Gateway(s)

Gateway(s)

This past month, Wiley announced the release of two new AI services: AI Gateway and AI Knowledge Nexus. At first glance, these seem to be similar and were announced together (and grouped together on the Wiley website), but they are actually very different things.

AI Gateway is an aggregation and distribution platform for large language models (LLMs). It is a content repository that can be accessed by a “connector” — an API (application programming interface) that uses the MCP (model context protocol). Confusingly, the connector is not called “Wiley AI Gateway” but instead “Scholar Gateway.” 

The Wiley website says that Scholar Gateway has “native support for Claude, Mistral AI, Perplexity, and more.” However, Scholar Gateway notes that the service only works with Anthropic’s Claude Pro and Mistral’s Le Chat at the moment.

Users of Claude Pro or Le Chat who add the connector and have an institutional subscription (although an institutional subscription to what is unclear) can authenticate via CONNECT (way too many brands…). Once authenticated, they can perform queries that bring in relevant information through Scholar Gateway. Meaning, a user can perform a query on, say, volcanos and it would bring in relevant information from scholarly articles from Wiley journals. In other words, it is a query-based platform envisioned largely for academic users.

Importantly, AI Gateway only returns snippets of information, with links back to full-text content on Wiley’s platform (requiring appropriate institutional access if the article is not OA). In this sense, AI Gateway is very much a sustaining technology, in that it supports and potentially extends existing institutional subscriptions.

While the AI Gateway is currently only supporting Wiley titles, it is envisioned as a wider distribution platform, including for other publishers who want to make their content available and have it similarly distributed to LLMs. It is unclear (to us at The Brief) exactly how that would work financially.

There is a lot to like about Wiley AI Gateway. It is a handy way to connect to a corpus of content from inside the LLM of your choice. It is subscription-based and maintains, and perhaps even increases, the value of scholarly content, as well as the value of Wiley’s institutional sales apparatus. It also connects LLM queries to full-text content with not just “attribution” but full-text links to sources.

It is also a hot mess from a branding and user experience perspective. The user flow is so complicated that it requires over a page of documentation. Wiley has announced its “AI Gateway,” but it is called “Scholar Gateway” in the user’s LLM of choice. Wiley is trying to appeal to other publishers, so they may be deliberately downplaying the Wiley brand. However, why not just call it “Scholar Gateway” everywhere? (They could add “Powered by Wiley” or something in that vein). Then to perform a query, we are told that we need an institutional subscription — but to what? To Scholar Gateway? To something else? To find out if we have an institutional subscription, we need to authenticate with CONNECT. What is CONNECT? (We at The Brief know what CONNECT is, but what about users less deep in publisher technology?) Is it from the same company as Scholar Gateway? AI Gateway? Atypon? Wiley? Is our subscription to CONNECT? It is a tangle of brands. And this is all before Wiley introduces content from other publishers into the mix.  

Focusing on the research use case (and putting the branding issues to the side), there remains another, familiar, problem. Why would a user want to query just a relatively small slice of the scholarly literature? Wiley is trying to expand this slice by inviting other publishers to the platform, but unless other publishers such as Elsevier and Springer Nature sign up, which seems unlikely, it will remain a subset of the wider research corpus. Therefore, it will have limited utility for most scholars, due, not to the technology, but the limitations of the content set. This may not matter in the long run, however, if LLMs query multiple connectors from different publishers simultaneously and bring together the results for users.

Wiley’s AI Knowledge Nexus (here abbreviated to “Nexus”) is very different. It is less a technology and more a sales and licensing service. Wiley is essentially offering its experience in negotiating AI licensing deals and its market research to other publishers (for a cut of the deal). The Nexus service includes one-time training deals, as well as other recurring and query-based models. It appears to be focused on the corporate market, operating on the hypothesis (a good one) that corporate R&D efforts will be increasingly incorporating AI technologies. Those technologies will be ravenous for curated and highly relevant research content.

