News & Views: How does the growth of a particular publisher’s open access content factor into the relative value of a Big Deal? Part 2: The Findings

Heather Staines • May 17, 2022

This month’s News & Views is the second part of an article derived from ongoing research involving Michael Levine-Clark, John McDonald, Jason Price, and Heather Staines. This content, and the previous April 2022 News & Views, is based on a session presented at the Charleston Conference in November 2021.


Introduction


If you missed our April News & Views containing Part 1 of this investigation, please review it here.

Now that data sources for both institutions and publishers have been identified and usage outliers removed, we can get to the analysis of whether growth in OA affects value.


First, let’s look at usage of our two publishers’ content at the broad level of Carnegie class. (Figure 1) To recap, there is a group of eight libraries in the R1 Very High Research activity category with an average student FTE of 37,000. Our 12 smaller institutions, the R2+ group, have an average student FTE of 8,700. We see a higher average usage in the R1 group, with 38-42 uses per student FTE. Meanwhile, the R2+ group average 13-15 uses per student. For both groups, open access gold usage accounted for approximately 10% of the controlled usage.

We can now look into whether the value of the Big Deal is declining.


Cost per Controlled Use (CPCU) 


The standard metric used to determine package value is cost per use (CPU). A decline in package value (increase in cost per use) would occur if there is an increase in total cost or a decrease in total usage (or both). However, to take into account the underlying usage of publisher-hosted gold open access articles, we need to remove that freely-available usage from the denominator. Cost per controlled use (CPCU) does just that. (Figure 2)


The next set of figures addresses each of three factors: total cost, total usage, and gold OA usage, depicting a detailed view and a summary view of each trend over the five years of the study. The last set of figures brings them back together, comparing cost per use and cost per controlled use over time. Each line represents an individual institution, and their colors are consistent throughout. The dashed lines are the Carnegie Very High Research institutions (or R1s for short). The solid lines include High Research Intensity (R2s) on down to Baccalaureate institutions (R2-Bac for short).


It will likely not be surprising to anyone that most institutions’ costs increased over time. Figure 3 also reveals that Publisher 1 shows more variation in the cumulative change in cost with a few R1 institutions showing quite large increases and two experiencing cost reductions. It is worth noting that the two institutions that had decreases for Publisher 1 also had decreased prices for Publisher 2.

Summarizing these same data (Figure 4), Publisher 1’s package cost (in purple) increased 4 to 6% per year and Publisher 2’s package cost (in red) increased 2 to 4% per year (with the error bars reflecting Standard Errors). Publisher 1’s cumulative cost increased more than Publisher 2’s in 2019 and 2020: these increasing costs create downward pressure on package value.

Next, we move to usage. To make it easier to see the trends, we plotted Total Item Requests on a log scale. (Figure 5) As one might expect, the large, research intensive institutions (dashed lines) had the highest usage for both publishers. Even with outliers removed, some institutions experienced large fluctuations in usage from year to year.

We also see differences in usage trends for the two publishers. Usage increased for 14 out of 20 institutions for Publisher 1 from 2017 through 2020, despite the start of ‘work from home’ in March of 2020. For Publisher 2 over the same period of time, 15 out of 20 institutions’ usage declined.

We can look at the average usage for each publisher to see whether total usage is declining. (Figure 6) For Publisher 1 (in purple), the answer is no. The increase seen from 2017 to 2020 amounts to almost 40% or an average of 13% increase per year. In contrast, usage of Publisher 2’s content declined about 15% in the same time period, starting in 2019 and resulting in an average decline of about 5% per year. Thus, in the case of Publisher 2, we see both an increase in price and a decrease in usage, both contributing to declining value.

The proportion of total usage of content that is freely available within the package is a third factor that can impact the value of journal packages over time. As detailed in Figure 7, that proportion is fairly small and fairly consistent across the institutions. These data are plotted on a 0 to 100% scale to highlight the fact that 85-95% of usage still comes from paywalled ‘controlled’ content. (Figure 7)

Now, let’s zoom in on that same data. In Figure 8, we see that trends in usage of publisher-hosted open access differ between these two publishers. Publisher 1, on the left, shows a clear increase in the proportion of OA usage. Publisher 2 is less clear with the proportion of hybrid usage of some institutions appearing to peak midway through.

