The DOI matches its Crossref record; 58 of the 63 references it cites were found in a registry or an index, and 5 are not checked yet; none of the 37 sources its claims cite is checked for support yet.
AI on the Shoulders of Giants: Using Kuhlthau’s Information Search Process to Improve AI Support for Information-Seeking
Open accessAbstract
Librarian-patron exchanges parallel chatbot-user interactions: Experienced librarians provide better reference service by prompting users with clarifying questions; likewise, chatbots can improve their responses with better-engineered prompts. A chatbot’s self-assessment mechanisms transform prompt components into predictions formed into responses, a process that enables large language models to continually learn and improve from prompts engineered to produce desired outcomes. Kuhlthau’s Information Search Process (ISP), created to describe library patrons’ information-seeking stages and dispositions, may be useful for understanding and improving users’ interactions with AI-driven chatbots. Kuhlthau has detailed seven stages of information-seeking, with intervention zones and reference mediation levels, for librarians to assist users in reaching satisfying outcomes in offline and online environments. We hypothesize that generative AI agents can use ISP’s zones of intervention and levels of reference mediation to improve their large language models to assist users better. Guided by the overarching question “To what extent does applying ISP to virtual reference transactions present strategies for improving prompt engineering?” we explored a recent virtual reference dataset ( N = 1,959) to determine whether users’ questions reflected Kuhlthau’s ISP and how librarian-patron interactions used clarifying questions and interventions to ascertain users’ information-seeking stage and provide satisfying responses. Our investigation demonstrated that authentic patron questions clustered in ISP’s focus formulation and information gathering stages. When exemplar questions were posed to a chatbot, it demonstrated a strong ability to detect and respond to the information need, often providing scaffolds to the next step of the information-seeking process. These results suggest that chatbots can be improved by blending web searching, database searching, and generative AI to seamlessly meet user needs, while task definition, topic selection, and search assessment should remain human-mediated aspects of information-seeking.
Checked 2026-11-18 · 5 references not checked yet
Overview
CoChat Summary
TL;DR
The paper applies Kuhlthau’s Information Search Process (ISP) to authentic virtual reference questions and tests whether ChatGPT 4o can respond in ways aligned with ISP stages, zones of intervention, and reference mediation levels. It argues chatbots can support middle ISP stages but should be blended with human mediation for task definition, topic selection, and search assessment.
Key findings
- In Florida Ask a Librarian transactions from October 2023 (N = 1,959), local help was most common (n = 641, 33%), followed by research (n = 608, 31%) and reference (n = 509, 26%); mean transaction duration was 6.0 minutes (median 4.1).
- Among research questions (n = 608), ISP Stage 3 Prefocus Exploration was most frequent (n = 381, 63%), followed by Stage 5 Information Collection (n = 70, 12%), Stage 1 Task Initiation (n = 63, 10%), Stage 4 Focus Formulation (n = 57, 9%), Stage 2 Topic Selection (n = 33, 5%), and Stage 6 Preparing to Write (n = 4); no questions reflected Stage 7.
- ChatGPT 4o successfully identified the zone of intervention for exemplar Stage 3, Stage 4, and Stage 5 questions; for Stage 3 and Stage 4 it offered bridges to the next ISP stage, but for Stage 5 it did not suggest additional questions to bridge to Stage 6 or revisit topic focus.
- The authors conclude chatbots can be improved by blending web searching, database searching, and generative AI, while task definition, topic selection, and search assessment should remain human-mediated.
Methods
- Exploratory study using Florida’s Ask a Librarian data from October 2023 (N = 1,959), including initial questions, timestamps, transaction duration, and number of questions; AAL provided only initial questions, not full transaction text.
- Deductively coded all questions by AAL categories (Research, Reference, Local Help, combinations) and coded Research questions (n = 608) for ISP stage, intervention zone, and reference mediation level; two coders resolved differences collaboratively.
- Ran frequencies and descriptive statistics, then posed three exemplar AAL Research questions representing ISP Stages 3, 4, and 5 to ChatGPT 4o and qualitatively analyzed responses for ISP-linked support elements.
Limitations
- Stated: AAL privacy policies meant the dataset lacked full reference transaction text, including librarians’ responses and follow-up questions; not all users fit ISP or seek information to complete a specific task; cited virtual reference and chatbot studies may reflect older technologies due to publication lag; the study is exploratory.
- Evident from the text: only three exemplar questions were posed to one chatbot version (ChatGPT 4o), and chatbot responses were analyzed qualitatively rather than systematically coded; the dataset covers one state service and one month, limiting generalizability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
0 Valid · 0 Not valid · 34 Not checked · 73 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“ISP Stage 3, Prefocus Exploration, was assigned to 381 (63 percent) questions”
plainn = 608- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We did not expect Stage 7 to appear in questions, because it is an independent stage; our analysis confirmed this assumption.”
boostedn = 608- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
In the paper
“Chatbots did not provide links to resources in subscription databases but, rather, to web-accessible reports, sites, and articles related to the question (Lai 2023).”
