The Research Platform Has a New Job Description
If researchers are no longer coming to publisher platforms to find and consume content, what exactly is the platform for? That question kept surfacing at NISO Plus in the conversations that came back from our team, dressed in different language depending on the session: usage metrics, headless infrastructure, AI keynotes, metadata workflows. But underneath all of it, the tension was the same. The traditional model was elegant in its simplicity. A researcher needed an article. They found it through a library discovery system or a database, authenticated against their institutional holdings, landed on the platform, read it, maybe downloaded it. The library was always an intermediary in that chain, but the chain itself was visible: usage data told a coherent story about value delivered, and the whole system was legible because the researcher's behavior was legible.
That model is under real pressure now, driven less by any loss of content value than by a fundamental change in how researchers get to it.
The Crocodile in the Room
The scholarly publishing industry has a name for what's happening to search traffic: the Crocodile Effect, a term borrowed from SaaS marketing to describe the widening gap between how many times content is surfaced in search and how many times anyone actually clicks through to read it. Kudos launched a cross-industry study called “Taming the Crocodile” to examine it in the scholarly context, and the list of sponsors, which includes Silverchair, Elsevier, BMJ Group, American Physical Society, Oxford University Press, and several others, suggests the industry recognizes this as something worth understanding carefully.
Zero-click search is already a reality for general web content, and AI-generated overview panels are accelerating it. A researcher asks a question, gets a synthesized answer, and moves on. The underlying articles that informed that answer may never receive a pageview or a referral. By every traditional measure, the content was never used. But it was. The value was delivered. It just wasn't measured, attributed, or compensated.
One participant noted that libraries actively promoting agentic tool use among their patrons are already seeing steeper drops in traditional usage metrics than libraries that aren’t. It’s a current data pattern with real budget implications for librarians trying to justify their subscriptions, and it points to something the industry needs to take seriously: what we “track” and what “happened” have started to diverge.
What Our Communities Are Telling Us with Their Behavior
There's a temptation to frame this as a crisis of access control, and some of that impulse is reasonable. Our researchers and our communities are telling us something through their behavior. They are bypassing platforms because they have found more efficient ways to get to what they need, and that efficiency is clearly valuable to them. The question worth sitting with is whether fighting that efficiency serves the mission, or whether there's a more honest and more durable path forward.
However, that newfound efficiency comes with a risk the industry hasn’t fully named yet. When researchers bypass authenticated platforms in favor of general-purpose AI tools, they are often drawing on content that has been absorbed into a model’s training rather than retrieved from a live, maintained source. The ideas may survive that process, but the argument structure, the version history, the retraction notice, the citation chain, largely don't. There's no mechanism to retract a finding that has been baked into a model’s training.
Scholarly publishing has always understood itself as being in service of something larger than any individual platform or business model. If the tools researchers are adopting are getting them closer to that content faster and more naturally, then the question is how publishing infrastructure can serve as the backbone of that experience rather than a barrier to it.
The Platform as Gatekeeper, Not Destination
This is where the role of the platform must evolve, and where the scholarly publishing industry has something genuinely important to contribute that the broader AI ecosystem does not yet have.
We have spent decades building the infrastructure of trust: peer review workflows, versioning, retraction management, persistent identifiers, access, and licensing frameworks. These are the systems that distinguish verified research from fabricated claims, final publications from early drafts, and current guidelines from outdated ones.
AI systems cannot see any of that unless we make it visible to them. And right now, most of them can't. A hallucinated citation looks structurally identical to a real one if the only difference is provenance, and provenance is exactly what current discovery infrastructure doesn't reliably communicate to machines.
The platform’s job is no longer to be just the place where a researcher lands. Its job is to govern the terms under which content participates in discovery, ensuring that whatever flows through those channels is verified, appropriately licensed, and accurately attributed regardless of where the researcher ultimately ends up.
That job gets harder as the unit of content itself starts to fragment. A journal article has always been a convenient container, but what an agentic workflow actually needs might be a single finding, a figure, or a methods section. Those fragments carry even less inherent context than a full paper does, which makes the trust infrastructure more critical, not less. The platform becomes the metadata layer that ensures that whatever travels, travels with enough information to be evaluated.
The Commercial Model Has to Travel with the Content
If the platform is no longer primarily a destination, the obvious question is how anyone makes money. It’s worth sitting with that directly rather than dancing around it, because the answer has real implications for how platforms need to be built, how publishers need to think about their commercial futures, and how libraries need to think about budget allocation in a world where traditional usage signals are becoming less reliable.
The pageview was never really the thing being sold. It was a proxy for access, a measurable signal that someone had gotten to the content through a licensed channel. What’s breaking down is the proxy, not the underlying meaning. A researcher whose AI assistant synthesizes findings from 10 licensed articles has consumed something worth paying for. The challenge is that the current commercial infrastructure has no way to recognize or capture that transaction.