When one looks at the averages in Figure 9, the differences are clear. The proportional use of Publisher 1’s hybrid open access articles is increasing by about 1 to 1.5% per year (in purple), while Publisher 2’s usage is effectively flat. The right side of the figure uses a 0 to 100% scale to remind us that the annual average change in hybrid OA usage has still been quite small.

Next we move to cost per use. (Figure 10) Once we put the costs and total usage together to calculate cost per use, we learn that institutions tend to experience flat or decreasing cost per use for Publisher 1 and flat or increasing cost per use for Publisher 2. Because the change in the proportion of hybrid use is so small, when we adjust for cost per controlled use (Figure 11), the lines only move up a bit, retaining their shape, to reflect the removal of hybrid open access use from the denominator.

In the summary chart, we compare cost per controlled use and cost per use directly. (Figure 12) The cost per controlled use, indicated by the solid lines, is always higher than traditional cost per use (dashed lines). The difference is driven by the proportion of hybrid open access.

To the extent that transformative agreements lead to more rapid increase in the proportion of hybrid OA content and use over time, then these lines should diverge, with cost per controlled use increasing faster than traditional cost per use. Given this, we suggest that cost per controlled use should be used as a standard measure going forward.

Now, when we bring cost per controlled use from the two publishers together on the same axis, we can see a clear difference in their multi-year trends. (Figure 13) The purple line (Publisher 1) shows a slow decrease in cost per controlled use over time, whereas Publisher 2 shows a fairly rapid increase in 2019 and 2020. Next, we can reverse the y axis to acknowledge that increasing cost per controlled use equates to a decline in value. Therefore, Publisher 1’s package (in purple) is slowly increasing in value, and Publisher 2’s (in red) shows a sharp decline. (Figure 13) So, value did decline for one of the two publishers examined. To review: (1) average cost increased 2 to 6% per year, (2) average usage changed by between +10 and -5% per year over a four year period (and it was by far the most variable across institutions), and (3) the average proportion of hybrid OA usage changed by about 1% per year.

Conclusion


Some final thoughts: (1) Overall usage was a stronger influence on the change in value than the small changes in the proportion of hybrid OA article usage. (2) Despite the range of research activity levels across our institutions, there wasn’t much difference in the proportion of the open versus controlled usage across the site-licensed institutions for either publisher. (3) COVID likely affected these trends, but precisely how was unclear. Did lockdown increase the usage or limit it? Did it affect our two publishers differently? We have no ‘non-COVID’ control unfortunately. (4) If the impact of transformative agreements on the rate of hybrid OA article output influenced these trends, the impact was quite small. Still, with more libraries negotiating transformative agreements, growth in the proportion of OA articles should accelerate. As long as usage in publisher packages continues to grow, cost per controlled use will increase more quickly than cost per use. This new cost per controlled use metric should help libraries track the return on investment from their journal package subscription payments as a growing proportion of underlying articles are free to read.



This article is © 2022 Delta Think, Inc. It is published under a Creative Commons Attribution-NonCommercial 4.0 International License. Please do get in touch if you want to use it in other contexts – we’re usually pretty accommodating.

News & Views: The Impact of Inflation on APC Costs

By Dan Pollock and Heather Staines • April 22, 2025
In March 2025, we looked at the latest Article Processing Charges (APCs) . This month we focus on how prices have risen relative to inflation. As APC price increases fall back to trend, what does this mean in real terms? Background Each year we survey the list Article Processing Charges (APCs) of more than 30 major and significant publishers. Going back to 2015, the dataset includes more than 20,000 unique titles and 150,000 title per year combinations. Going into 2025, we saw price increases fall back to long-term trends from their unusually high increases in 2024. Fully OA (“gold”) journal list prices across our sample rose by around 6.5%, compared with a 9.5% increase this time last year. Hybrid list prices rose by an average of 3%, compared with 4.2%. Last year’s price rises were above long term trends, but overall we found they were rising below inflation. How does this hold for this year’s price increases? We again use the global Consumer Price Index (CPI) as our inflation index, as we consider it to represent the most realistic view of our marketplace. Prices exclude zero APCs, so we can see the effects for instances when publishers choose to charge APCs. Are APCs becoming cheaper or more expensive? The chart below shows how increases in all list APCs work out in real terms for both hybrid and fully OA journals.

