even spread across ChatGPT and Perplexity. Thomas Peham, chief executive of OtterlyAI, put it plainly: "A licensing deal does one clear thing: it tilts your citations toward ChatGPT. It does not guarantee you show up everywhere, and it is not the only way to win."
A tilt is not the same as a lift. A publisher whose citations concentrate on one platform has swapped a diversified position for a dependent one, which is a familiar trade in this industry and has rarely aged well.
Why this is a correlation and not a proof
The study does not show that signing a deal caused the gap, and its own coverage in the trade press was careful about this. The Media Copilot's report of 24 August 2026 set out the limits clearly.
OpenAI has largely signed prominent, English-language news organisations that were plausibly more likely to appear in AI answers before any contract existed. Normalising by citations per cited page helps with size, but it does not control for brand authority, topic mix, publishing cadence, link profile or editorial format. It is entirely possible that the causal arrow runs the other way, and that OpenAI signed the publishers its systems already leaned on.
Four further caveats are worth carrying. The analysis covers a single month. It has not been peer reviewed. No methodology has been published explaining how comparable unlicensed publishers were selected, which is the load-bearing assumption in the whole exercise. And both publishing companies sell products in the adjacent market, one in distribution and one in AI visibility monitoring, which is not disqualifying but is a reason to read the framing as well as the numbers.
The honest summary is that this is the best available evidence and it is one study.
Citations went up. Clicks went down.
Even if the premium is real and causal, it buys a smaller prize than it did a year ago.
In October 2025, David Buttle wrote in Press Gazette that publishers with OpenAI licensing deals were seeing a ChatGPT clickthrough rate almost seven times higher than publishers without agreements, drawing on Tollbit's State of the Bots report. The mechanism was visible: the deals that secure real-time access also secure the large feature boxes ChatGPT displays for certain news prompts. Buttle's conclusion was blunt. "Get a deal and you'll get paid and get (some) traffic. If you don't have a deal, you'll get neither."
That advantage did not hold. Tollbit's Q3 and Q4 2025 report, covered by The Register in February 2026, found that clickthrough rates for websites with AI licensing deals had fallen to 1.33 per cent in Q4, a 6.5-fold decrease across the year. Clickthrough from AI apps generally fell from 0.8 per cent in Q2 2025 to 0.27 per cent in Q4. Over the same period the ratio of bot to human visits went from roughly one AI bot visit per 200 human visits in Q1 to one per 31 by Q4, and human visitors declined 5 per cent between Q3 and Q4.
The composition of that bot traffic matters for anyone thinking about licensing as a training-data question. Training scrapes fell 15 per cent between Q2 and Q4 2025 while retrieval bot traffic rose 33 per cent and AI search indexer traffic rose 59 per cent. The volume has shifted from ingestion to live retrieval. Olivia Joslin, Tollbit's cofounder and chief operating officer, told The Register that "AI will become the primary reader of the Internet" and that "the odds are very much stacked against publishers".
Set the two datasets side by side and the position is this: a licence appears to correlate with being cited more often on ChatGPT, in a year when being cited on ChatGPT sends almost nobody to your site. Citation is now closer to a branding event than a traffic event.
There is a quality caveat on citation volume too. The Tow Center's 2024 examination of ChatGPT citations, reported by TechCrunch, found the model misattributing and fabricating sources, including for publishers that held licensing agreements. A count of citations is not a count of correct citations.
The finding that matters most to unlicensed publishers
The headline of the study is about the licensed. The most useful result is about everyone else.
News is a small share of what AI cites at all, at 7.2 per cent of total citations. Within that slice, concentration is severe: five media groups, Future plc, Forbes, People Inc., Conde Nast and Hearst, capture 69 per cent of all citations going to licensed publishers. A deal is not a level playing field even among those who have one.
Now the counterweight. Across 16 United States industries the study examined, niche and trade outlets, most of them unlicensed, carried the majority of news citations in 15 of them, collecting 213 per cent more AI citations than mainstream media. The most-cited news domains in the dataset include unlicensed sites such as NerdWallet, Healthline and Bankrate. And the content assistants cite most is commercial and evergreen: best-of lists, buying guides and product reviews accounted for 46.9 per cent of citations to licensed publishers.
