Press Ranger and OtterlyAI report that pages from publishers with OpenAI licensing agreements received 10.2 ChatGPT citations per cited page in June 2026, compared with 6.9 for unlicensed publishers: a 48% premium. That is a material platform-specific difference for anyone selecting earned-media targets for ChatGPT visibility. It is not, however, evidence that signing a licensing agreement caused the full difference. markets.businessinsider.com
The study establishes a deal-correlated citation skew worth taking seriously. It does not settle whether that skew comes from licensing access, preferential treatment, structured delivery mechanisms, the pre-existing authority of licensed publishers, or a combination of all four. For GEO practitioners, that distinction changes the action: do not treat an OpenAI deal as a universal quality signal, but do treat coverage on a licensed publisher’s page as a potentially higher-value route into ChatGPT’s cited-source set.
What changed: a large citation dataset was matched to licensing records
The underlying analysis matched 129.3 million citations across more than 20 million cited URLs against 91 confirmed AI-publisher licensing agreements, mapped to 314 publisher domains. Citations were captured during June 2026 from ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Gemini, and Claude; the deal record was backed by public sources through July 28, 2026. globenewswire.com
That is more useful than a vague claim that licensed publishers “show up more.” The reported measure is citations per cited page, rather than total citations. This normalization matters because it reduces the simple scale explanation: a large publisher cannot explain the premium solely by having more pages in the observed cited pool. elmohq.com
But per-page normalization is not causal identification. It controls for page volume, not the reasons particular publishers became attractive licensing partners in the first place. The source analysis explicitly notes that OpenAI did not license random publishers and that per-page normalization does not control for the underlying citability that may have made those publishers worth licensing. elmohq.com
That is the core limitation. A publisher can have high editorial authority, strong brand recognition, deep archives, commercially useful evergreen pages, and established inclusion in AI citations before a deal is announced. If those characteristics increase both the likelihood of an OpenAI agreement and the likelihood of being cited, the observed 48% contains selection effect alongside any licensing effect.
The platform asymmetry is the strongest part of the finding
The 48% figure becomes more interesting because the result is not reported as a generic licensing premium across every engine. OpenAI-licensed publishers averaged 10.7 citations per cited page across all seven measured platforms, versus 7.3 for unlicensed publishers, a 46% premium. Yet the study says the effect concentrates on ChatGPT: publishers with OpenAI deals received 57.9% of their total AI citation volume from ChatGPT. markets.businessinsider.com
More importantly, OpenAI was reportedly the only licensor with a clear advantage on its own platform. Google-licensed publishers were cited on Google AI Overviews at a slightly lower rate than comparable unlicensed publishers, while Perplexity-licensed publishers reached parity on Perplexity. globenewswire.com
This asymmetry weakens the simplest explanation that licensing merely identifies inherently superior publishers. If the deal cohort’s authority alone drove the result, a broad advantage across systems would be easier to expect. Instead, the reported advantage is strongest where the OpenAI relationship should matter most.
It still does not prove intentional citation preference. The study does not document the matching method for “comparable unlicensed publishers,” and it does not provide a controlled before-and-after analysis around signing dates. The available material also does not isolate whether contracts change crawl permissions, source feeds, content accessibility, or citation selection. The best-supported conclusion is narrower: OpenAI agreements and elevated ChatGPT citations occur together, at a large measured margin, in this June snapshot. elmohq.com
The 112% exclusive-deal result raises the stakes—and the questions
Publishers that had signed with OpenAI and no other AI company reportedly earned 112% more ChatGPT citations per page than unlicensed publishers. The release presents this as evidence that the relationship is especially concentrated on the platform tied to the deal. markets.businessinsider.com
For causal inference, though, exclusivity is not a clean experiment. An exclusive OpenAI cohort may differ from both multi-licensed and unlicensed publishers in ways the supplied material does not disclose. We do not know its size, publisher mix, topical mix, historical citation rate, dates of agreements, or whether the cited pages were created before or after agreements. Those omissions do not negate the result; they limit the precision of what can be inferred from it.
The study period is also one month: June 2026. That makes the analysis a valuable observation, not a durable estimate of a stable premium. The third source characterizes the design as correlational and says no methodology paper accompanies the release. elmohq.com
Authority remains a credible competing explanation
The observed concentration among licensed publishers is a reason to resist reading “licensing” as the whole story. Five media groups—Future plc, Forbes, People Inc., Condé Nast, and Hearst—accounted for 69% of citations to licensed publishers in the study. globenewswire.com
That degree of concentration is compatible with a world in which a limited number of large, already-citable publishers drive much of the cohort result. It may also be compatible with a world in which licensing gives those already-strong publishers better access or distribution into ChatGPT. The supplied evidence does not quantify either contribution, so attributing the full 48% to the contract would overstate what was measured.
Content composition is another confounder the public release flags without fully resolving. Best-of lists, buying guides, and product reviews made up 46.9% of licensed publishers’ citations. markets.businessinsider.com
Those formats may be especially useful for answer generation, but the release does not provide the equivalent format distribution for the comparison group. Until it does, practitioners should avoid assuming that outlet licensing status outperforms format, topical relevance, and the page’s independent citability.
What this means for GEO media strategy
The practical implication is not “buy access to licensed publishers.” Most brands are not parties to publisher licensing contracts, and the study itself says a licensing deal is not a prerequisite for AI visibility. It identifies unlicensed NerdWallet, Healthline, and Bankrate among the most-cited news domains. globenewswire.com
It also finds that niche and trade outlets, mostly unlicensed, held the majority of news citations in 15 of 16 U.S. industries and collected 213% more AI citations than mainstream media. markets.businessinsider.com
That creates a two-part GEO targeting model:
- For ChatGPT-weighted visibility: include relevant OpenAI-licensed publishers in the media map, but assess their actual coverage fit, page formats, and existing citation presence rather than assuming the deal alone creates value.
- For cross-engine visibility: keep vertical, trade, and niche publications central. The reported Google and Perplexity results provide no basis for paying a licensing-status premium on those platforms. elmohq.com
- For measurement: distinguish a cited publisher page, a brand mention on that page, and the accuracy of the resulting answer. Citation frequency alone does not establish that an answer represents either the source or the brand correctly. elmohq.com
News itself represented only 7.2% of all citations in the study. globenewswire.com
So the 48% result is important, but its addressable surface is narrower than the headline implies. It changes the expected value of some publisher placements for ChatGPT. It does not demonstrate that licensing is the decisive ranking factor, replace source-quality work, or generalize across answer engines. The credible reading is deliberately modest: licensing is now a measurable variable in ChatGPT-oriented earned-media planning, while causality remains unproven.