News & Views Webinar: Open Access Charges 2025

By Heather Staines • April 7, 2025
We are proud to share a video recording of our March News & Views companion online discussion forum! Join us to hear the latest trends around APC data, including APCs for both fully OA and hybrid journals. We'll talk about what we're seeing in relation to recent years and discuss the broader context for the APC market. If you missed the session, or if you attended and would like to watch/listen again, or share forward with friends, please feel free!

Author/Researcher Survey: Impact of Potential Funding Reductions on Research, Publishing, and Conference Attendance

By Lori Carlin • March 28, 2025
Delta Think is currently spearheading an industry market research survey to authors and researchers across the scholarly community designed to provide insight into the impact of potential US federal funding reductions on their research. The survey addresses topics such as publication volume, their ability/allowance for peer review, conference participation and attendance, influence on their research scope and topics, and more. Working in collaboration with nearly 25 scholarly societies, we are launching this initiative to capture the real-world impact of these potential changes in order to help societies better plan and support their members, researchers, and authors. The results of the survey will provide scholarly publishers with systematic, quantitative voice-of-market data to inform evidenced-based strategy development and scenario planning in a rapidly changing funding landscape and policy environment. The survey opens this week, with each participating society distributing the link to their own communities. All participating societies will receive an in-depth analysis of the full survey results, filtered by various demographics such as country, career stage, and discipline, as well as options for Delta Think to analyze their specific community data or the raw data from their specific community so they can analyze it themselves. Delta Think has designed the survey and will conduct all the analysis of the results.

News & Views: Open Access Charges – Price Increases Back on Trend

By Dan Pollock and Heather Staines • March 13, 2025
This month we look at our latest data about Article Processing Charges (APCs). Per article pricing is a fundamental building block for all paid publishing models, so our review provides an invaluable insight into how costs of open access continue to evolve. APC prices in general continue to increase, but at a slower rate compared with this time last year. Important nuances in the distribution of prices continue to affect the value and cost of paid publishing models. Background Each year we survey the list Article Processing Charges (APCs) of a sample of major and significant publishers. Covering more than 20,000 titles going back to 2016, our dataset represents one of the most comprehensive reviews of open access pricing. To compare like for like, we consistently analyze non-discounted, CC BY charges. We take a snapshot at the end of every January, so we can track yearly changes while controlling for the different times of year that publishers may update prices. Our statistics exclude zero or unspecified APCs, although these are included in our underlying data. This allows us to understand trends where publishers choose to charge APCs without skewing averages. We run separate analyses around APC-free models. Headline Changes Going into 2025, we have seen APC pricing increasing but falling back to long-term trends. Fully OA APC list prices across our sample have risen by around 6.5% compared with 9.5% this time last year. Hybrid APC list prices have risen by an average of 3% compared with 4.2% this time last year. Maximum APCs for fully OA journals remain at $8,900. Maximum APCs for hybrid journals now top out at $12,690 (up $400 from last year). Big jumps in prices happened last year, driven by high inflation. In 2020-2021 prices were driven up when high-impact journals began offering OA options for the first time. In both cases, increases subsequently fell back to averages. Underlying trends continue. There are around 2.6x more hybrid journals than fully OA ones, down from 2.9x a year ago. Hybrid journals follow (or, rather, set) a similar pattern to the market overall. On average, fully OA prices are around 64% of those of hybrids. This is a couple of percentage points higher than long term trends. Around 31% of our sample of fully OA journals charge no APCs. (We have separately analyzed the number of articles in OA journals.) Price rises vary significantly by discipline. Arts & Humanities and Social Sciences have seen particularly large average increases, especially in fully OA journal prices. Price Distribution Market-wide headline price changes mask important nuances. We have discussed previously that the most important nuance lies in the spread of prices within a given publisher’s portfolio. For example, if the bulk of a publisher’s journals lie toward the lower end of its pricing, with just a few journals priced at the high end, the average (mean) price will be higher than most authors pay. The following figures show how the spread of prices plays out in the market across our sample of publishers. The figures are outlines of histograms, showing how many titles sit in various price bands over the successive years of data we have curated. The red line shows the most recent year’s prices. The lines become greener as they go further back in time. Subscribers to Delta Think’s Data and Analytics Tool can see full details of axes. Hybrid Prices The spread of price bands for hybrid journals is shown in Figure 1 below.

