OPM launches workforce-data chatbot that shows the queries behind its answers

The Office of Personnel Management has launched a public beta chatbot that lets people ask about the federal workforce in plain English. FWD Chat answers questions about hiring, departures and employee demographics—and shows the database query behind successful answers. That easier entry point comes with limits on which data the chatbot can reach.

An answer with a trail back to the data

Released publicly Wednesday, FWD Chat sits on OPM’s Federal Workforce Data website. Suggested questions include how many federal employees were hired last year, average salaries by state and the most common types of employee departures. Answers come with suggested follow-up questions and data downloads.

Demographic questions can cover age, veteran status and education level, Federal News Network reported. The chat interface builds on a January overhaul that replaced OPM’s older FedScope search tool with a website offering dashboards and downloadable data. The new release adds conversation to that existing set of tools.

OPM chief information officer Adam Starr told FedScoop that the chatbot uses OpenAI’s GPT-5.6. It translates questions into Structured Query Language, or SQL—the language used to search databases—and runs those queries against agency data. A “How was this generated” dropdown reveals the exact query behind each successful response.

Starr said much of the work before release involved balancing a conversational style with answers drawn directly from the data. The written explanations required trial and error. OPM’s intended boundary is narrow: describing workforce records, rather than interpreting what policy should follow from them.

FWD Chat is built to report what the data says, not to offer commentary or policy recommendations.

Adam Starr, OPM chief information officer, in emailed responses to FedScoop

Easier access is not fuller coverage

The chatbot draws only from OPM’s Enterprise Human Resources Integration datasets, Starr said. Those are also the records behind the website’s data downloads, table-building tools and developer interface. It does not bring in outside sources, and some information visible elsewhere on the website remains beyond its reach.

• The chatbot cannot yet combine information from different data tables, a capability Starr said OPM wants to add.
• Administrative leave use and federal recruitment figures shown in some interactive charts are not included because they are not published as individual records.
• Race and ethnicity data are absent from the chatbot, as they are from the workforce website, FedScoop reported.

Early reactions reflect that divide. Gabe Menchaca, a Niskanen Center senior policy analyst who worked on workforce policy in the Biden White House, praised the chatbot’s usability. Former OPM official Bernhard Kluger welcomed the agency’s accessibility efforts but wanted a stronger commitment to evidence-based policy and public dialogue.

Kluger pointed to a specific unanswered question: how many probationary employees were terminated in 2025. His criticism concerned the information people could obtain, not simply how easy the interface was to use.

A beta built around unanswered questions

Starr said the project took roughly three months. OPM plans to collect feedback over the next month and use it in subsequent releases; the tool is updated twice a month. Officials are especially watching whether error messages help users and which questions people expected the chatbot to answer.

AI engineer and data scientist Abigail Haddad, who participated in user testing, told FedScoop she had already seen improvement. After trying one of the same queries again, she said it performed better. That is an early tester’s observation, rather than a broad measure of the chatbot’s accuracy.

OPM also intends to connect its workforce data to America.gov, the administration’s government-information chatbot. Starr said the agency plans to publish a Model Context Protocol server—a standardized connection that gives AI tools access to data—and coordinate integration with the America.gov team. That connection remains a plan, not a completed feature of this beta.