How AI Is Changing Social Security Customer Service

AI moves in as staff shrinks

The chatbot was rolled out as the SSA moved to shed more than 12 percent of its staff amid the Trump administration’s push to shrink the federal workforce. The agency currently has about 51,400 workers, down from about 57,000 before the reductions.

Jack Smalligan, a senior policy fellow at the Urban Institute, a nonprofit research organization, says Social Security’s customer service issues require both human and technological solutions.

“SSA is struggling enough right now with its service levels that it can use the AI tools to improve service and not reduce staff,” says Smalligan, whose research focuses on retirement and disability issues.

The SSA says it is successfully speeding up phone service. The “average speed of answer” for 800-number calls, which spiked to 30 minutes in January, was down to seven minutes in September, according to agency data.

That metric largely reflects the time it takes for the caller to choose between self-service or speaking with a live agent, says Kathleen Romig, director of Social Security and disability policy at the nonprofit Center on Budget and Policy Priorities, but it does not include the time customers wait for a callback, as about three-quarters of callers do.

The SSA no longer discloses all phone wait times on its public performance page, but it did report them recently to Sen. Elizabeth Warren’s office in response to an inquiry from the Massachusetts Democrat on customer service.

According to that report, sent in September, the average hold time for a caller over the previous 11 months was nearly an hour, and the wait for a callback was nearly two hours. (Those averages declined sharply over the summer, to about 22 minutes on hold and 1 hour and 4 minutes for a callback in August. That coincides with the SSA’s shift of about 1,000 field office workers to handling 800-number calls, Romig notes.)

“Bisignano has been touting improvements in average speed to answer, claiming single-digit wait times,” Romig says. “The average speed of answer doesn’t reflect a typical [caller] experience.”

AARP called the 800 number about 25 times in recent weeks to test what kind of information AI was providing. The results, although not scientific, suggest that the bot has trouble in some areas.

On some calls, for example, it was unable to answer basic questions about spousal and survivor benefits, Eskovitz says. On others, it failed to distinguish between retirement benefits and Supplemental Security Income (SSI), a safety-net benefit administered by the SSA that differs significantly from Social Security in eligibility and payment policies.

Here’s an example of one exchange:

AARP: How do I estimate my Social Security benefit?

SSA BOT: We are here to help you by providing the critical services you need. We want to help you as quickly and safely as possible. Our website is the best way for most people to do business with us. Visit us at www.ssa.gov. If you’re finished, feel free to hang up. Otherwise, how can I help you today?

AARP: How do I estimate my Social Security benefit?

SSA BOT: The easiest and most convenient way to apply for benefits or get benefit estimates is by using our online application at ssa.gov. Please note, your retirement application can only be submitted up to four months before you want to start receiving your benefits. If you are unable to apply online, or are applying for survivors’ or death benefits, say “agent.” If you’re finished, feel free to hang up. Otherwise, how can I help you today?

Clients of Community Legal Services of Philadelphia, which helps low-income residents and people with disabilities access Social Security benefits, have had similar experiences, says Kristen Dama, a managing attorney with the organization.

She recalls one client who called the SSA to find out if a request for reconsideration of a disability benefits overpayment had been processed; the bot responded with information about a federal railroad retirement program and domestic partnerships.

Dama says the bot has improved dramatically since its rollout. Because it is powered by generative AI, it learns as it goes. Still, it’s “part of the litany of complaints that we hear about customer service,” she says. Some clients don’t know what to ask and get frustrated, she says; others find themselves cut off when the bot decides their question has been answered and disconnects.

Many of the legal-aid group’s clients have mental health issues or cognitive problems and “really need a person who’s trained on the phone, who can really kind of help them get to the point, help them articulate what they need,” Dama says. “An AI chatbot can’t do that.”

Progress on disability claims?

The SSA spokesperson says that in the coming year, the agency will continue its “digital-first approach to further enhance customer service” by refining the phone bot’s performance and deploying other AI tools.

“SSA will also introduce new service features to assist callers while they wait, expand language support to reach more customers and automate processes such as mobile phone enrollment,” the spokesperson says. These initiatives “reflect SSA’s commitment to using technology to deliver faster, more accessible and customer-focused service.”

One area in which the agency was already testing and using AI is processing disability claims — for example, using a different AI tool to organize applicants’ voluminous medical records. Smalligan says AI could be very beneficial in this arena.

“Individuals applying for disability benefits can come to SSA with 1,000 pages or more of medical evidence, and so the process of having SSA adjudicate that claim can be very time-consuming and very complex,” he says. “AI has the potential to kind of organize and make sense of that history of medical evidence,” identifying gaps in the record and expediting decisions.

That can be a “win-win,” he adds, if it speeds up disability claims for applicants and reduces the administrative burden on SSA staff.

At the same time, he says, the SSA needs to be transparent about how it is using AI and aware of its pitfalls — including generating inaccurate or misleading information, known as “hallucinations.”

Dama says AI could be used to “green-light” benefit awards for people who “so clearly meet the disability standard that nobody needs to spend a lot of time developing the case.” But she worries about “a computer algorithm … making some sort of black box determination that someone isn’t disabled without having a human involved.”

In a September 2025 AI strategy document, the SSA lists expanding the use of AI in providing frontline services and processing claims among its priorities but says it will “continually reinforce humans at the center of AI development and AI-augmented processes.”