Books: The Hidden Goldmine in Your 2025 Content Strategy

By Diane Harnish and Meg White • February 27, 2025
User information needs as well as funding models are evolving rapidly, as evidenced by Clarivate’s recent move to phase out perpetual access purchases for print, eBooks, and digital collections by the end of 2025. Taking a hard look at how these assets contribute to your portfolio and overall organizational strategy has never been more critical. A holistic books program assessment can help you think intentionally about how books and book-based content can help meet customer and market needs. Publishing and Product Strategy A market-driven publishing and product strategy begins with an understanding of customer information needs. What markets, segments, information needs, and challenges are present? How can customer information needs be addressed? What role can our book content play? How can we differentiate our solutions? Can our book content contribute to a unique value proposition? Thinking creatively about how your content meets market needs is critical; think solutions, not printed pages and chapters. Commercial Strategy A detailed commercial strategy, supported by proper resources, is fundamental to success. Leveraging a clear understanding of customer preferences and delivering messaging that resonates with your specific market segments and use cases is essential. What are the best methods to generate market awareness? When and how should we communicate with key audiences? What messages resonate best? What sales and marketing capabilities do we have internally? Where do we need to partner to reach core audiences? How do we meet global needs? Do we have the appropriate access, pricing, and distribution models in place to meet customer expectations? What do we need to do directly? Where should we cultivate successful channel partnerships? And you don’t have to go it alone; a commercial strategy is best formulated and executed by a combination of internal and external resources. Technology Infrastructure Is your technology optimized to support your book program? From agile content management systems to product platforms to customer relationship management tools, the right tools enable your content and commercial strategy. What systems do we need to ensure efficiency in our publishing processes and quality and integrity in our content? What technologies and platforms do we need to build market-responsive products? What systems do we need to communicate effectively and meaningfully with our customers, including authors? Are we best served by building these systems or partnering? Successfully integrating and leveraging new technologies, such as AI, requires a fundamental understanding of markets and customer information needs . The Numbers Financial metrics are a key measure of the health of any program. An in-depth assessment of a program’s recent performance is a vital tool to help identify strengths, weaknesses, and gaps, and help to surface areas for improvement and corrective action. A financial analysis will clarify: What is our book and content annual output? Is it sufficient to support our strategy and meet customer and market needs? What is our cost structure? Our pricing strategy? Do they align with industry and market norms and expectations? Do we have the appropriate mix of internal and external resources in place to support our strategy? How can we best align our financial performance to contribute to the organization’s larger strategy? Beyond red ink or black ink, financial analysis will provide answers to these questions. Assessing Your Book Program Delta Think partners with publishers to do the foundational analysis necessary to understand how your book and book-based content can be a vital part of your content portfolio and support your organization’s goals and objectives. Our processes, including program benchmarking, stakeholder interviews, surveys, and workshops, combined with expert landscape research and analysis ensure you are building a content strategy that is market-focused and customer-driven. Contact Delta Think at info@deltathink.com to set up a time for a video call to learn more. We will also be attending the London Book Fair, March 11-13, 2025, if you’d like to schedule an in-person chat.

News & Views: What proportion of OA is Diamond OA? Can AI help us find out?

By Dan Pollock and Ann Michael • February 20, 2025
Overview A recent post on the Open Café listserv posed a question about the true extent of fee-free open access publishing, but it noted the incomplete coverage of the data cited. We have more comprehensive data, but just as we started our analysis, DeepSeek’s release sent markets into turmoil. The stage was set for a timely experiment. We first answer the question using our data. Then we see how the AI did. Background What proportion of open access is not paid for by APCs? In discussing this, a recent Open Café listserv post cited studies by Walt Crawford – a librarian, well-known in the academic library and OA communities for his analysis of open access. He has paid particular attention to “diamond” OA journals, which charge neither readers nor authors. His studies are based on data from the Directory of Open Access journals ( DOAJ ). Excellent though both sources may be – and, full disclosure, we contribute to the DOAJ – the DOAJ’s remit covers only fully OA (“gold”) journals. As listserv founder Rick Anderson noted, “By counting only articles published in DOAJ-listed journals, Crawford’s studies radically _undercount_ the number of APC-funded OA articles published – because DOAJ does not list hybrid journals, which always charge an APC for OA and which produce a lot of genuinely OA articles (though exactly how many, no one knows).” Using our data Actually, we do know … or at least have some fair estimates of hybrid OA. Our data allows us to determine the share of open access output in APC-free journals, as follows.

