US jobs data in 2026: what the BLS publishes and how to read it

The US Bureau of Labor Statistics published two sets of employment figures at the end of August 2026, and they are not the same kind of number. One counts jobs that already exist, down to county level. The other projects where jobs will be in ten years. Together they cover most of what a marketing or research team needs from labour market data. Read without care, they will mislead you, because one carries no uncertainty at all and the other is mostly uncertainty.

What does the county file actually count?

The County Employment and Wages release for the first quarter of 2026 covers the 376 largest counties in the United States. National employment stood at 154.8 million, up 0.1 per cent over the year, and the average weekly wage rose to $1,654, up 3.9 per cent. Of the 376 counties, 358 recorded a rise in average weekly wages.

The county detail is where it earns its place. Licking County in Ohio had the largest employment increase over the year at 3.6 per cent. Washington, DC, and Arlington County in Virginia had the largest falls, both at 4.5 per cent. Average weekly wages rose 23.4 per cent in St Tammany Parish, Louisiana, and fell 9.6 per cent in San Francisco County. Those are four very different local economies sitting inside one national figure of plus 0.1 per cent, which is the argument for reading public data at the resolution your decision is actually made at.

One thing to know before you use it. This release comes from the Quarterly Census of Employment and Wages, which is a count rather than a sample. It draws on employer reports covering more than 95 per cent of US jobs. So it carries no margin of error, because nobody in it was estimated from anybody else. 

What the projections model forward

The second release, Employment Projections 2025 to 2035, is the opposite kind of number. It has total US employment rising from 170.3 million to 176.2 million by 2035, a gain of 5.9 million jobs or 3.5 per cent. Private healthcare and social assistance is projected to add more than 2.2 million jobs, about 37 per cent of all new jobs over the decade. Nurse practitioners are the fastest growing detailed occupation at 41.0 per cent, with data scientists at 34.6 per cent.

The Bureau is candid about what this is. The release states that "this precision in the data does not account for the inherent uncertainty of predicting long-term changes in the labor market", and advises focusing on "the direction and relative size of projected changes, rather than on the precise value estimates". Use it to work out which industries are growing and which are shrinking. Do not use it to size a market in 2035 to the nearest hundred thousand.

Three kinds of number, and what each one supports

Almost every figure a marketing team takes from public data is one of three things, and which one it is decides what you may claim from it.

• A count. Everyone in scope was counted, as in the county employment file. No error band. You can quote it precisely and compare small differences between places.
• A survey estimate. A sample stands in for a population, as in the American Community Survey, the US Census Bureau's annual household survey. It comes with a margin of error, and at small geographies that band can be wide enough to reverse the ranking you were about to act on.
• A projection. A model output, as in the ten-year employment projections. It supports direction and relative size, not a precise value.

Getting this wrong is quiet rather than loud. Nothing breaks. You simply end up defending a decision with a number that was never built to carry it.

What a marketing team can do with it

Three uses stand out. For business-to-business targeting, the industry and occupation cuts show where your buyers work and which sectors are adding staff. For segmentation, the wage figures give you an income anchor by place that does not rest on self-reported survey answers. For timing, a county with employment down 4.5 per cent over the year is not the county to open a regional campaign in this quarter, whatever the national trend says.

The honest limit is that jobs data describes jobs, not the people who hold them. It will not tell you the household a worker goes home to, who else lives there, or what they spend. For that you need household data, which is why the jobs figures work best joined to the census and survey sources covering the same geographies. Two definitions are worth having straight first: class of worker, which separates private, public and self-employed workers, and not in labour force, which covers everyone the employment figures leave out.

This is also where a persona either holds up or does not. If your customer persona says your buyer earns a particular amount in a particular county, someone should be able to open the file that number came from and see whether it was counted, estimated or projected. Cambium AI builds population models from verified public data for that reason, and publishes how the figures are put together so the trail is there when a decision is questioned. Which method suits which decision is a separate question, and there is more on matching the method to the stakes elsewhere on this blog.

None of this needs commissioning. Both releases are free, published on a schedule, and are among the sources government departments and central banks plan against. The work is knowing which number you are holding.

If you want a population picture where every figure can be traced back to the file it came from, that is what Cambium AI is built for. See it in Cambium AI →