That’s a key finding from “Uncovering the Results of AI Chatbot Use In the Public Sector: Evidence from U.S. State Governments,” the latest study from Dr. Mila Gascó-Hernandez, research director at the University at Albany — State University of New York’s Center for Technology in Government, and Tzuhao Chen, a University at Albany doctoral student.
Their initial studyfocused broadly on chatbot adoption and implementation, but this new paper narrows in on outcomes and effectiveness to explore the impact of chatbots on workload, citizen satisfaction and communication efficiency.
“Through our research, we wanted to focus on the impact chatbots have had on government agencies and the people they serve, from the perspective of public employees,” Gascó-Hernandez said. “To do this, we developed a framework based on research in innovation and digital transformation to evaluate these impacts.”
Gascó-Hernandez and Chen’s findings suggest that by combining quantitative and qualitative feedback, engaging in interagency collaboration, and adopting evolving AI technologies responsibly, government agencies can leverage chatbots to create more responsive, inclusive and effective public services.
Their research revealed four main outcomes: optimized workloads, improved communication, shorter wait times, and an expanded range of use cases.
While some agencies reported reduced workloads due to automated responses handling common queries, others found demands for oversight from bots had increased. Using built-in analytics tools did enhance communications, their research showed, by helping agencies identify frequently asked questions and populating clearer responses.
By offering a self-service option, chatbots have significantly reduced wait times. Building on this success, agencies are now broadening how they use chatbots to services like application processing, connecting citizens to live agents for more comprehensive support.
Yet, even with this forward progression, Chen noticed a persistent gap: "Very few chatbots have the feature to collect feedback or measure user satisfaction — and that’s something agencies might need to focus on moving forward.”
Agencies discussed their intentions to bridge this gap, as Chen and Gascó-Hernandez examined the pros and cons of quantitative and qualitative chatbot feedback.
“Both types of feedback came up in our interviews,” Chen said. “Quantitative feedback — such as ratings, resolution rates, frequently asked questions, peak usage times, or the number of unanswered questions — provides a quick snapshot of general trends in citizen needs and satisfaction. However, it doesn’t explain the ‘why’ behind the numbers, particularly what citizens expect from the chatbot.”
While qualitative feedback provides valuable insights into user experiences and motivations, it’s harder to gather since citizens rarely complete surveys after using chatbots, Gascó-Hernandez noted. In contrast, quantitative data is readily accessible through analytics tools.
Study findings also highlighted an evolving perspective on the need for a figurative “human touch.” Existing studies suggested, generally, that citizens and customers still tend to prefer human interaction over chatbots, Gascó-Hernandez said.
“That said, our interviewees noted that a subset of citizens — particularly those comfortable with technology — often prefer not to engage with a person if a chatbot can effectively address their inquiries,” she said.
“That’s why we should embrace chatbots more. Our users want quick, easy access to information, without having to search endlessly for it,” said one unidentified interviewee featured in the study.
This perspective is widely accepted, the study revealed, among users with prior experience using chatbots in their personal lives. “The families we serve interact with companies that use chatbots for customer service,” an agency representative interviewed in the study said. “I believe they bring those expectations with them when they interact with us. When chatbots are done well, they become part of that positive experience.”
Agency staffers interviewed were, however, divided on whether chatbots will fully replace traditional FAQs. “Chatbots are more interactive and behavioral, whereas FAQs feel like a descriptive essay,” a study participant said.
This doesn’t mean traditional communication methods are no longer needed; several interviewees said they have different preferences for how they like to access information. This is why even with expanded chatbot use, Chen said, it remains essential to maintain multiple communication channels to accommodate diverse needs.
Chatbot implementations still face familiar hurdles like data selection and accessibility. Research revealed an obvious but important fact: understanding the nuances of residents’ language preferences requires ongoing updates to a chatbot’s knowledge base.
While chatbots today are limited to answering informational queries, future models might begin incorporating data from other agency databases, which Chen said would undoubtedly lead to more comprehensive privacy protections, particularly around personally identifiable information.
Chatbot governance is an evolving field, too, with unlimited potential if guided by clear-cut standards. Gasco Hernandez said she envisions a time when routine performance reviews of chatbot conversation logs will become standard, and generative AI and natural language processing will help identify trends and adapt chatbot responses using “ethical standards guiding implementation.”
