Share Your Experience: Using AI to Access Federal Statistics
The National Center for Science and Engineering Statistics (NCSES) is working with NORC at the University of Chicago and the Georgetown University Massive Data Institute on a project to understand how large language models (LLMs) handle questions about federal statistical data. Your feedback will help us evaluate the readiness of federal data for LLMs and improve how these tools serve data users like you.

Why your input matters:
Your input will help us evaluate LLM performance and identify ways to make these tools more accurate and useful for federal data users.

How you can help:
Please share your experiences by answering any of these questions:
1. What kinds of prompts do you use with LLMs to ask questions about federal data? Tell us about what goes into crafting your prompts for LLMs.
2. Can you share examples of prompts you've tried? Did you find the responses accurate or helpful? If you have any specific prompts related to Census Bureau or Bureau of Economic Analysis data, please share those.
3. Have you run into any challenges when using LLMs for federal data questions?

OMB Control Number: 0690-0030, expiration date 07/26/2026

------------------------------
Beth Jarosz
------------------------------

Thanks Beth, do you want federal data users to post examples here? Attached is a simple example of a query (prompts highlighted in yellow) that shows the challenges in using ChatGPT (V. 5.2) to access ACS data. I often find that AI will confidently and cheerfully provide an answer, and then when I note that it's incorrect it will reply with "You're exactly right!" and can often correct it's errors. But it still can't deliver full tables from data.census.gov, as far as I know.

Share Your Experience: Using AI to Access Federal Statistics
The National Center for Science and Engineering Statistics (NCSES) is working with NORC at the University of Chicago and the Georgetown University Massive Data Institute on a project to understand how large language models (LLMs) handle questions about federal statistical data. Your feedback will help us evaluate the readiness of federal data for LLMs and improve how these tools serve data users like you.

Why your input matters:
Your input will help us evaluate LLM performance and identify ways to make these tools more accurate and useful for federal data users.

How you can help:
Please share your experiences by answering any of these questions:
1. What kinds of prompts do you use with LLMs to ask questions about federal data? Tell us about what goes into crafting your prompts for LLMs.
2. Can you share examples of prompts you've tried? Did you find the responses accurate or helpful? If you have any specific prompts related to Census Bureau or Bureau of Economic Analysis data, please share those.
3. Have you run into any challenges when using LLMs for federal data questions?

OMB Control Number: 0690-0030, expiration date 07/26/2026

------------------------------
Beth Jarosz
------------------------------