Nexus is a straightforward business-to-business licensing service, and therefore does not suffer from either the same branding issues or the same content limitations as Gateway. Corporate R&D departments and technology developers know that they are going to have to license content from multiple publishers, so the fact that Nexus has a limited content set is less of a concern. Nexus also builds on Wiley’s existing expertise in institutional sales and licensing — an area where there are real advantages to scale. Bringing smaller publishers into the service helps all parties. It helps smaller publishers gain access to markets that would be challenging and expensive for them to reach on their own. It helps Wiley by increasing Wiley’s own scale economics — they book their costs against a larger licensing business. And it helps corporate customers by reducing the number of brokers with which they need to deal.

Wiley is staking out a leadership position as both an aggregator and a broker for AI licensing deals, with their Gateway product focused on institutional researchers and their Nexus product aimed at licensing to the corporate sector. They are also perhaps the first publisher to articulate a strategy for how they (and other publishers) might play a role in a landscape intermediated by LLMs and other AI technologies. Notably, these are both recurring revenue initiatives and move the conversation beyond one-time training deals to more sustainable models.

Backflip

As the scholarly publishing world has sought out better, more equitable and sustainable business models for open access (OA), the Subscribe-to-Open (S2O)model has garnered a significant amount of attention. It has also garnered some skepticism regarding its long-term sustainability.

From the point of view of a publisher contemplating S2O, an essential consideration is therefore what happens if skeptics prove correct? While flipping to OA is, of course, the entire point of S2O models, the model only works if the journal can present a credible path back to a standard subscription model. This credible path is important both in terms of market messaging (“If we don’t meet our threshold, we will flip back to a standard subscription model”) but also for a publisher’s own planning and approval processes (“If it doesn’t work out, we can transition back”).

While we have seen some journals that have pledged to move to S2O fail to meet their subscription targets, we have yet to see any S2O program flip a journal back to the subscription model.

At least that was true until AIP Publishing announced an end to its recent S2O pilot, moving the two journals involved back to the hybrid model. The publisher states that while the two journals did hit their subscription revenue thresholds, they also saw some attrition, with “a few institutions who canceled their subscriptions.” Submissions to the journals saw initial growth following the announcement of the pilot, but they fell in the next year to levels below those seen for the rest of the portfolio. Usage rose, but not significantly more than was seen across AIP Publishing’s hybrid portfolio.

The S2O model was not seen as offering “a path for scaling or monetizing the journals’ steady expansion,” nor an ability to cover the ongoing increases in operational costs. AIP Publishing’s goal in ending the S2O pilot is to streamline things and reduce complexity around their OA offerings, moving the two journals into their portfolio-wide Author Select, AIP Fusion, and Read and Publish programs.

While on one hand, this news is a mark against the sustainability of the S2O model, on the other hand it serves as a concrete proof point that experimenting with S2O is not necessarily a one-way street.

That said, the AIP Publishing pilot was a relatively short foray into S2O. A credible path back from S2O becomes more tenuous over time. Under a standard subscription model, a journal’s content is accessible only via subscription (journals adopting Bronze OA present a different case). Once a journal flips to S2O, each subscription year of content that is made open access remains open in perpetuity. If the journal flips back to a subscription model after, say, five years of S2O, the previous five years of content will remain open access, which may reduce the value of the subscription in the eyes of some customers. This makes it harder to flip back (or backflip!).

Giants

When it comes to journals, how big is too big? Is there a size limit where it becomes impossible to effectively maintain quality and integrity controls? Or are journals like sauropods, where paleontologists keep uncovering the bones of ever larger creatures from the Jurassic layers of the earth? (We imagine future information archeologists announcing ever larger journal discoveries from long forgotten layers of the Internet.) Recent developments from Elsevier and Springer Nature suggest there may be some ecological constraints.

