I push teachers to bring AI into their work, so I pay attention when a study measures what students are already doing on their own. Lund and colleagues (2026) surveyed 236 U.S. university students in late summer 2025 about how they look for information, comparing AI chatbots, search engines, and the academic library.
Before the findings, a caveat the authors put up front. The sample leans heavily toward graduate students, computing and information sciences, and international students, so the adoption rates probably run ahead of a representative undergraduate population. I’ll read the numbers with that in mind.
The headline reassures anyone worried that AI has swept everything aside. Google is still where most academic searches begin. Lund et al. found 45.5 percent of students start at a search engine, 27.9 percent at the library homepage, and 26.6 percent at an AI chatbot. AI is a real player in the mix, but the field hasn’t reorganized itself around the chatbot.
Who Reaches for AI First
Age splits the picture cleanly. Among students 18 to 25, about half start at Google, a third at AI, and only a sixth at the library. The 31-and-older group flips it, with most starting at the library and AI tied with Google well behind. For older students the library is still the front door. Younger ones have already moved AI ahead of it. Lund et al. also note an unusual split among men, who are likelier than women both to never touch AI and to use it daily, while women settle at steady, moderate use.
The most striking finding concerns international students. Lund et al. report they use Google at the same rate as domestic students but lean on the library far less and on AI far more, roughly 17 monthly AI uses against 8 for domestic peers. The extremes are stark. About 58 percent of domestic students never use AI alone for academic research, against only 8 percent of international students.
The authors call this substitution, with AI taking the role the library plays for everyone else. As they write, “international students are major consumers of information from AI, often supplanting information that might be gathered from sources like the university” (p. 13).
That substitution has a likely cause worth naming. Lund et al. connect it to the language barriers and library anxiety earlier work documented (Lu and Adkins, 2012; Peters, 2010, cited in Lund et al., 2026). A general chatbot is easier to approach than a catalog that assumes fluency in U.S. academic conventions. This echoes what Hysaj and colleagues (2025) found about multicultural students leaning on generative AI to navigate unfamiliar academic writing expectations. The students reaching for AI hardest may be the ones the library is serving least well.
A Tool Students Use, Sometimes Without Trusting It
Two numbers pull against each other. About 19.6 percent of respondents report no confidence at all in the completeness of AI answers, yet only 15.3 percent never use AI for academic work. Some students lean on a tool they openly distrust. At the same time, 58.7 percent are somewhat or very confident in AI completeness, a higher trust level than they give conference proceedings.
The strongest statistical result is a correlation between how often students use AI and how satisfied they are with it, holding across seven information types. No such pattern appears for search engines, where satisfaction stays flat no matter the frequency. Lund et al. are careful, noting they can’t tell which way the arrow points.
Their reading is that “the more an individual uses AI, the more they are likely to express strong perceptions of usefulness and satisfaction with AI tools” (p. 16). I’d add the caution they imply. Students who already like AI may simply use it more, so familiarity and preference are tangled together.
Most students don’t work from one source. Around 40 percent start with a library database or search engine and turn to AI to fill the gaps, and a smaller group does the reverse. The pure-AI-only group is the smallest of all. Lund et al. tie this to Bates’s berrypicking model (cited in Lund et al., 2026), where searching is iterative and pulls from whatever source fits the next step. That matches how I watch capable students actually work.
What This Means for Libraries
The operational message is the one libraries should hear. If international students route around the catalog because AI is easier to reach, that is an engagement problem the library can’t afford to wave off. Lund et al.’s suggestion is concrete and, to me, the most useful one in the paper.
Build AI into library services directly. An AI layer using retrieval-augmented generation (RAG), drawing on the library’s licensed collections, could give students AI-style answers grounded in vetted scholarship, lowering the friction that now sends them to a general chatbot.
I find that framing is practical. The library works best here as the bridge between students and responsible AI access, not as a rival to it. That aligns with how McCrary (2026) reframes the future of libraries around AI-mediated research, and with the case LaFlamme (2025) makes for scaffolding AI literacy directly into library instruction so students aren’t left to figure it out alone.
The most candid part of the paper is its restraint. Lund et al. don’t claim AI has won. Their conclusion is that “AI is generally found not to be an outright replacement for other types of information sources, but one of many tools that students use in seeking information” (p. 16).
The sample skews graduate, computing, and international, the data is self-report, and the satisfaction link can’t be pinned to a direction. They name all of it, and they point to interviews and a hands-on task study as next steps. That restraint is what makes the paper credible. It catches a three-way habit forming in real time, and it hands libraries a role in shaping it before the habit hardens.
References
- Hysaj, A., Dean, B. A., & Freeman, M. (2025). Exploring the purposes and uses of generative artificial intelligence tools in academic writing for multicultural students. _Higher Education Research & Development_, 44(7), 1686–1700. https://doi.org/10.1080/07294360.2025.2488862
- LaFlamme, K. A. (2025). Scaffolding AI literacy: An instructional model for academic librarianship. The Journal of Academic Librarianship, 51(3), 103041. https://doi.org/10.1016/j.acalib.2025.103041
- Lund, B. D., Teel, Z. A., Mohammed, Y., Jagathpally, A., & Wang, T. (2026). Artificial intelligence (AI) and information seeking: A comparative exploration of AI chatbots, search engines, and library resources as information sources among university students. Journal of Librarianship and Information Science. https://doi.org/10.1177/09610006261438484