Chatbots Open the Door to State Government Access

A Foundation for the Future

Mississippi won a NASCIO award for MISSI, and the chatbot has fielded 3 million inquiries to date. Although AI technology has evolved significantly since MISSI made its debut, Orgeron notes that the need for a strong foundation of rich, accurate content has remained the same. That data, he says, enables MISSI to handle complex, multiagency queries that would confound traditional search tools.

When residents ask about starting a business, for example, MISSI can pull together requirements from the Secretary of State, Department of Revenue and other agencies into a single, coherent response. “The only reason a state portal like this works and is effective is that the data is clean,” Orgeron says. “You can’t train the model on something that is going to render an outdated result.”

MISSI has hunted down information from more than 121,000 sources on behalf of citizens, with 291 of those sources referenced more than 50 times. In the near future, Orgeron predicts, AI-enabled chatbots will gradually be replaced by AI agents that can take action.

“I think the tsunami is really going to be agents,” Orgeron says. “These agents are going to do things on your behalf, and they are eventually going to be the drivers of digital government.”

Improved Citizen Services

Connecticut launched its chatbot Robin in 2019, years before the generative AI boom. Named for the state bird, Robin was built on Salesforce’s machine learning-based chatbot platform, and the solution also leverages Google Analytics and Microsoft Power BI.

Although the state may modernize the solution to take advantage of large language models in the next year, Robin has already brought together information from multiple agencies for citizens while reducing the burden on state employees, says Max Gigle, deputy director of digital product in the state’s Bureau of Information Technology Solutions.

“Before, folks would call the Secretary of State’s office and ask how a business hires somebody in the state,” Gigle says. “That’s not the Secretary of State’s job, and so the help desk would have to guide them to the right agency. Even after an entrepreneur connected with the Department of Labor, there was no guarantee they had checked all the boxes to get served. With the chatbot, the person intuitively understands what they need to do and how to work with state government.”

“The volume of calls didn’t necessarily decrease,” Gigle says. “But employees were able to spend their time answering more nuanced and difficult questions.”

Years after Robin’s debut, entrepreneurs continue to be one of the largest user groups. Gigle notes that large numbers of citizens also use the chatbot for inquiries about job training programs and to apply for benefits such as the Supplemental Nutrition Assistance Program or Women, Infants, and Children support.

Without access to generative AI, Robin cannot make inferences based on contextual analyses of different data sets. But Gigle notes that the tool also doesn’t hallucinate. Unlike some LLMs, Robin will simply state when it is unable to track down a piece of information. That not only improves reliability, he says, but also gives the state valuable feedback.

“We can see where people are asking questions and not getting the information they need,” he says. “That helps us continually improve on the chatbot.”

‘A Dramatic Shift’

Officials were deliberate in their approach to rolling out the Ask Indiana chatbot. The state’s two-year launch process involved running the same 50 questions past one AI model after another to compare their effectiveness, using the questions again for each subsequent model upgrade.

“We would evaluate them based on a rubric,” says Graig Lubsen, director of digital experience and external affairs for the state. “We wanted to make sure it was super accurate. There have been instances where chatbots have been rolled out and provided either bad advice or even illegal advice. We wanted to embrace the technology, but we were really cautious as we approached this.”

Ultimately, OpenAI’s ChatGPT model won out. The chatbot grounds all of its answers in existing state content, and it can pull together information and resources from multiple departments to provide more holistic answers to citizen queries. Before the tool was rolled out to the public, it was tested by IT directors, agency communications professionals, legal counsel and citizen focus groups.

In 2025, the state made Ask Indiana the default search tool on the IN.gov homepage, suggesting queries such as “Where can I apply for Medicaid?” and “How can I renew my driver’s license?”

“We’ve seen a really dramatic shift in how users are interacting with the site,” Lubsen notes. During the seven months after making Ask Indiana the default search tool, the state saw 2.7 million fewer simple searches on its homepage, replaced by 370,000 questions to the chatbot. The results provide links to citations, and Lubsen says the tool has cited more than 1 million pages or documents from across the state’s enterprise content management system.

“There are great opportunities for states to use generative AI,” Lubsen says. “You can’t let it run wild, and it has to be constantly checked. But we have seen the results. Here in Indiana, people are getting to the information they want faster, with fewer questions, and they’re spending less time on the site. I want citizens to find what they want as quickly as possible and then go back to living their lives.”