Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
By
The Office of Personnel Management released an artificial intelligence chatbot focused on answering questions about federal workforce data in plain English. Now, it wants people to try to break it.
FWD Chat, or Federal Workforce Data Chat, was officially released to the public Wednesday in a testing phase, with the intention of making improvements on the tool as officials get user feedback. Sample prompts invite users to ask how many federal employees were hired last year, what the average salary is by state, and the most common types of employee separations. Outputs are accompanied by suggested follow-up questions and data downloads.
“It’s a beta. Use it, try to break it, and hit the feedback button,” Adam Starr, chief information officer of OPM, said in a LinkedIn post announcing the tool.
In emailed responses to FedScoop, Starr said the chatbot is built on an OpenAI model, leverages a data programming language, and is planned for integration with the Trump administration’s America.gov chatbot. Per Starr, the project took the agency roughly three months to complete.
“Part of a beta is getting the product into people’s hands, seeing how they use it and what they like, and iterating,” Starr said. “We’re especially watching the error messages: whether they’re useful, and which questions people expected FWD Chat to answer. Unanswered questions tell us what to improve next.”
The agency’s plan is to collect feedback over the next month and make improvements in its next releases, he said. The tool itself is updated twice a month. “At OPM, we’re building a culture of rapid releases and iteration, which is still relatively new for government,” Starr wrote.
FWD Chat comes as an expansion of the personnel agency’s efforts to improve the usability of its data interface — which predate the current administration. In January, OPM unveiled a major overhaul of its data products, replacing its antiquated legacy search tool, FedScope, with a new website complete with multiple dashboards and data download features.
Despite the Trump administration’s controversial efforts to downsize the federal workforce and remove certain data, that data website generally won praise from the federal data community as a long overdue change that improves usability. FWD Chat similarly has some early support.
Gabe Menchaca, who focused on workforce policy in the Biden White House and is currently a senior policy analyst with the Niskanen Center’s State Capacity Initiative, told FedScoop the chatbot is “a great step forward for usability” and commended the efforts in the last three to four years at the agency to make data products usable.
Abigail Haddad, an AI engineer and data scientist who was previously part of the Department of Homeland Security’s AI Corps, was recently part of user testing for the chatbot and said there has already been progress since then.
“I could see they’d made improvements after the testing we did,” Haddad said. “I tried one of the same queries again and it did better.”
According to Starr, the beta chatbot is built on OpenAI’s GPT-5.6 and translates each question into a data programming language known as SQL — pronounced “sequel” and short for Structured Query Language — and runs against the agency’s data.
Users can see those underlying queries themselves. Each successful response is accompanied by a “How was this generated” dropdown that identifies the exact SQL query used to get the answer.
Prior to its release, Starr said most of the refinement was the chatbot’s balance between being conversational and providing answers directly from the data.
“FWD Chat is built to report what the data says, not to offer commentary or policy recommendations,” Starr said. “The written description that comes with each answer took trial and error to get right.”
Per Starr, the tool currently relies on most of what’s currently available through Federal Workforce Data and no additional sources.
Specifically, that source is the agency’s Enterprise Human Resources Integration (EHRI) datasets, which is the same data behind the data website’s application programming interface, (API), data downloads and table-building tools.
Not yet included in the chat are data that populate a few interactive charts on the website, such as administrative leave use and federal recruitment, as that data isn’t published as record-level, he said.
There are also limitations to the chatbot. For example, Starr said FWD Chat cannot yet combine from different data tables, though it’s one of the agency’s desired improvements.
Although there was immediate praise for FWD Chat, some workforce policy experts said there is still information needed from the agency.
Bernhard Kluger, an OPM official during the Obama administration, told FedScoop that while he continues to be “impressed and appreciative of the agency’s commitment to transparency and accessibility in data,” he wishes OPM was also committed to evidence-based policy and having a dialogue with the public.
Kluger pointed specifically to the chatbot’s inability to answer a question about how many probationary employees were terminated in 2025 — a category of new and promoted employees leveraged by the Trump administration for mass terminations.
Notably, when the Federal Workforce Data website launched earlier this year, it included new data about payroll and recruitment, but it did not include race and ethnicity data that the Trump administration previously eliminated from FedScope. That remains to be true for the chatbot.
Ultimately, Starr confirmed it is the plan to make the data from Federal Workforce Data available in America.gov — the Trump administration’s recently launched chatbot that’s meant to serve as a front door or directory of federal government services and information.
Starr said the agency intends to make the model context protocol (MCP) server publicly available and will coordinate with the America.gov team for integration.
MCP is an open-source standard designed to improve how artificial intelligence tools parse data by providing them with necessary context. Those servers are already being created elsewhere in government, as officials try to ensure that the public can get authoritative and timely results about federal data when using AI chatbots.