What the sources say
- Not checked yetcites Lai 2023 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The chatbot did respond to the questioner's request for information from trade publications with highlights from those resources, including promising focus areas.”
plainstates a causeSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 2 assumptions not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“most questions (n = 641, or 33 percent) related to local information”
plainn = 641- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
31 more claims
Create a free account to see all 107 claims and to verify the 27 references still open.
- Probably not a substantial claim. 57
- It may state two separate claims as one. 25
- The quote only partly states this claim. 2
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ Abstract
“Our investigation demonstrated that authentic patron questions clustered in ISP's focus formulation and information gathering stages.”
§ ISP Stage, Intervention Zone, and Reference Mediation Level Frequencies
“ISP Stage 3, Prefocus Exploration, was assigned to 381 (63 percent) questions”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“| Research and Local Help | 1 | 14.4 | 6.5 |”
Reported 6.5 · recomputed 14.4
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ ISP Stage, Intervention Zone, and Reference Mediation Level Frequencies
“sixty-three (10 percent) reflected ISP Stage 1, Task Initiation, and thirty-three (5 percent) reflected Stage 2, Topic Selection, thus suggesting a zone of intervention including process intervention, with the reference staff in an advisory role.”
§ Limits to Study Interpretation
“| Stage 1, Task Initiation | Zone 5, Process | Level 5, | Assistance in understanding the | | Stage 2, Topic | Intervention | Counselor | process, developing search |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ ISP zones of intervention and levels of reference mediation
“This zone of intervention tends to occur in ISP Stage 4, Focus Formulation, because topic focus and visualization of the preferred outcome are often aided by assistance in identifying, locating, and using a variety of relevant resources (Kuhlthau 2004).”
§ Limits to Study Interpretation
“| Stage 5, Information | Zone 2, Single | Level 2, | Locate a user-specified journal | | Collection | Source | Locator | article |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Data Analysis
“Using just questions with any element of research (n = 608)”
§ Limits to Study Interpretation
“| Research | 608 | 7.3 (7.1) | 4.5 | | Local Help | 641 | 4.4 (5.6) | 3.0 | | Reference and Research | 61 | 7.7 (8.1) | 5.1 | | Reference and Local Help | 15 | 8.2 (4.5) | 14.4 | | Research and Local Help | 1 | 14.4 | 6.5 |”
Citations
Cites
ChatGPT for good? On opportunities and challenges of large language models for education
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · Learning and Individual Differences · doi:10.1016/j.lindif.2023.102274 · Cited by 7,326 · cited 1×
“Because chat-bots are pretrained on vast amounts of data with transformer architectures known as large language models (LLMs) (Kasneci et al. 2023), current content-generating (generative) AI tools such as the chatbot ChatGPT have an exceptional ability to generate humanlike text.”
AI-based chatbots in customer service and their effects on user compliance
Adam, M., Wessel, M., & Benlian, A. (2021) · Electronic Markets · doi:10.1007/s12525-020-00414-7 · Cited by 1,195 · cited 1×
“Chatbots improve customer interactions with fast responses to easy customer questions so human employees can focus on complex problems (Adam et al. 2021).”
Developing and validating a service robot integration willingness scale
Lu, L., Cai, R., & Gursoy, D. (2019) · International Journal of Hospitality Management · doi:10.1016/j.ijhm.2019.01.005 · Cited by 1,013 · cited 3×
“…interactions depends primarily on perceived usefulness and ease of use informed by their expectations and perceptions of whether the chatbot met or exceeded a human’s performance (Lu et al. 2019; McLean and Osei-Frimpong 2019; Thomas et al. 2020) in a warm, supportive manner (Terzis et al. 2012).”
What factors contribute to the acceptance of artificial intelligence? A systematic review
Kelly, S., Kaye, S.-A., & Oviedo-Trespalacios, O. (2023) · Telematics and Informatics · doi:10.1016/j.tele.2022.101925 · Cited by 1,008
Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm
Reynolds, L., & McDonell, K. (2021) · Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3411763.3451760 · Cited by 952
Hill, J., Randolph Ford, W., & Farreras, I. G. (2015) · Computers in Human Behavior · doi:10.1016/j.chb.2015.02.026 · Cited by 856 · cited 1×
“…hinder overall interaction (Ashktorab et al. 2019), mainly when users’ queries include profanity and grammatical mistakes that impair the chatbot’s capabilities and result in poor responses (Hill et al. 2015; Lu et al. 2019; McLean and Osei-Frimpong 2019; Thomas et al. 2020; Zarouali et al. 2018).”