News & Views Special Edition: How much scholarly publishing is affected by US Presidential Executive Orders?

By Dan Pollock and Ann Michael • February 11, 2025
Overview Following the 2024 US election, the new US administration has instructed employees in some key federal agencies to retract publications arising from federally funded research. This is to allow representatives of the administration to review the language used, to ensure it is consistent with the administration’s political ideology. In this special edition of News & Views, we quantify how many papers might be affected and estimate their share of scholarly publishers’ output. The initial numbers may be small, but we suggest the effects on scholarly publishing could be profound. Background On 20 January 2025, Donald J. Trump took office as the 47th President of the United States. Within hours he signed an Executive Order 1 (EO) 14168 proclaiming that the US government would only recognize two sexes, and ending diversity, equity, and inclusion (DEI) programs inside federal agencies. The following day, his administration instructed federal health agencies to pause all external communications – “such as health advisories, weekly scientific reports, updates to websites and social media posts” – pending their review by presidential appointees. These instructions were delivered to staff at agencies inside the Department of Health and Human Services (DHSS), including the Food and Drug Administration (FDA), the Centers for Disease Control (CDC) and Prevention, and the National Institutes of Health (NIH). The events that followed are important, as they directly affect scholarly papers and our analysis. A memo on 29 January instructed agencies to “end all agency programs that … promote or reflect gender ideology” as defined in the EO. Department heads were instructed to immediately review and terminate any “programs, contracts, and grants” that “promote or inculcate gender ideology.” Among other things, they were to remove any public-facing documents or policies that are trans-affirming and replace the term “gender” with “sex” on official documents. By the start of February, more than 8000 web pages across more than a dozen US government websites were taken down . These included over 3000 pages from the CDC (including 1000 research articles filed under preventing chronic disease, STD treatment guidelines , information about Alzheimer’s warning signs, overdose prevention training , and vaccine guidelines for pregnancy). Other departments affected included the FDA (some clinical trials), the Office of Scientific and Technical Information (the OSTP, removing papers in optics, chemistry and experimental medicine), the Health Resources and Services Administration (covering care for women with opioid addictions, and an FAQ about the Mpox vaccine). Around this time, it further emerged that CDC staff were sent an email directing them to withdraw manuscripts that had been accepted, but not yet published, that did not comply with the EO. Agency staff members were given a list of about 20 forbidden terms, including gender, transgender, pregnant person, pregnant people, LGBT, transsexual, nonbinary, assigned male at birth, assigned female at birth, biologically male, biologically female, and he/she/they/them. All references to DEI and inclusion are also to be removed. The effects of the EO Commenting on the merits of policy and ideology lies beyond our remit. However, when these matters affect the scholarly record – as they clearly do here – then they are of interest for our analyses. Specifically, what might the effects of the EO be on the publication of papers, and what effects might accrue from withdrawal of research funding? If federal agencies are being instructed to withhold or withdraw submissions, then, to quantify what this might mean to publishers, we have estimated the volume of output from a few key federal agencies. It is summarized in the following chart. 

New Year – New Technology: What Problem Are You Trying to Solve?