A recent piece in The Scholarly Kitchen, drawing on conversations with publishing leaders across the industry, named the structural problem clearly: the subscription model and the AI licensing model are converging, and in some cases cannibalizing each other, with publishers who have bundled AI rights into subscription pricing potentially undercutting their own premium licensing deals. The authors called it a paradigm problem, not a pricing one, and most publishers, by their own admission, haven’t reckoned with what it means for either revenue line.
The path forward is going to require a genuinely open commercial mindset from publishers, including those with nonprofit or society missions who may be less accustomed to thinking in these terms. The reading experience and the commercial transaction can be decoupled, and for some organizations accepting that will feel like a significant cultural shift, not just a technical challenge but a financially destabilizing one. Smaller publishers and societies in particular face real uncertainty about what their revenue looks like on the other side of this shift, and the people whose jobs depend on current models have legitimate reason to find this moment uncomfortable. The platform’s role is to make those new pathways real. Governance of the channels is only meaningful if the channels carry commercial terms that can be enforced and tracked, which means building the infrastructure for licensing and usage instrumentation that reports on value delivered regardless of where the researcher ultimately ends up.
What Scholarly Publishing Systems Are Built For
Scholarly publishing platforms are uniquely well positioned for what the emerging discovery landscape actually needs.
Take identity. Author disambiguation, institutional affiliation, ORCID integration, the kind of persistent identifier infrastructure that makes attribution trackable across systems and across time: when an AI system surfaces a finding, the ability to trace it back to a specific author and a specific version of a specific article is not a nice-to-have. It is the difference between knowledge and noise. Scholarly content also doesn’t stand alone: it exists in relation to prior findings, corrections, retractions, commentary, and responses. That relational context is encoded in citation structures and editorial relationships in ways that AI systems pulling from the open web simply cannot replicate.
Then there’s access governance. The licensing and authentication infrastructure that publishers and platforms have built over decades, imperfect as it is, represents a serious capability that AI systems need and largely don't have. Knowing who has access to what, under what terms, with what audit trail, is genuinely hard, and it is infrastructure the industry has spent years building.
Underpinning all of it is curation. Controlled vocabularies like MeSH give biomedical content a precision of retrieval that keyword search cannot replicate. Retraction and correction workflows ensure a flagged paper is identified as such across the catalog, but only if the publisher maintains that infrastructure upstream. Preprint-to-published linking connects a submitted manuscript to its peer-reviewed version to any subsequent corrections in ways that are simply lost when an AI system pulls from an open repository without that editorial chain behind it.
The question is whether the industry treats these capabilities as the foundation for what comes next, or lets them atrophy while waiting for the discovery landscape to stabilize.
Building the Plane While Flying It
Nobody knows whether Model Context Protocol (MCP) becomes the long-term standard for AI-to-content integration, or whether something else emerges, and we are unlikely to know for some time. But the foundational work is the same regardless of which protocol wins: clean structured metadata, authentication that extends into agentic workflows, and usage instrumentation capable of capturing value even when that value doesn't look like a pageview. These investments compound, and they make content more discoverable through any channel, human or machine, in whatever configuration comes next.
The preparation for a “headless future” is not a bet on any specific technology. The delivery mechanism has changed many times over the history of this industry, from print to CD-ROM (remember those?!) to the web to mobile, and it will change again. What has not changed is that the value was never really in the container. It was in the peer review, the editorial judgment, the version control, the retraction notice, the persistent identifier that makes a finding traceable a decade after publication. AI-mediated discovery does not diminish that value. If anything, it makes it more legible, because the difference between a trustworthy source and an untrustworthy one becomes consequential in ways that a casual web browse could obscure.
The Responsibility That Comes with the Infrastructure
The scholarly publishing community built the current infrastructure of knowledge for a reason, and that reason was not pageviews. It was the integrity of the scientific record and the responsible dissemination of knowledge to the people and communities who need it. The tools researchers are now adopting to discover, surface, summarize, and access content are being rapidly integrated into genuinely consequential decisions in clinical practice, policy, and public understanding of science.
If those tools are drawing on content that lacks proper attribution or conflates preliminary results with established consensus, the downstream harm is real. The scholarly publishing industry is not the only stakeholder responsible for preventing that harm, but we are among the most capable of addressing it. If we treat educating and structuring the tools that are reshaping discovery as part of our mission, that changes the conversation about where platform investment is warranted and what it means to be a responsible participant in the discovery ecosystem right now. Scholarly publishing platforms were not built for the pageview era. They were built for this one. The job description has changed, and the industry that established the standards for trusted knowledge is exactly the one positioned to fulfill it.