Over 10 years, Springer Nature’s Cureus Journal of Medical Science went from publishing around 500 articles per year to more than a staggering 25,000 in 2024. Last year, Clarivate put Cureus “on hold” for unspecified irregularities. This past month, Clarivate de-listed the title from the Web of Science and its Impact Factor has been rescinded. De-listing can prove a near extinction event for a journal, as was the case for MDPI’s International Journal of Environmental Research and Public Health, once the second largest journal in the market and now reduced to less than 1/10th the size at its peak.

James Butcher points out that the third largest journal in the market last year, Elsevier’s Heliyon, which is also “on hold” with Clarivate, has already seen a precipitous decline in publication output. Heliyon is on track to publish 85% fewer articles in 2025 than in 2024. Butcher suspects that this is the result of Elsevier “trying to put its house in order” to avoid being de-listed. Similar quality control considerations may be behind recent reductions in the article outputs of the portfolios of both MDPI and Frontiers.

All of this raises questions about the viability of extremely large journals. We’ve now seen multiple journals rocket to the top of the charts (as measured by article output) only to shrink back down under the weight of research integrity issues. A particular challenge is that very large journals tend to operate under an OA article processing charge or transformative agreement model, with relatively modest price points. They also are, by definition, not highly selective journals. A highly selective journal may accept fewer than 20% of the papers it receives. A highly selective journal also tends to have much higher revenue per article as compared with a large, less selective title. This means that the highly selective journal can spend more, on a per-article basis, on research integrity checks and the overall editorial process — simply put, editors can spend more time with papers. Very large journals have a huge set of papers to review and less money to work with per paper.

Despite these pressures, giants still roam the research savannahs. Scientific Reports, the largest journal yet to exist, is on pace to publish more than 44,000 articles this year.

Manifestos

As John Warren notes, there is nothing like a good publishing manifesto. Like a podcast, everyone needs one. The last couple of decades alone have brought us Budapest, Bethesda, Berlin, and DORA (to name some of the more notable entries in the genre). Cambridge University Press & Assessment (CUPA) launched a podcast last year, which perhaps only served to highlight the organization’s lack of a manifesto. The august institution remedied this gap this past month with the release of “Publishing Futures,” which calls for “radical change” in academic publishing (to be fair, they call this missive a “report,” but the British are a modest people and can undersell even when they have a manifesto at hand). The Publishing Futures manifesto (we are going to take the liberty of calling it one) argues that scholarly publishing is being flooded with vast amounts of low-quality materials, access inequities (whether for readers or authors) persist, and no one seems to want to pay the rising costs for all this growth (these points seem indisputable).

As with most critiques of scholarly publishing, the real culprits at the heart of the arguments are the career reward and funding mechanisms of academia. Publishers, essentially a service industry that exists to meet the needs of the research community, are unlikely to be able to dictate to that community how it should run things. CUPA Managing Director Mandy Hill is clear that this would be the tail wagging the dog, and that these are not things publishers can either dictate or achieve on their own. Her hopeful outcome from the report is that it will raise awareness and bring together stakeholders “with the idea of convening conversations to drive collective action.”

Not to be outdone, the Royal Society’s journal Royal Society Open Science has published the Stockholm Declaration — an unabashed manifesto named after a capital city and everything.

To be clear, the Stockholm Declaration is not authored by the Royal Society, but “a group of scientists and stakeholders” who gathered at a June 2025 meeting at the Royal Swedish Academy of Sciences. Compared to the CUPA manifesto, the Stockholm Declaration is much more radical (you can get the gist from this entry in the Royal Society’s blog). The Stockholm signatories effectively want to toss out the existing publishing system and replace it with “non-commercial publishing models where journals are community-owned and authors retain their copyright.” Speaking of copyright, they appear to want to issue CC BY licenses for all existing (already published) works, calling for “free, open access and public rights to use digital archives and their AI-based analysis, data or texts, including material which is already in the permanent scientific record” (emphasis ours). They also want to create some kind of publishing police, calling for “developing and enforcing laws and regulations by independent bodies for quality control, and draft legal frameworks to sanction authors, paper mills, journals, or publishers.” To their credit, they (echoing the CUPA manifesto) also call for reform of academic reward systems, proposing “incentive systems to merit quality, not quantity.”

