OpenAI licensing deals are commercial agreements in which OpenAI pays a publisher for the right to train on and surface its content inside ChatGPT, and new data has finally put a number on whether they change anything about citation rates. A 129.3-million-citation study published in August 2026 by Press Ranger and OtterlyAI found that pages from publishers with an OpenAI licensing deal earn 48% more ChatGPT citations than pages from publishers with no deal at all, a premium that widens to 112% among the publishers who signed with OpenAI exclusively. The finding matters well beyond the roughly two dozen media companies that have actually signed one, because it forces a question every site doing AEO eventually has to answer: is a licensing cheque the thing that is actually buying those citations, or is something else happening that a licensing deal cannot replicate for the rest of us.
What the study actually measured
The study compared citation volume across seven AI search platforms in June 2026, covering 129.3 million individual citations in total. Publishers with a confirmed OpenAI licensing deal averaged 10.2 citations per page on ChatGPT specifically, against 6.9 citations per page for publishers with no deal at all, the 48% gap that has driven most of the coverage. Publishers who had signed with OpenAI and no other AI company saw a noticeably larger effect again: 112% more citations per page than unlicensed publishers, roughly double.
OpenAI-only versus multi-platform deals
The gap between the 48% headline figure and the 112% figure for OpenAI-exclusive publishers is itself informative. Publishers who had also signed separate deals with other AI companies saw a smaller ChatGPT-specific premium than those who had signed with OpenAI alone, which is consistent with citation volume being redistributed across platforms rather than simply added on top. A publisher with deals everywhere gets more even coverage across engines; a publisher with a deal only with OpenAI gets concentrated coverage on ChatGPT.
- 10.2 citations per page on ChatGPT for licensed publishers, against 6.9 for unlicensed publishers, a 48% premium overall
- 112% more citations per page for publishers whose only AI licensing deal was with OpenAI
- 57.9% of licensed publishers' total AI citation volume came from ChatGPT alone, against a more even split across ChatGPT and Perplexity for unlicensed publishers
- The dataset covers seven AI search platforms and 129.3 million citations recorded across June 2026
Correlation, not proof that the deal did it
OtterlyAI's own chief executive, Thomas Peham, was careful about how he framed the result: a licensing deal 'does one clear thing: it tilts your citations toward ChatGPT', which is a claim about direction, not about causation. The study cannot show that signing a deal caused the higher citation count, only that publishers with a deal and publishers without one look different on this metric.
The obvious confound
Publishers who sign OpenAI licensing deals are not a random sample of the web. The deals reported so far have gone to national newspapers, wire services, and major media groups with large content archives, constant news output, and, in most cases, strong existing search and AI visibility before any contract was signed. Some, possibly a large share, of the 48% premium could simply reflect that OpenAI is licensing outlets that were already going to be cited more often, rather than the licence itself changing anything about how ChatGPT retrieves and ranks their content.
Read a citation-premium statistic like this one the way you would read any AEO correlation claim: check whether the study demonstrates a mechanism you could actually act on, or just a difference between two groups that were already unlike each other before the comparison started. The same caveat applies to research showing backlinks correlating with citation share without proving which one causes the other.
The mechanism that most plausibly explains it
Most publicly reported OpenAI publisher deals do not amount to a simple legal permission slip. They typically bundle training and display rights with something more concrete: a structured, direct content feed into OpenAI's retrieval systems, alongside faster crawl or ingestion priority, and in several announced cases an integration that sits outside OpenAI's ordinary web-discovery path entirely. If that feed-level access, rather than the payment itself, is what actually drives the premium, it would explain why the lift concentrates so heavily on ChatGPT instead of spreading evenly across engines: a direct data feed only ever feeds the one platform it was built to serve.
Whether or not this specific causal story is correct, it points at the right question to ask about any AI-citation intervention: does it change how fast and how reliably a model can find and trust your content, or does it just change a legal agreement that sits alongside content the model could already see? Interventions of the first kind are the ones worth prioritising with limited time.
What this means if you will never sign one
For nearly every site reading this, an OpenAI-scale licensing deal will never be on the table. The deals reported so far have gone to national newspapers and major media groups, not SaaS companies, agencies, local businesses, or independent publishers, and there is no indication that is about to change. What is worth taking from the study is not the deal itself but the plausible mechanism behind it, direct, structured, fast-updating content access, because that part is something any site can build without a contract or a payment to OpenAI.
- Publish structured, machine-readable content (Article schema, a genuinely current RSS or JSON feed) rather than relying on a general crawl to notice a change on its own
- Get your update cadence fast: correct caching headers so a real content change reaches a re-crawl in hours, not the weeks a stale Cache-Control or Last-Modified header can quietly cause
- Maintain a clean identity layer, Organization schema, a brand.json file, and a consistent About page, so a retrieval system has an unambiguous entity to attach a citation to
- Confirm GPTBot and OAI-SearchBot both have unrestricted access in robots.txt; a licensing deal does not exempt a publisher from the ordinary crawler-access checks, and the absence of one does not excuse skipping them either
The concentration risk licensed publishers are taking on
The 57.9% ChatGPT concentration figure is worth sitting with even if you will never see a licensing offer of your own. Citation density differs sharply by platform already, and a publisher whose citation mix tilts that hard toward a single engine is exposed if that engine's algorithm, market share, or licensing terms move against it. The unlicensed publishers in the same dataset kept a more even split across ChatGPT and Perplexity, which is the safer default posture for any site that is not being paid specifically to concentrate its visibility on one platform.
If you are actually offered a deal
A small number of publishers reading this may genuinely be approached by OpenAI or another lab. The one question worth asking before signing is whether the agreement includes a structured feed integration into the platform's retrieval systems, or only a training-and-display licence with no feed component. Only the former plausibly explains the citation premium in the study; a training-only agreement, however well it pays, is unlikely to move citation numbers on its own, since it changes what OpenAI is legally allowed to do with your content without changing how quickly or reliably its systems can find it.
The realistic takeaway is not that a licensing deal is the secret to AI citation. It is that whatever a licensing deal buys, fast, structured, direct access to your content, is buyable in miniature by any site willing to do the technical work, without a contract or a cheque from OpenAI at all.