October's 165,000 is the first of several versions of that number. The same report already shows how much those versions can move: August was revised up from 133,000 to 186,000 and September from 29,000 to 60,000. The White House counted that as 84,000 more jobs than first reported.
A jobs number actually gets revised in two different ways, for different reasons, and coverage often blurs them together.
What happens to October's 165,000 from here?
It will be revised at least twice more. The Bureau of Labor Statistics (BLS) publishes its first payroll estimate about three weeks after the month ends. The second estimate comes a month later and the final sample-based estimate two months later. After that the number stays fixed until the annual benchmark. Nobody knows yet which direction October's revisions will go.
That schedule tells you where the other months stand. In this report, August got its final monthly revision. September got only its first, so it can still change in next month's report.
Even "final" isn't quite final. In BLS's own example, the March 2023 benchmark, published in February 2024, revised data from April 2022 through December 2023. If that pattern holds, the March 2026 benchmark due early next year will also reach the late-2026 months, October included.
Why do the monthly revisions happen?
Mostly because some employers report late. The Congressional Research Service says monthly revisions happen largely because some private employers report late, whatever their size, industry, location or pay frequency. BLS publishes the first estimate quickly with the responses it has, then adds the stragglers over the next two months.
Fewer businesses are answering the survey at all, which makes this gap bigger. An Edgeworth Economics analysis found the payroll survey's response rate fell from just over 60% in 2016–2019 to under 43% in recent years, and its authors expect substantial revisions to continue. That figure comes from 2025 analysis. We couldn't find a published 2026 response rate.
What is the annual benchmark revision?
It's a yearly check against a near-complete count of jobs. The survey sample covers about 30% of total employment. So once a year BLS compares its estimate for March with a count taken mainly from unemployment-insurance tax records, which nearly every employer has to file. That process recalculates nearly two years of unadjusted data and five years of seasonally adjusted data.
Late responses aren't the issue here. BLS says the benchmark corrects for sampling and modeling error. The biggest modeling error comes from business births and deaths.
What is the birth-death model, and why does it matter?
It's how BLS estimates jobs at businesses too new to be in its survey yet. BLS describes an "unavoidable lag" between a business opening and when it can be added to the sample, so a model fills in jobs gained and lost as firms open and close. A BLS official put it bluntly to Scripps News: "Non-farm payroll is a model."
Michael Horrigan, a former BLS official now at the Upjohn Institute, says business births and deaths are the key reason for the gap between survey estimates and the tax-record count. After errors during the pandemic, BLS changed the model in 2024: it began updating its forecasts quarterly instead of yearly and added a regression component.
The overcounting continued anyway. Fed Chair Jerome Powell described an "almost predictable overcount" tied to the model, and Marketplace reported that BLS had overestimated job creation and revised it down three years in a row. BLS changed the model again starting in January 2026, aiming to reduce errors in the monthly estimates and shrink the annual revisions.
How big were the recent revisions, really?
They were large in raw numbers but small as a share of all jobs. The final 2024 benchmark was down 598,000 and the final 2025 benchmark down 898,000. The 2025 revision started as a preliminary estimate of -911,000, released in September 2025 and covering April 2024 through March 2025. That was 0.6% of employment, and the final version went into the data in February 2026.
The direction mattered more than the size. BLS's guide says that historically, benchmark revisions to total employment have been very small. The guide doesn't mention that three years in a row they all went the same way, down. Econoday notes that payrolls have a strong tendency to be revised lower.
Did the new model fix the problem?
It's too early to say. The preliminary benchmark for March 2026 was -79,000, or -0.1%, much smaller than the last few years. But nobody has shown whether that's because of the new birth-death model or because the job market was simply steadier.
Be careful with what you read about that figure. Before the release, some forecasts expected an upward revision, and Goldman Sachs had projected an increase. At least one headline reported that BLS had revised payrolls higher for the first time since 2022. BLS's own figure is a downward revision of 79,000.
That number is still preliminary. The final March 2026 benchmark comes out early next year and could differ. Last cycle, the change from preliminary to final was small: -911,000 to -898,000.
Are big revisions a sign the data is rigged?
The sources here explain the revisions as statistical problems, not manipulation. In August 2025, President Trump fired BLS Commissioner Erika McEntarfer, claiming the agency's data were "rigged." Horrigan's view, as the title of his Upjohn piece puts it, is that the revisions are concerning, but not for the reasons the administration thinks. He points to business births and deaths instead. The American Action Forum noted that the 2026 benchmark revisions got little media attention compared with the fight a year earlier.
Does the unemployment rate get revised too?
Yes, but less often. The unemployment rate comes from a separate household survey. Its seasonal adjustments are calculated each month, but earlier months are revised only at year's end, when the last five years are updated. So October's 4.1% unemployment rate, down from the month before, won't change month to month the way the 165,000 payroll figure will.