Steve Beyatte, founder of Press Ranger, named this as the real result. "The bigger surprise is not that OpenAI's deals pay off on ChatGPT, it is that the largest opening for PR teams sits with the trade and niche outlets nobody licensed."
That is the finding a specialist publisher should act on. Depth on a defined subject beats general authority in retrieval, because retrieval is question-shaped rather than brand-shaped. A trade title with genuine coverage of a narrow area is competing on the axis assistants actually score, and it does not need a contract to do it.
What drives citation if you do not have a deal
Nothing in the study suggests a licence is a prerequisite for AI visibility, and its own recommendations point elsewhere. The levers available to an unlicensed publisher are the ordinary ones, and they are not mysterious.
Be retrievable. A page that cannot be fetched and parsed cannot be cited, whatever its quality. Answer specific questions completely and early in the page, because assistants extract self-contained passages. Publish the service formats the data shows assistants prefer. Cover a subject deeply enough that you are the obvious source on it. And measure which assistants cite you for which prompts, rather than assuming the pattern is uniform, because the study shows it is not.
Those are the same levers a licensed publisher pulls. The licence, on this evidence, adds a platform-specific tilt on top of them rather than replacing them.
What a citation is worth, and what it is not
A citation is a retrieval event. Something fetched your page, used it, and named you. On current clickthrough rates it is unlikely to bring a reader, and unless you hold a contract it does not bring a payment either. The industry has spent three years arguing about whether licensing changes what AI reads. The narrower and more awkward question is what a publisher gets when AI reads them and does not send anyone back.
This is the gap blankspace works in. Our infrastructure sits at the CDN edge, identifies Live Search Agent traffic as it arrives, and treats that request as a monetisable event in its own right rather than as a lost pageview. It is a different transaction from a licence: a licence prices access to the archive as an input, while edge monetisation prices the live retrieval. A publisher can reasonably pursue both, and neither is a substitute for the other. It is worth saying plainly that no vendor in this category, ours included, can guarantee that a model will honour what it is served, and the layer remains contested.
The decision framing this study supports is narrow. If a deal is on the table, take it for the money and the certainty, not for the citations, because the citation effect is one-platform, unproven as causation and worth less traffic than it was. If a deal is not on the table, which is the position of almost every publisher reading this, the evidence says citation share is still winnable without one, and that specialists are already winning it.
Frequently asked questions
Does an AI licensing deal guarantee more citations?
No. The one study to measure this found a 48 per cent citation premium on ChatGPT for publishers with an OpenAI deal, but no comparable advantage for Google-licensed publishers on Google AI Overviews or for Perplexity licensees on Perplexity. The effect appears to be specific to OpenAI and to ChatGPT, and the study establishes correlation rather than causation.
Do licensed publishers get more traffic from AI assistants?
Much less than they used to. Tollbit data reported in October 2025 put the ChatGPT clickthrough rate for OpenAI-deal publishers at roughly seven times that of publishers without deals. By Q4 2025 the same source put licensed sites at a 1.33 per cent clickthrough rate, a 6.5-fold fall across the year. Referral rates from AI apps generally fell to 0.27 per cent over the same period.
Can an unlicensed publisher get cited as often as a licensed one?
In some categories they already are cited more. The Press Ranger and OtterlyAI study found that across 16 United States industries, niche and trade outlets, mostly unlicensed, carried the majority of news citations in 15 of them and collected 213 per cent more AI citations than mainstream media. Specialist depth appears to matter more than licensing status outside general news.
What kind of content gets cited most?
Commercial evergreen content. Best-of lists, buying guides and product reviews made up 46.9 per cent of citations to licensed publishers in the June 2026 data. This is service journalism that answers a specific purchase or comparison question, which is the shape of a large share of the prompts assistants receive.
Is being cited the same as being paid?
No, and conflating them is the most expensive mistake in this area. A citation records that an AI system retrieved and used your page. It carries no payment, and at current clickthrough rates it usually carries no reader either. Payment comes from a licensing contract, if you are offered one, or from monetising the retrieval event itself at the infrastructure layer.