This is truly radical stuff — a proper manifesto! But where is the podcast?

Karger

In late breaking news as we wrapped up this issue, Oxford University Press (OUP) has purchased Karger Publishers. Although the terms of the deal were not released, Karger brings around 100 (owned, not society-partnered) medical and healthcare journals to OUP, and a backlist of 9,000 books. In a market where publication output is increasingly seen as key to success, many midsized publishers are taking this pathway and acquiring smaller organizations (see Sage’s recent purchase of Mary Ann Liebert, Inc., as another example). While the Karger purchase significantly increases the number of journal titles at OUP, the journals acquired from Karger are relatively small, publishing just over 5,300 articles in 2024. Adding those articles into OUP’s total doesn’t move the needle much in terms of overall size (as measured by article output): In 2024 it would have moved them from being the 11th largest publisher in the market to 10th (just sneaking ahead of Frontiers).

But article output is not everything. The Karger portfolio comes with subscription revenues that can be added to OUP’s packages. The biomedical and clinical focus of the portfolio also likely played a role in the acquisition calculus. Karger’s titles help bolster OUP’s footprint in medicine, which brings with it revenues in advertising, reprints, and licensing (including AI licensing), and helps with offerings aimed at the healthcare market.  

Briefly Noted

For those tracking the ongoing decline of the American scientific enterprise, a new report from the Quincy Institute for Responsible Statecraft shows just how much things have changed. “China grew its Nature Index output by 95 percent from 2020 to 2024, while America’s rose just 9.5 percent. That’s not a competition — it’s a rout.”

After the US federal government mistakenly fired, and then re-hired, the publishing staff of the Centers for Disease Control and Prevention’s Morbidity and Mortality Weekly Report, the New England Journal of Medicine and the Center for Infectious Disease Research and Policy will begin offering their own public health alerts as an alternative source of information.

Publishers across all sectors are seeing steady declines in Google referral traffic as AI Overviews absorb more queries and users increasingly turn to LLMs instead of traditional searches. We wrote about this trend and its potential implications in the July issue of The Brief. Notably, The Economist has outlined its own roadmap for the post-search world. The publisher is investing in content innovation, on-site AI tools, and a renewed focus on brand marketing, discovery, reader engagement, and differentiated subscriber benefits — while remaining cautious about AI licensing deals. While scholarly and STM publishers operate in a very different market and use different business models, The Economist’s approach offers useful inspiration. It’s a good time for associations and publishers to take a harder look at how greater focus on audience engagement, content innovation, and user experience can help build deeper, more direct relationships with readers, authors, and members. In a post-search world, community and brand may well prove more important than ever.

Time spent on social media apparently peaked in 2022, and has since gone into steady decline, down 10% by the end of 2024. The one exception is use in North America, which over that period has continued to climb and now sits at a level 15% higher than that seen in Europe.

How good is your metadata? Find out using Crossref’s tremendously useful new Participation Reports tool.

Two new metrics caught our eye this month, although the utility of each remains somewhat elusive. The first is Scite’s new “Annual Rankings Report,” based on a “ratio of supporting to contrasting citations for any research entity.” It’s hard to grasp much rhyme or reason in the resulting rankings, which have a few expected journals at the top (Nature and Science at numbers 1 and 2), followed by some seemingly out of place journals, such as Scientific Reports (6th) ranking above the New England Journal of Medicine (8th), and Cell and The Lancet way down in 30th and 36th place, respectively, well behind PLOS One (9th). As with most-things-Scite, the citations that they can determine sentiment for (supporting or contrasting) are a tiny percentage of each journal’s total citations, and these rankings may be random noise or perhaps more dependent on journal style, linguistic quirks, or field-dependent citation behavior than on journal quality or influence. One could easily see a route to gaming this metric, simply by having copy editors tweak, to a more supportive tone, the phrasing related to citations to papers in one’s own journals. Perhaps more interesting, although probably not the “implosion” of the Impact Factor one supporter suggests, is the KGX3 engine from the Thomas Kuhn Foundation. KGX3 attempts to adapt knowledge-mining approaches in use for things such as consumer sentiment and economic intelligence toward identifying paradigm-shifting research in real time, rather than waiting decades for its impact to be recognized. While the resulting tool, Preprint Watch, takes into account the “epistemic role” of a preprint and whether it represents a “conceptual shift,” it does not, however, verify whether that paradigm-shifting, unreviewed research is valid or actually makes sense in any way. Perhaps there is some value in the tool for editors seeking flashier papers to bring attention to their journals.

