The AI citation strategy for everyone who will never sign a licensing deal

There are two ways to read a study that proves licensing deals buy citations. The first is fatalistic: the engines are pay-to-play now, the deals are signed, and the rest of us are competing for scraps. The second is that a 91-deal universe is a very small club, and a study large enough to measure the club precisely is also large enough to show exactly where everyone outside it is winning. The second reading is better supported by the actual numbers.

Press Ranger and OtterlyAI released the study on August 20, 2026, covering 129.3 million citations across more than 20 million cited URLs on seven platforms: ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Gemini and Claude. The citation data comes from June 2026, with licensing deals verified through July 28. It is the largest attempt so far to answer a question every marketing team has been asking informally since the first licensing announcements: does the deal actually move citations, and by how much.

The two-tier citation economy the data just made visible

The headline number is clean. OpenAI-licensed publishers earn 10.2 citations per page on ChatGPT. Unlicensed publishers earn 6.9. Across all seven platforms the figures are 10.7 and 7.3, a 46% premium rather than 48%, which tells you the advantage is real but only mildly contagious across engines. Publishers holding OpenAI-only deals, with no other platform agreements, post 112% more ChatGPT citations than unlicensed competitors.

The concentration numbers matter as much as the premium. Five media groups, Future plc, Forbes, People Inc., Conde Nast and Hearst, capture 69% of all licensed publisher citations. So the club is 91 agreements across 314 domains, and inside the club, two thirds of the benefit lands on five owners. Anyone modeling a licensing deal as a realistic path is modeling an outcome available to a few dozen organizations on earth, most of which already had it.

There is a second cost attached to the deal that rarely gets discussed. Licensed publishers now draw 57.9% of their total AI citation volume from ChatGPT alone. Thomas Peham, OtterlyAI's CEO, put it directly in the release: a licensing deal does one clear thing, it tilts your citations toward ChatGPT. That is a fine trade if ChatGPT stays where it is. It is a concentration risk of exactly the kind we flagged when Reddit's ChatGPT citation share collapsed inside a single month after a sourcing change nobody outside OpenAI saw coming.

That last bar is the one that reframes the whole study. News accounts for 7.2% of all AI citations. The licensing fight, which has consumed most of the public argument about AI and publishing for two years, is a fight over a small minority of the citation surface. The other 93% is documentation, guides, comparisons, reviews, forums, trade coverage and product pages, none of which is governed by a licensing agreement of any kind.

Why an AI citation strategy cannot start with a licensing deal

For a brand rather than a publisher, the deal question is not merely difficult, it is categorically unavailable. Nobody is signing a content licensing agreement with a cybersecurity vendor's blog. So the useful question is not how to get into the club, it is which of the club's advantages are actually transferable, and the study is unusually helpful on that point because it breaks the advantage into parts.

Part one is the deal itself, which is not transferable. Part two is format. Service journalism, meaning best-of lists, buyer guides and reviews, produces 46.9% of licensed publisher citations. That is a format advantage, not a contract advantage, and it is completely available to anyone who can either publish those formats or get included in someone else's. Part three is the outlet tier, and this is where the study produces its genuinely surprising finding: trade and niche outlets earn 213% more AI citations than mainstream media across 15 of the 16 US industries examined.

“Steve Beyatte, who founded Press Ranger, framed the finding this way: 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.”

Sit with the size of that number for a moment against the size of the licensing premium. A deal buys 48%. Getting covered in the trade press rather than the national press is associated with 213%. If those two numbers were both on a media plan, nobody would spend a second on the first one. Yet most enterprise communications budgets are still weighted toward tier-one placements, for reasons that made complete sense when a Wall Street Journal mention was the ceiling of credibility and nothing was reading the trade press at scale except the trade.

Something is now reading the trade press at scale. That is the change, and it argues for a different allocation than the one most brands are running, in the same direction we argued when we looked at how earned media converts into AI citations rather than into referral traffic.

The five plays

Each play below states the move, the specific steps, and the condition that tells you it is finished. They are ordered by how quickly they produce measurable citation movement, not by how hard they are.

