The BLS Slashed Nearly 900,000 Jobs From Its Books: What Happened and Why It Matters

The Bureau of Labor Statistics released its Preliminary Benchmark Revision for the Current Employment Statistics (CES) survey on September 9, 2025, estimating that total nonfarm employment for March 2025 would be revised down by approximately 898,000, or 0.6 percent, relative to the sample-based, seasonally adjusted published estimate. The release was posted at 10:00 a.m. Eastern Time.

A benchmark revision, as defined by the BLS, is the difference between the actual count of employment for March and its corresponding sample-based estimate after removing the effect of the estimate's error. The CES survey uses a sample of business establishments to produce monthly payroll estimates. Each year, the BLS reconciles those estimates against the Quarterly Census of Employment and Wages (QCEW), which draws on state unemployment insurance tax records covering virtually all US employers. Think of it as a regular audit: the sample-based estimates are the initial, fast read; the QCEW is the gold-standard headcount that arrives later.

The preliminary revision, published in late summer, offers an early read on the direction and magnitude of that reconciliation before final figures are issued the following February.

Reuters reported that the preliminary revision estimate suggested a downward adjustment of around 550,000 nonfarm jobs, or about 45,000 per month, when the initial signal was published on September 9, 2025. The subsequently reported figure from the BLS national benchmark article, a downward revision of 898,000 jobs, indicates a larger adjustment than the earlier Reuters estimate implied.

When the final benchmark revision for the 2025 cycle was issued alongside the January 2026 Employment Situation news release in February 2026, the BLS reduced US employment growth through March 2025 by 862,000 jobs. That final figure was smaller than the 911,000-job reduction the BLS had estimated in its preliminary August revision, as reported by Reuters. The gap between the preliminary and final figures narrowed by roughly 49,000 jobs.

The 10-year track record for annual benchmark revisions to CES data shows an absolute average of 0.2 percent of total nonfarm employment. The 0.6 percent revision associated with the March 2025 benchmark exceeds that historical average. Whether the final 862,000-job reduction, which translates to approximately 0.6 percent of total nonfarm employment, holds as an outlier or becomes part of a trend toward larger revisions will depend on subsequent cycles.

The BLS has scheduled the Preliminary Benchmark revision for March 2026 for release on August 28, 2026, at 10:00 a.m. Eastern Time. That release will provide the first preliminary signal on whether the sample-based estimates for the March 2026 reference month are tracking the actual universe count or diverging as they did in the prior cycle.

In the most recent Employment Situation report, the BLS revised the change in total nonfarm payroll employment for May 2026 down by 66,000, from an initially reported +129,000 to +63,000. The June 2026 change was revised down by 37,000. These monthly revisions are part of the standard sample-based estimation process and are distinct from the annual benchmark revision, which applies a full-count adjustment to the March reference month.

The broader context here is one of serial downside revisions layered on top of a benchmark cycle that already painted a weaker labor market than the CES sample initially indicated. The preliminary revision of roughly 898,000 was itself larger than what Reuters' earlier reporting flagged, and the final figure of 862,000, while smaller than the preliminary 911,000 estimate, still represents a material downward adjustment. Combined with the consecutive monthly downward revisions in May and June 2026, the pattern is consistent with a CES sample that has been overstating payroll growth relative to the administrative universe data.

For market participants, the key question is whether the August 28, 2026 preliminary benchmark release will show another large downward adjustment. If it does, the reliability of the monthly CES prints as a real-time labor market indicator will face further scrutiny. The Fed's policy stance depends heavily on the employment leg of its dual mandate, and persistent downside benchmark revisions complicate the picture by suggesting that the headline payroll numbers investors react to in real time may be overstating labor market strength by a non-trivial margin.

Final benchmark revisions to state establishment employment data follow a separate timeline. The BLS issues those with the publication of the January State Employment and Unemployment Summary news release each March. The state-level revisions use the same full-count methodology but apply it to subnational geography, where sample sizes are smaller and revision volatility can be greater.