What is OPM’s FWD Chat beta meant to do with federal workforce data?

How is FWD Chat meant to handle a question?

The goal is to let people ask about workforce statistics in ordinary language rather than construct a database query themselves. Starr told FedScoop that FWD Chat converts questions into SQL, a language used to request information from databases, and runs the resulting queries against OPM data. That is the agency official’s description of the process, not a result independently confirmed by a technical assessment.

OPM’s release notes say users can inspect the dataset, measure, filters and time period used for an answer. They also describe follow-up questions and options to open results in Table Builder or export them as a CSV file. FedScoop reported that Starr said successful answers expose the associated SQL query.

Those features can show what information and conditions went into a result. They do not, on their own, prove that the chatbot interpreted a question correctly, generated an appropriate query or returned a complete answer. OPM warns users that responses can contain errors or vary across similar questions, and recommends checking important figures in Table Builder or the relevant Workforce Insights page.

Starr told FedScoop that the beta uses an OpenAI model and identified it as GPT-5.6. OPM’s public chatbot page and release notes do not name a model. The account of the model and query process comes from Starr, and the reported material provides no independent evaluation of the system’s accuracy.

What workforce information can it reach?

Federal Workforce Data, or FWD, is OPM’s public platform for exploring and downloading information about the federal civilian workforce. OPM describes it as a successor to the older FedScope platform. Much of its data comes from the agency’s Enterprise Human Resources Integration program, or EHRI, which collects employee and payroll information from federal agencies.

The underlying datasets cover different kinds of records. OPM describes Status as a monthly workforce snapshot, Dynamics as personnel actions during a reporting period, and Payroll as pay-period records that can include pay and time information. OPM says the broader FWD data are updated monthly; the chatbot page shows data as of August 2026.

Starr told FedScoop that FWD Chat draws on most of the data then available through FWD, particularly EHRI information, but cannot yet combine different tables. That constraint means a request requiring information from separate datasets may not be answerable through the beta, even if relevant material appears elsewhere on the site.

Some figures displayed in FWD charts also were not yet available through the chatbot, Starr said. He cited administrative-leave use and federal recruitment, explaining that those data had not been published at record level. OPM’s guide also identifies gaps in FWD’s demographic information, including disability status, gender, race or national origin, and ethnicity; FedScoop reported that race and ethnicity data are absent from the chatbot.

The coverage limits appeared in one example in FedScoop’s report: the chatbot could not answer how many probationary employees were terminated in 2025. The report supplied no verified termination count. OPM’s chatbot page also advises users to confirm significant figures in other FWD tools, rather than treating the conversational answer as definitive.

Is the America.gov connection in place?

No completed connection is established. Starr told FedScoop that OPM plans to make an MCP server publicly available and coordinate with the America.gov team so workforce data can be accessed through that administration chatbot. This is a reported plan, not confirmation that America.gov can currently query FWD data.

The proposal would add a route to the workforce data beyond OPM’s own beta, but the report does not provide a launch date, technical design or list of questions the proposed connection could handle. It also does not say that the two chatbots already share data or that the planned server is public.

Starr said OPM planned to collect feedback over the following month and use it to guide later releases. He also said the chatbot itself is updated twice a month, while OPM describes the underlying workforce data as updated monthly. The public information does not explain how those schedules work together or what changes resulted from the feedback plan.

What has been established about answer quality?

OPM’s public materials establish that FWD Chat is a beta, describe ways to inspect results and warn that answers can be wrong or inconsistent. They do not identify the model or provide a measured accuracy rate. The technical details about GPT-5.6, SQL and the inability to combine tables are Starr’s account to FedScoop, not independently tested findings.

The report includes individual reactions, not a systematic evaluation. Gabe Menchaca, a senior policy analyst with the Niskanen Center’s State Capacity Initiative, described the tool as “a great step forward for usability,” FedScoop reported. AI engineer and data scientist Abigail Haddad, who took part in user testing, told the outlet, “I tried one of the same queries again and it did better.” Her account describes one repeated query, not performance across the chatbot’s questions.

Bernhard Kluger, a former OPM official, praised the agency’s work on transparency and access, FedScoop reported, while calling for more emphasis on evidence-based policy and public discussion. The unanswered question about probationary employees illustrates a practical limit: users may not know from a conversational interface alone whether a missing result reflects the chatbot, the available records or the way the data are organized.

Starr said FWD Chat is built to report what the data says rather than offer policy recommendations. Whether it does that reliably remains untested in the material reported. Until OPM publishes an evaluation or otherwise establishes how consistently the beta handles questions, the agency’s own advice is the clearest guide: check important results against its underlying workforce tools.