Most states chose to do this initially to get the new program up and running faster, but then did not recalculate benefits based on actual earnings.
The report comes as 12 million Americans could lose their unemployment benefits next month, when certain relief programs are set to expire, and as congressional Republicans and Democrats and the White House have been unable to reach a deal on another coronavirus relief package.
But the GAO has called on the Labor Department to retroactively fix its figures and make clear that its weekly data no longer offers an accurate count.
Thomas Costa, acting director of the GAO's education, workforce and income-security team, said the office didn't have exact estimates of the magnitude of the inaccuracies in Labor Department data or for the total number of PUA claimants who have been underpaid.
The PUA program, created under the CARES Act signed into law in March, provides payments to Americans who are unemployed for reasons related to the COVID-19 pandemic but who would otherwise not be eligible to receive state unemployment insurance benefits, including gig workers and those who are self-employed.
It's possible the department has both overestimated and underestimated the number, the GAO says.
"Specifically, 27 of the 41 states reported average weekly PUA benefits paid that were within 25 percent of the state's minimum PUA benefit amount; 10 of these states reported average benefits within 10 percent of the minimum", the report says.
"However, the number of claims has not been an accurate estimate of the number of individuals claiming benefits during the pandemic because of backlogs in processing a historic volume of claims, among other data issues".
States have also been inconsistent in reporting claims week to week.
The GAO said the problems in data collection and reporting were making it hard for policymakers to get a reliable picture of what unemployment was doing during the pandemic. For example, Arizona reported no PUA data in the week ending July 4, 2020, after reporting 2.3 million claims in the previous week.
In ordinary times, weekly initial jobless claims serve as a pretty good estimate for layoffs and the number of newly unemployed people, GAO said. In August, the Labor Department changed the methodology it uses to address seasonal fluctuations in the claims data.
Unemployment fraud, which appears to be particularly prevalent in the PUA system, has also skewed claims counts, the GAO said.
For example, in a sample of 20 states, the number of continued claims submitted in the PUA program through June 27 was almost 20 million more than the total number of people who had submitted an initial claim, the GAO found. In some instances, fraud has inflated claims counts and in other cases, it has had the opposite effect.
Figures for the week ending November 21 showed 778,000 new jobless claims being filed, a rise of 30,000 from the week before.
The GAO has recommended that the Department of Labor revise its weekly news releases to clarify that the numbers it reports do not accurately estimate the number of Americans claiming benefits, as well as pursue options to report the correct data.
States collect information to "pay unique individuals claiming UI benefits", GAO said, meaning they could provide the Labor Department with an accurate weekly count.