Trustworthy products for the public good: the USA.Gov AI chatbot experience
GSA—Public Experience Portfolio
Year: 2025
Due to the nature of government work, certain information may be redacted, names changed and/or technical information withheld.
Navigation
Proof of concept that will improve government services
Our customer service agents are experiencing a high volume of calls that can be solved by researching available content on our site. Since we have contractual agreements with the contact center regarding the amount of daily/monthly calls and chats, there is a large percentage of people who are not being serviced. Once we reach the quota of communications, the contact information is disconnected from the site’s menu. The contact center only operates within a limited time frame, Monday through Friday.
The Proof of Concept (POC) would use AI as a chat bot to communicate USAgov information and highlight GSA specific offerings such as GSA Auctions, unclaimed money and financial hardship information.
Our goal was to create a natural language dialogue tool and reduce overwhelming the contact center with easily identifiable information on our website and ensure reliable and vetted information—removing the possibility of inaccuracies or room for interpretation. This would enable the American public to get immediate answers to frequently asked questions without scouring information on webpages or interacting with a contact center agent.
The team consisted of one project manager, two developers and one UX designer (me).
Insights and lessons learned
- In a project as complex as the USA.gov AI Chatbot, parameter chains played a critical role in breaking down information into manageable, visual components. Mapping what could be controlled early in the process helped align stakeholders, reduce ambiguity, and prevent scope creep as the system evolved.
- The complexity of variables made it difficult to establish explicit AI control upfront. Without real user data, we relied on informed assumptions and strategic best-bets—critical tools that shaped early design decisions and guided user testing.
- I approached the chatbot as an AI agent, emphasizing conversational design principles to maintain a natural flow. This included offering users clear ways to revisit previous topics, exit loops, or clarify intent—creating a circular, user-led experience that mimics real human interaction.
- To support a smooth conversational experience, the chatbot also needed fallback responses for unsupported topics, along with multiple guided flows that could interpret user intent through keywords and phrases—ensuring users always had a clear path forward.
- While the long-term vision is for the chatbot to handle a wide range of user needs—including potential direct integration with the contact center—we recognized the importance of designing for phased improvements. Given possible contractual limitations, we planned for a fallback that directs users to the contact page, ensuring a clear handoff even in the worst-case scenario.
- Although I was familiar with the basics of conversational design, this project pushed me into a deep exploration of dialogue and interaction principles. It involved extensive research into conversational design best practices, making the experience both intellectually rewarding and genuinely enjoyable.
- As the project evolved, it became clear that maintaining consistency was critical. Implementing Figma variables and design tokens brought structure to the system, enabling us to scale the design more intentionally and efficiently.
- Although dark mode is not currently implemented on the USA.gov site, it’s a consideration for future updates. Should it be introduced, the chatbot users would benefit from this feature, enhancing accessibility and user experience.