What the trade outlet advantage actually costs to capture

The 213% figure will get quoted without its cost structure attached, so it is worth stating the cost structure. Trade placements are cheaper per unit than tier-one placements and more numerous, but they are not free, and the work is different in kind. A national reporter wants a story with stakes. A trade reporter wants specificity: the numbers, the deployment detail, the thing that a practitioner in that industry will recognize as true. Communications teams built entirely around narrative pitching tend to be poor at the second one and do not always know it.

There is also a coverage-quality trap. Trade outlets vary enormously in whether the engines actually read them. A trade site running syndicated vendor press releases with no editorial layer will not carry citation weight, whatever its domain metrics say, and there are a lot of those. Play one includes the citation-presence check for exactly this reason, and skipping that check is the fastest way to spend a quarter earning placements that no engine ever reads. Our link building and digital PR work puts that check before the pitch list, not after it.

The final cost is patience of a specific kind. Citation presence builds through repeated appearance across independent sources, which is why single placements rarely move anything and why the fifth placement in a category tends to move more than the first four combined. That is a real effect and it is also a convenient excuse for a program that is not working, so play five exists to tell the difference.

Measuring an AI citation strategy with no deal underneath it

A brand without a licensing deal has one measurement advantage over a publisher with one: nothing in your citation profile is contractual, so every change you observe is a change you caused or the engine caused, and those two are separable with enough history. Publishers with deals cannot cleanly separate the deal's effect from their editorial work. You can.

Use the study's own units so your numbers are comparable to something outside your account. Citations per page. Engine mix as a percentage. Asset type as a segment. Then hold two external reference points: the 6.9 unlicensed ChatGPT figure and the 7.3 all-platform figure. If your owned pages are running well below those, the problem is extraction and structure rather than authority. If they are running near or above them, the constraint is coverage volume, and the budget belongs in plays one and two.

This is also where cross-engine comparison earns its keep, because the engines disagree with each other more than most dashboards admit, and we walked through how much in our engine overlap analysis. A single blended visibility score will hide exactly the concentration this playbook is built to prevent. Report per engine, then blend if an executive audience needs one number, and never the other way around. If dashboards are the constraint, that is a reporting and analytics problem with a known fix, not a research problem.

One caution on timing. The measurement environment is moving underneath everyone right now. In late August, Google began automatically expanding AI Overviews into full AI Mode responses for some queries, with no click required, which Google described as dynamic expansion for topics where its systems determine it is most useful. That change alters where citations appear and how many links a user sees before the traditional results, and it will move your numbers without any change in your work. Annotate the date in your reporting and keep the annotation for a year, the same discipline that keeps an enforcement event from being misread as a content problem.

The first thirty days

Week one is inventory and instrumentation. Tag last year's coverage mainstream versus trade, build the citation-presence check for 20 to 30 trade targets, and stand up citations-per-page tracking segmented by asset type. Nothing here requires new content and all of it is required before any of the rest produces a readable result.

The reason to run this in that order is that the two failure modes look identical at day thirty if you have not instrumented first. A brand with strong structure and no coverage and a brand with heavy coverage and unextractable pages both show a flat citation count. They need opposite interventions, and the only thing that tells them apart is the segmented per-page number the instrumentation produces. Teams that skip week one usually spend the quarter on whichever intervention their existing staffing prefers, which is not a strategy, it is an org chart.

None of this closes the licensing gap, and it is not meant to. A 48% per-page premium held by 314 domains is a real structural advantage and it will stay real. But it applies to 7.2% of the citation surface, it concentrates two thirds of its benefit in five media groups, and it comes bundled with a 57.9% dependency on a single engine's sourcing decisions. The 213% trade advantage sits in the other 93%, it is available on merit, and it is not owned by anyone. Start there, in the categories where your buyers actually read, and be the specific source the roundup writer cites because you made the comparison easy. When that is working, technical categories with fast-moving buying committees reward it faster than almost anywhere else.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.