By Lori Carlin • January 23, 2025
Emerging technologies are reshaping how we create, distribute, and consume content. Publishers face the critical task of making smart technology investments to stay competitive and enable strategic objectives. How do you ensure that your next tech purchase aligns with your organization's needs and goals? Enter the needs assessment process – your roadmap to making informed, strategic technology decisions. From defining clear objectives to creating a comprehensive RFP, these best practices will help you navigate the decision-making process with confidence and ensure that your investments deliver value for your organization and your customers. Technology is not a solution; it is a tool. The temptation to adopt technology without a clear definition of what you are trying to achieve is an all too common (and usually very costly) mistake. Does your strategy include delivering a more personalized experience for your users? A customer data platform may be the right technology. Interested in using AI to build research integrity into your editorial process? Perhaps it’s time to revisit the capabilities of your editorial management system. Looking to support education and learning for students, faculty, and professional learners? Maybe it is time to evaluate formal learning management systems. Once you are confident about what you are seeking to achieve, the real work begins. Here are the key components that will help lay the foundation for a successful process from inception to deployment: Analyze Current State: Audit existing systems and processes to understand current capabilities and limitations. Conduct a Gap Analysis: Identify gaps between current capabilities and desired future state. Collect and Analyze Data: Gather qualitative and quantitative data from staff, users, customers, industry benchmarks, and about existing systems. Consider Resources and Constraints: Assess available resources, including budget, skills, and time. Research Solutions: Investigate potential technologies and/or types of solutions that could address identified gaps. Prioritize Needs: Work with stakeholders to prioritize needs based on impact and feasibility. Create RFP: After identifying prioritized needs and potential solutions, develop an RFP that clearly outlines project objectives, specific requirements, evaluation criteria, budget, and timelines. Distribute the RFP: Identify vendors with fit for purpose solutions and capabilities and distribute. Evaluate Proposals: Review vendor responses against established criteria and prioritize them based on how well they meet your needs. Plan for Adoption and Training: Consider the change management aspects of introducing new technology and processes. Be sure to develop a plan for user adoption, training, and ongoing support in your new systems. Technology as a Strategic Ally A methodical needs assessment is not just a procurement exercise – it is a strategic opportunity to reimagine how technology can transform your organization. The most successful technology investments are those that solve real problems, align with organizational goals, and empower your team to work more efficiently and creatively. Don’t fall into the trap of just moving what you are currently doing over to a new system. This is an ideal occasion to think about how you would design workflows and processes if you were to start from scratch and use that framework to evaluate the new capabilities available. You don’t want to duplicate what you are doing today; you want to step back and take the opportunity to build something better whenever possible. Customer Data Platform? Editorial Management System? Learning Management System? Something Else? Delta Think partners with publishers to do the foundational and implementation work required to ensure that technology decisions match the organization’s capabilities, fit the budget, and are grounded in voice-of-customer data. Our processes, including stakeholder interviews, surveys, and workshops, combined with expert landscape research, analysis, and assessments, underpin technology decision-making that is market-focused and customer-driven. If your 2025 objectives depend on or are enabled by technology, we’d welcome the opportunity to help you learn, plan, achieve. Please contact us today to start the conversation.

News & Views: How reliable are the headlines you read in reports?

By Dan Pollock and Heather Staines • January 14, 2025
A number of sources provide information about patterns in the overall scholarly journals market. However, as we so often mention in our analyses, important nuances lie beneath the headlines. This month we explore just how much variation exists and highlight the importance of specificity. Background As part of our annual market updates, we estimate the proportions of open vs. subscription access content each year. Over the last few years, we have observed how OA has approached 50% of output, but we note that it has yet to punch through that number. However, this headline varies greatly depending on your area of publishing. An example from physics The chart below shows the nuances across just a few of the 200+ subjects that we track.

News and Views: How much content can AI legally exploit?

By Dan Pollock, Ann Michael • December 10, 2024
This month’s topic: How much content can AI legally exploit? Scroll down to read about this topic, along with the latest headlines and announcements. Delta Think publishes this News & Views mailing in conjunction with its Data & Analytics Tool . Please forward News & Views to colleagues and friends, who can register to receive News & Views for free each month. Delta Think will be attending several upcoming conferences, including APE (Jan 14-15), NISO Plus (Feb 10-12), and Researcher to Reader (Feb 20-21). We would love to see you there – please get in touch or visit our Events page to see all the meetings we will be attending. How much content can AI legally exploit? O verview During the recent PubsTech conference , we were asked how much content could be legitimately used to train artificial intelligence systems without being specifically secured through a licensing agreement. In considering this question, we find some counterintuitive results. Background Generative AI (genAI) is a type of artificial intelligence that can create new content—text, images, music, and more – by analyzing patterns in massive datasets. These models are typically trained on publicly available data scraped from the web. In the US, developers often invoke the “Fair Use” copyright doctrine to justify this training, claiming it is limited to specific purposes (training) and transformative in nature (different from the original use). In reality, the legal position is complex and evolving , with many rights holders and their representatives – unsurprisingly – taking the opposite view. Even if legal clarity emerges, different geographies and jurisdictions will likely reach different conclusions. The legal complexities of AI and copyright law are beyond our scope. However, for scholarly publishers, particular issues apply. Half of our output is open access , and open access content is designed to be reusable. Open or not, content has varying restrictions on onward use – for example, non-commercial use is often allowed with attribution. How much scholarly content is exploitable?  For the purposes of analysis, we will assume that the license under which content is published will have a material bearing on the legitimacy of its use to train AI systems. Therefore, looking at share of licenses, we might be able to answer our question.
More Posts