A new Google Scholar–based tool offers impact metrics for researchers emphasizing whether they were the first or last author on the paper. This might make sense for fields where the practice is that the person who did most of the work is the first author and the person in whose lab the work was done is the last author, but this is not a universal custom and there are many fields where author order practices vary (e.g., in economics and mathematics, author order is often determined alphabetically). Even in fields with first/last author distinctions, any uptake of this tool would likely be ruthlessly (and quite easily) gamed. Would simply using CRediT author statements make more sense?

In other metric news, Altmetric has now added podcast mentions of research papers to its colorful donut. One thing to note is that, according to their documentation, Altmetric is not tracking the actual content of those podcasts, but rather seeking out “research mentions in the text descriptions of podcast episodes.” So, if you’re running a podcast that discusses research papers and care about inclusion in this metric, be sure to add citations to your text podcast descriptions.

In the latest example of why community-owned, grant-funded research infrastructure can be a tenuous proposition, Wellcome has had to step in with an emergency bailout of the FlyBase genetics database. FlyBase lost its funding when the National Institutes of Health (NIH) froze grants to its host institution, Harvard. While the Wellcome money is (sorry) welcome, it only represents a temporary reprieve and partial coverage for the personnel the resource needs. However, Wellcome optimistically notes that “This award will provide the necessary time for the Principal Investigators to establish a new longer term funding stream for the Cambridge FlyBase team.”

A study from Taylor & Francis and DataSeer offers an optimistic view of uptake of open data practices. They uncovered “much higher adoption than we had anticipated,” at least as determined by inclusion of data availability statements in papers, which occurred at 1.5 times the expected rate. A response from Glenn Hampson of the Science Communication Institute suggests this is perhaps an oversimplification of a complex set of behaviors, and that “we have a very long way to go before we can start claiming that we’re moving the needle on anything.”

After its acquisition by ReaderLink fell through, Baker & Taylor, founded in 1828 and one of the world’s largest distributors of materials to libraries, is going out of business.

Doug Armato, Director of the University of Minnesota Press for 27 years, has announced his retirement.

Geoffrey Bilder has joined MIT Press to lead technical innovation and strategic partnerships.

Chris Reid will be joining Wiley as senior director of product management for the Atypon Experience Platform, focused on core platform development and AI.

Maverick Publishing Specialists’ founder and President Martin Marlow is retiringafter 18 years at the firm.

Allen Powell has been appointed Interim CEO at EBSCO Information Services (an EBSCO Industries company). Powell replaces Annie Callahan, who had joined the organization in July of last year.  

And finally, a couple of online conversations that we found enlightening this month: first, Elsevier’s Michael Magoulias offers estimates of how much basic research in the US is actually funded by the federal government — 60% around 60 years ago, which had dwindled to 40% in 2022 and likely even less in coming years. Second, Lisa Janicke Hinchliffe and AJ Boston discuss just what is counted as an “institute of higher learning” by those compiling statistics, which apparently includes a significant number of beauty and barber schools.

***
“Cindy the Netherlands.” — The Wall Street Journal, apparently falling prey to AI copy editing applying the rules for country names to a story about Cindy Holland, a Netflix executive