The Work Requirement Is Really a Paperwork Requirement

The most powerful part of a work requirement is not the work. It is the form.

I understand why the phrase polls well. Most Americans do not think public aid should be detached from any expectation of effort, and I do not think that instinct is crazy. The new federal rules for Medicaid and the Supplemental Nutrition Assistance Program, or SNAP, are framed in exactly that moral language: if an adult can work, train, study, volunteer, or do community service, the public may ask for 80 hours a month before continuing support. Congress built that framework into the 2025 reconciliation law, Public Law 119-21, which created a Medicaid “community engagement” requirement and expanded SNAP work rules after the law was enacted on July 4, 2025, according to the Congressional Research Service¹.

But when I look at the evidence from actual Medicaid and SNAP rollouts, I do not see a labor-market program. I see an eligibility machine. Its main product is not work. Its main product is churn: missed notices, misunderstood exemptions, online portals, monthly reports, procedural denials, and people disappearing from the rolls even when they are working, exempt, sick, unstable, or simply bad at navigating government paperwork.

That distinction matters because the national version is no longer theoretical. The 2025 law requires states to apply Medicaid work requirements to adults covered through Affordable Care Act expansion programs beginning by the end of 2026, or sooner if states choose, and the requirement generally turns on 80 hours a month of work, community service, specified work programs, or half-time education, according to CRS¹. SNAP’s able-bodied-adults-without-dependents rule also uses an 80-hour monthly benchmark, though the exact activities and exemptions differ between the two programs, according to the same CRS comparison¹. This is not a small pilot. It is a mass administrative screen being bolted onto two of the country’s core safety-net programs.

The Arkansas lesson is still the place to start. In 2018, Arkansas became the first state to enforce Medicaid work requirements. Adults ages 30 to 49 in the state’s Medicaid expansion program had to report 80 hours a month of work or another qualifying activity, or show they were exempt, to keep coverage. By late 2018, about 17,000 adults had been removed from Medicaid, and the first major evaluation found that the policy increased the uninsured rate without increasing employment or hours worked, according to a Commonwealth Fund summary of the New England Journal of Medicine study⁴.

The mechanism is the key. If the policy were mainly sorting people who refused to work from people who complied, Arkansas would be awkward but defensible. That is not what the study found. Nearly all targeted beneficiaries had already met the requirement or should have been exempt, and one-third of affected adults with Medicaid or marketplace coverage had not heard anything about the policy, according to the Commonwealth Fund⁴. Harvard’s own account of the research put the compliance figure above 95 percent, finding that the target population was already working, doing qualifying activities, or likely eligible for an exemption, according to the Harvard T.H. Chan School of Public Health⁵.

That is the policy in miniature. The state did not discover a large class of people refusing to work. It discovered that low-income adults often fail complex reporting systems.

SNAP tells the same story from a different angle. The Urban Institute studied nine states that reinstated SNAP time limits for able-bodied adults without dependents after the Great Recession, using administrative data and linked SNAP-Unemployment Insurance records, and found that reinstatement substantially reduced SNAP participation while showing no evidence of increased employment or annual earnings, according to Urban’s 2021 report⁶. The biggest drop usually came between the third and fourth month, exactly when people first became sanctionable under the time-limit rule, according to the same Urban Institute study⁶. That timing looks less like a jobs program succeeding and more like a deadline doing what deadlines do.

A newer paper sharpens the point. A 2026 National Bureau of Economic Research working paper by Jason Cook, Elizabeth Cox, and Chloe East examines SNAP work requirements through procedural denials, using age-based exemptions and waiver periods to isolate administrative burden, and finds that work requirements significantly increase both procedural and overall denials, according to the NBER working paper record⁷. “Procedural denial” is the phrase that should haunt this debate. It means the case can die because the person did not complete the process, not necessarily because the person was substantively ineligible.

