The library provides access to over 3 million e-books (based on our 23-24 reporting to ARL). Most of that access is acquired through e-book packages and EBSCO is a big one, with over 370,000 titles in a range of subject areas. EBSCO sends us quarterly reports on the usage of those titles, with statistics on views and downloads by title, purchase type, publisher and subject area.\u00a0\u00a0<\/p>\n\n\n\n
Last week I received our report for March through May of 2025. The csv file was 5620 rows, so I asked ChatCPT to help me make sense of it. I uploaded the data and used this prompt:\u00a0<\/p>\n\n\n\n
\n\nAnalyze this file please. It is an excel spreadsheet that lists the usage of electronic books at an academic library. I would like you to provide top level insights into the usage by numbers and content trends.\u00a0<\/p>\n\n\n\n
\n<\/div>\n<\/div>\n<\/blockquote>\n<\/blockquote>\n\n\n\nWithin seconds, ChatGPT provided me with this: <\/p>\n\n\n\n
\ud83d\udcca Overall Usage Statistics<\/em><\/strong>\u00a0
Total Accesses<\/strong>: 15,892\u00a0
Abstract Requests<\/strong>: 15,061 \u2013 indicates strong exploratory or research use\u00a0
Online Views<\/strong>: 13,363 \u2013 the dominant method of engagement\u00a0
Chapter Downloads<\/strong>: 1,984\u00a0
Full Downloads<\/strong>: 545 \u2013 suggests some selective reading patterns\u00a0
Turnaways<\/strong>: 483 \u2013 may indicate unmet demand or access limitations\u00a0
Print & Email Requests<\/strong>: minimal (Print: 239, Email: 10)\u00a0
\ud83d\udcda Top Subject Areas by LC Subclass<\/em><\/strong>\u00a0
These reflect areas with the highest interest:\u00a0
E<\/strong> \u2013 U.S. History\u00a0
LA<\/strong> \u2013 History of Education\u00a0
PN<\/strong> \u2013 Literature (General)\u00a0
ML<\/strong> \u2013 Literature on Music\u00a0
DT<\/strong> \u2013 African History\u00a0
HQ<\/strong> \u2013 Women’s Studies, Sexuality, Family\u00a0
HV<\/strong> \u2013 Social Pathology, Criminology, Social Work\u00a0
RC<\/strong> \u2013 Internal Medicine\u00a0
PS<\/strong> \u2013 American Literature\u00a0
LB<\/strong> \u2013 Theory and Practice of Education\u00a0
These trends suggest:\u00a0
Strong interdisciplinary interest spanning humanities, health, and education.\u00a0
Particularly high engagement with history, literature, education, and social sciences<\/strong>.\u00a0<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\nIt was interesting that the use of abstracts was so high, and I was surprised that U.S. history<\/strong> was the top subject. I was impressed with this output, and the speed with which the data was analyzed. But of course, there are experts here at the library who are in a far better position to evaluate the analysis \u2013 its accuracy and its value. I asked Karen Kohn, Collections Analysis Librarian, to review ChatGPT\u2019s output with me.<\/p>\n\n\n\n
Karen was patient. Typically she looks at usage data for an entire year rather than a quarter. That aligns with our renewal cycle, and of course we report these figures to organizations like ARL on an annual basis. <\/p>\n\n\n\n
Let\u2019s also keep in mind that EBSCO is one of dozens of e-book packages and platforms that we license. While EBSCO covers all subject areas, a trend within this package is not necessarily what we\u2019d see in JSTOR or ACM (Association of Computing Machinery). <\/p>\n\n\n\n
WIth those caveats, Karen and I continued our conversation with ChatGPT. <\/p>\n\n\n\n
I asked to see an analysis of \u201cturnaways.\u201d <\/p>\n\n\n\n
\ud83d\udcc9 Turnaways Analysis<\/strong>\u00a0
Turnaways occur when users attempt to access a title but are denied, often due to license restrictions (e.g., limited user seats).\u00a0
\ud83d\udea9 Top Turnaway Subject Areas (LC Subclass)<\/em><\/strong>\u00a0
HF<\/strong> \u2013 Commerce \/ Business Administration (257 turnaways)\u00a0
BL<\/strong> \u2013 Religion \/ Mythology (105)\u00a0
DT<\/strong> \u2013 African History (33)\u00a0
E<\/strong> \u2013 U.S. History (27)\u00a0
RA<\/strong> \u2013 Public Health (15)\u00a0
This suggests a notable unmet demand in business, religion, and global history\/public health content<\/strong>, which may merit review of current licensing models or potential acquisition of multi-user or unlimited access licenses.\u00a0
Let me know if you\u2019d like a visual of subject turnaways or recommendations on mitigating high-turnaway titles.\u00a0<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<\/p>\n\n\n\n
Karen says, <\/p>\n\n\n\n
\nI can see places where it\u2019s especially important to have the person writing the prompts think carefully about what to ask and look closely at the results.\u00a0<\/em><\/p>\n\n\n\n
There are a handful of serious outliers in this dataset, it doesn\u2019t always know when to check for them. At one point ChatGPT points out that What the Best College Students Do accounts for 90% of Belknap Press\u2019s total usage. Later the report notes that there are 257 turnaways in the HF section, which \u201csuggests a notable unmet demand,\u201d but it doesn\u2019t point out that 256 of these are for a single book. \u00a0<\/p>\n<\/blockquote>\n\n\n\n
She\u2019s right. There needs to be a human who is examining the data, looking for inconsistencies and prompting ChatGPT when there are results that need clarification. One of the ironies of using AI tools is this: we ask an AI tool because we don\u2019t know an answer, but to best evaluate the output, we need some knowledge of the subject.\u00a0Karen continues, explaining that:<\/p>\n\n\n\n
\nThis is analysis that I could do in Excel using pivot tables, but I recognize that not everyone has the same Excel knowledge, even people who look at collections data as part of their jobs. As someone who is comfortable in Excel, I really like to do things manually so I have more control over how the data is displayed and can look closer at specific details when I have questions, but not everyone wants to or can make their own Excel tables.<\/em><\/p>\n<\/blockquote>\n\n\n\n
Fair point. Look out for Part II of this post, in which I ask ChatCPT to draft training materials on creating a Excel Pivot chart.\u00a0And a big thank you to Karen Kohn who helped to remind us of the need for human intervention and a critical eye when incorporating AI tools into our work with library assessment. <\/p>\n\n\n\n
<\/p>\n","protected":false},"excerpt":{"rendered":"
The library provides access to over 3 million e-books (based on our 23-24 reporting to ARL). Most of that access is acquired through e-book packages and EBSCO is a big one, with over 370,000 titles in a range of subject … Continue reading