AI Use on the Rise

Research from Fractl offers objective evidence that AI tools are prolific and quickly becoming must-have support for all kinds of tasks. It indicates that one in eight gen AI users treat their chatbots as colleagues; 60% use these tools as work assistants. Even more striking and, perhaps, concerning, 54% have had extended conversations lasting more than an hour with AI systems, and 29% have shared secrets or deeply personal information with their chatbot.

As AI tools rapidly transform reference services, the question isn’t whether to adopt automation—it’s how to implement it wisely while preserving the irreplaceable human expertise that defines quality information work. 

The Transformation of Reference Services

AI excels at tasks that were once time-intensive manual work for librarians. “Artificial intelligence has brought reference services from simple indexing to synthesis where the tool reads millions of documents to give a singular answer,” says Gor Gasparyan, co-founder and CEO of Passionate Agency. His team finds that AI successfully handles approximately 80% of tasks previously done by human agents manually searching databases. 

Edward Tian, founder and CEO of GPTZero, agrees that AI transforms delivery while keeping human intervention central. “AI excels with the highest volume/lowest judgment-type tasks, including fast retrieval of information; clustering related information; summarizing large, dense documents, and answering repetitive requests made after hours,” he says.

Chris Kirksey, founder and CEO of Direction.com, sees AI redefining which services will continue to exist. “AI is automating routine functions such as database searches, content summaries, and providing citations at a speed that has never been possible before,” he says.

However, the consensus among experts is clear: AI handles scale brilliantly but struggles with nuance and judgment.

What AI Does Well—and What It Doesn’t

Fred Freeman, co-founder of GlobeScribe, puts it succinctly: “AI is exceptionally good at scale: processing huge volumes of information, identifying patterns, and delivering fast, consistent results.” However, he stresses, what remains irreplaceably human is reasoning. “An algorithm can present options, but only a person can understand what matters in a specific context, what is appropriate for a particular reader, and what carries cultural or emotional nuance,” Freeman says. “Taste, empathy, and cultural awareness cannot be automated.” 

Denise Agosto, a professor at Rutgers University’s School of Communication and Information, urges a fundamental reframing in terms of how AI is approached. “We want to avoid thinking about tasks as appropriate for either humans or AI, separating AI systems from the humans who design and use them,” she says. “Instead, we want to think about humans using AI to solve problems, explore questions, and deepen our thinking. AI doesn’t operate without human guidance.” 

In its current state, gen AI can assist with writing tasks, help generate questions to explore, serve as copy-editing assistants, and handle mundane writing tasks. “AI tools don’t just spit out useful content, though,” Agosto cautions. “For effective AI use, humans must review, reflect on, add to, question, and personalize the content they produce.”

Bruce Bachenheimer, a clinical professor of management at Pace University, highlights a critical limitation of using gen AI tools. “People don’t know what they don’t know. So performing an internet search or entering an AI prompt may be helpful but of limited value,” he says. “Intelligence comes down to the ability to learn, not simply accumulating information, facts, and figures. It’s what one is able to do with that information—through critical thinking, logical reasoning, and strategic analysis—that is the challenge and where human skills remain important.”

Finding the Right Balance: Practical Strategies and Metrics That Matter

Librarians have an opportunity to serve as the critical filter that AI cannot provide. As Freeman suggests, use AI for scale and humans for meaning. “Let AI handle the vast and repetitive processing such as classification and thematic analysis, while librarians concentrate on the nuanced work of understanding users, shaping collections, and guiding people to the information that matters most.” 

Tian takes a layered approach. “Allow AI to perform the first-pass retrieval of information, then employ humans to verify the information retrieved, provide context to the verified information, and make final recommendations.” One research group he assisted reduced initial query time by 60% using AI triage while librarians validated citations and provided expert commentary, maintaining high accuracy.

So how can organizations monitor their AI use to ensure they have the right balance between AI and human tasks? Kirksey recommends tracking accuracy, user satisfaction, turnaround time, and frequency of human intervention. “Increases in errors or user complaints signal adjustments are needed,” he says. “Identifying patterns in AI weaknesses helps determine how to divide tasks between AI and humans.” 

Another opportunity to monitor effectiveness, suggests Yad Senapathy, founder and CEO of Project Management Training Institute, is monitoring “the effectiveness of the system, the frequency at which errors are escalated to higher levels of support, and the impact that those errors have on end users. When employees continually spend increasing amounts of time correcting errors caused by an AI system, or users choose to avoid using the system to answer complex questions, then it is likely the system requires adjustment.”

From both a policy and practice standpoint, Agosto stresses that organizations must discuss appropriate use and set clear AI use policies. “We can’t assume that students, employees, or others will know what types of AI use are appropriate and inappropriate without setting clear use guidelines and communicating them to all members of the organization,” she says.

As AI tools continue to evolve, information professionals need to stay informed, flexible, and comfortable with ambiguity. Things are changing quickly in this space. Understanding the tools and their capabilities and limitations, as well as having policies, practices, and metrics in place to continually monitor the accuracy of output, are critical.

Creating workflows that leverage AI’s speed and scale while ensuring human judgment and oversight will yield the best outputs of both.