CensusBot
Tapping AI to help reporters use high-value, high-complexity data
With over 1000 tables, almost 28,000 variables, and hundreds of thousands of geographies, American Community Survey data can be overwhelming. Reporters struggle with understanding geographic summary levels, knowing what tables are available, and understanding the definitions of technical terminology. This project aims to build a conversational AI tool that taps general Census documentation as well as our knowledge from building Census Reporter to help users more casually ask for guidance and get plain language answers that help them find data from the American Community Survey, whether through Census Reporter or other means.
In this Knight Lab Studio project, students will build CensusBot, a conversational AI assistant designed to help journalists navigate the overwhelming complexity of American Community Survey data. Through user research, AI platform evaluation, and iterative prototyping, the team will create a tool that translates technical Census terminology and structures into plain language guidance. The resulting prototype will demonstrate how AI can bridge the gap between high-value demographic data and the reporters who need it, potentially serving as a model for making other complex datasets more accessible.
Faculty and Staff Leads
Project Details
2026 Spring
Important Questions
- How do journalists currently search for and use Census/ACS data, and where do they encounter barriers?
- What kinds of questions do reporters ask when looking for demographic data, and how can we map those to ACS tables and variables?
- What AI platforms and approaches are best suited for helping users navigate complex, structured data through natural language? What role might emerging integration patterns like Model Context Protocol or LLM skills play?
- How do we balance conversational flexibility with data accuracy and precision?
- How might a chat interface integrate with the existing Census Reporter experience?
Sample Milestones
- Weeks 1-3: Conduct user research with reporters who use Census data to document pain points, common questions, and barriers. Research and evaluate AI platforms and conversational interface approaches, including awareness of emerging integration patterns like MCP and LLM skills. Inventory existing Census documentation and Census Reporter institutional knowledge.
- Weeks 4-6: Design conversational interface patterns and knowledge base structure. Develop and test AI prompting approaches using ACS documentation and Census Reporter insights. Prototype with sample queries and refine based on results.
- Weeks 7-9: Build functional prototype integrating selected AI platform with Census knowledge base. Test with journalists and iterate based on feedback. Develop integration concepts showing how CensusBot would work with Census Reporter.
- Week 10: Finalize working prototype and comprehensive documentation.
Outcome
Students will produce a working conversational AI prototype that helps journalists navigate American Community Survey data. They will document user research findings, catalog the knowledge sources that can help reporters work better, and describe their technical findings for potential future development. The project will demonstrate how AI can make high-complexity, high-value datasets more accessible to journalists without requiring deep technical expertise.