My Chatbot Companion - a Study of Human-Chatbot Relationships
Skjuve, M., Følstad, A., Fostervold, K. I., & Brandtzaeg, P. B. (2021) · International Journal of Human-Computer Studies · doi:10.1016/j.ijhcs.2021.102601 · Cited by 797 · cited 1×
“Because users spend most of their time seeking information while ignoring the question identification and focus formulation stages, they will benefit from chatbots’ improved ability to detect users’ moods and mind-sets (Skjuve et al. 2021).”
Brandtzaeg, P. B., & Følstad, A. (2017) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-70284-1_30 · Cited by 788 · cited 1×
“Users appreciate good user interface and visual attractiveness (Yang and Lee 2019) and are often dismayed when chatbots’ attempts to mimic humanlike interaction seem inauthen-tic (Brandtzaeg and Følstad 2017; Lu et al. 2019; Tsiourti et al. 2018).”
Information-seeking, curiosity, and attention: computational and neural mechanisms
Gottlieb, J., Oudeyer, P.-Y., Lopes, M., & Baranes, A. (2013) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2013.09.001 · Cited by 784
McLean, G., & Osei-Frimpong, K. (2019) · Computers in Human Behavior · doi:10.1016/j.chb.2019.05.009 · Cited by 762
Research works printed but not linked: 4
- Kuhlthau, Carol C. 2001. "Rethinking Libraries for the Information Age School: Vital Roles in Inquiry Learning." Keynote address at the International Association of School Librarian- ship Conference & International Research Forum on Research in School Librarianship, Auckland, July 9. https://wp.comminfo.rutgers.edu/ckuhlthau/iasl_notes/.
- Kuhlthau, Carol C., Jannica Heinström, and Ross J. Todd. 2008. "The 'Information Search Process' Revisited: Is the Model Still Useful?" Information Research 13 (4): 355. https:// InformationR.net/ir/13-4/paper355.html.
- O'Leary, Daniel E. 2024. "Do ChatGPT 4o, 4, and 3.5 Generate 'Similar' Ratings? Findings and Implications." IEEE Intelligent Systems 39 (5): 78-81. https://doi.org/10.1109/MIS .2024.3441136. · doi:10.1109/mis.2024.3441136
- Vaswani, Ashish, Noam M. Shazeer, Niki Parmar, et al. 2017. "Attention Is All You Need." In Advances in Neural Information Processing Systems 30 (NIPS 2017), edited by I.
Cited by
Alon, L., & Levkovich, I. (2026) · Computers in Human Behavior Artificial Humans · doi:10.1016/j.chbah.2026.100254 · Cited by 10
Modelling the impact of generative artificial intelligence on information behaviour research
Wilson, T. D. (2026) · Journal of Documentation · doi:10.1108/jd-03-2026-0150 · Cited by 4
Equipping a Virtual Reading Promoter for Digital Libraries with Retrieval-Augmented Generation
Li, Z. (2025) · ACM/IEEE Joint Conference on Digital Libraries · doi:10.1109/jcdl67857.2025.00054 · Cited by 3
Complex questions and quality answers: Comparing ChatGPT and Gemini as research collaborators
Ravuri, B., Mardis, M. A., Moja, O., Ayinde, L. A., & Emdad, F. B. (2026) · Journal of the Association for Information Science and Technology · doi:10.1002/asi.70056 · Cited by 3
Self-citationThe case for friction in AI-mediated information seeking and learning
Archambault, S. G., & Kizhakkethil, P. (2026) · Information Research an international electronic journal · doi:10.47989/ir31263104
Metadata
- Authors
- Ravuri, B., & Mardis, M. A.
- ORCID
- Mardis, Marcia A.
- Venue
- Library Trends
- Publisher
- Project MUSE
- Published
- 2025-02
- Volume
- 73
- Issue
- 3
- Pages
- 267-296
- ISSN
- 0024-2594 (ISSN-L)
- 1559-0682
- OpenAlex
- W4410829856
- Cited by
- 9
- References
- 58
- Citations by year
- 2026: 8 · 2025: 1
- FWCI
- 4.62
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by-nc-nd
- www.crossref.org/license (vor, 116 d)
- Locations
- Library trends (publishedVersion, cc-by-nc-nd)
- Topics
- Physical Sciences › Computer Science › Information Systems › Information Retrieval and Search Behavior
- Social Sciences › Library and Information Sciences › Library Science and Information Literacy
- Social Sciences › Library and Information Sciences › Web and Library Services
- Keywords
- Kuhlthau's Information Search Process, information seeking, chatbots, virtual reference services, prompt engineering, large language models, generative AI agents, clarification questions, librarian-patron interaction, information needs, information gathering
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.