The strongest defense of work requirements deserves a fair hearing. It says that Arkansas was bad design, not destiny. States can automate verification, use wage databases, match SNAP and Medicaid records, permit self-attestation when data are incomplete, offer good-cause exceptions, and keep genuinely exempt people out of the reporting maze. SNAP already has general work rules and ABAWD rules with exemptions for people unable to work because of physical or mental limitations, pregnant people, caregivers, people in treatment programs, and others, according to the USDA Food and Nutrition Service⁹. Welfare reform in the 1990s also coincided with major increases in single mothers’ employment, from 60 percent in 1994 to 72 percent in 1999, and random-assignment welfare-to-work demonstrations showed positive employment and earnings effects, according to Brookings¹⁰.

I think that argument proves less than its supporters need. Cash welfare in the 1990s was a different program, aimed at a different population, during a strong labor market, with large changes in the Earned Income Tax Credit, child care, health coverage, and state employment programs all moving at once, as Brookings¹⁰ itself emphasized. Medicaid is health insurance. SNAP is food aid. Taking away a doctor visit or grocery money is not the same as redesigning a cash-assistance program around job placement and earnings supplements. If policymakers want an employment program, they should fund one. A reporting requirement attached to medical coverage is not the same thing.

Georgia shows why the administrative design is not a side issue. Georgia’s Pathways to Coverage program conditions Medicaid coverage on work or other qualifying activities at initial eligibility and through monthly reporting, and the Government Accountability Office found that administrative spending reached $54.2 million out of $80.3 million in total demonstration spending from October 2020 through March 2025, according to the GAO⁸. Most of that administrative spending was federally funded, and GAO said implementation required eligibility-system changes, outreach, and other administrative work, according to the same GAO report⁸. When a coverage program spends more building and running the gate than paying for care, the gate is not incidental. It is the program.

The fiscal math points in the same direction. The Congressional Budget Office estimated that the Medicaid community-engagement provision would reduce federal Medicaid outlays by $325.6 billion from fiscal 2025 through fiscal 2034, while the SNAP work-rule provision would reduce SNAP outlays by $69 billion over the same window, according to CRS¹. KFF’s analysis of the Medicaid provisions says the work requirement is the largest source of Medicaid savings in the enacted package and that the savings will largely stem from coverage losses, according to KFF². That is the quiet tell. If a work requirement mainly moved people into jobs with employer insurance and higher income, the savings story would be a labor-market story. Here, the savings story is a disenrollment story.

Supporters can still say that some savings are legitimate because aid should be reserved for people who meet program conditions. I accept the premise in the abstract. A state that can prove, with reliable linked data, that a nonexempt adult is neither working nor doing any allowed activity has a stronger case for limiting benefits than a state that merely mailed a notice and waited. But that is not the standard the evidence says these systems reliably meet. The recurring pattern is that people lose benefits when the state cannot see them clearly, not when the state has carefully verified that they are refusing to work.

The April 30, 2026, KFF survey of state Medicaid officials makes the next phase look dangerous rather than reassuring. KFF found that seven states were planning either more restrictive verification approaches or early implementation, while states were facing time, cost, and guidance constraints before the January 1, 2027, implementation point described in the survey, according to KFF³. The same survey also found that 18 states planned to use new data sources to automate verification and that many wanted self-attestation for medical frailty, according to KFF³. That is better than indifference. It is not enough to erase the core risk, because automation works best for stable payroll jobs and clear categorical exemptions, while low-wage work often includes variable hours, informal care obligations, self-employment, gig work, seasonal work, illness, and bad records.

So I would judge this policy by a blunt metric: not how many press releases say “work,” but how many eligible households vanish. The burden of proof belongs on the government that adds the paperwork. If states can show independent data proving meaningful employment or earnings gains, low procedural-denial rates, and disenrollments concentrated among verified nonexempt noncompliance, I will revise my view. Until then, the evidence says the work requirement is best understood as a benefit cut administered through bureaucracy.

My prediction is concrete: by the first full renewal cycle after Medicaid enforcement begins in 2027, the states using longer look-back periods, quarterly checks, or limited self-attestation will show higher coverage losses than states relying on automatic matches and broad good-cause protections, without matching gains in employment. Watch the procedural-denial rate. That number will tell us whether this was really about work, or whether the paperwork was the point all along.