US jobs: every sector on one chart

Where the jobs came from

Split every job in the country into six groups. The bars below show how many jobs each group added or lost in September 2026. A minus number means that group shed jobs while the total may still have risen.

• Health and education +20,000
• Leisure and hospitality +10,000
• Professional and business -9,000
• Making and building things +18,000
• Shops, transport, finance and the rest +7,000
• Government -17,000

The six parts add up to +29,000 jobs, which is exactly the headline figure. They have to: every job in the country sits in one of the six and in no other.

These add up exactly, and they have to. Every job sits in one of the six groups and in no other, so the six changes must sum to the change in the total. Nothing here is estimated or weighted.

Every sector, side by side

The ten parts of the private economy plus the three levels of government, at the same moment, over three different lengths of time. Sorted by the last twelve months, so the sectors doing the work sit at the top and the ones shedding jobs sit at the bottom.

The ten private sectors add up to private jobs exactly, and the three government rows add up to government jobs exactly. Those two totals then add up to the headline. Nothing is left over.

The official count against ADP

Two organisations count private jobs every month, and they line up sector for sector. One surveys businesses, the other counts payslips at the firms it processes wages for. Where the two part company is often worth more than either number on its own, because it shows how much the answer depends on who you ask.

ADP covers private jobs only. It never sees a government payroll, so there is no public sector row to compare. Both columns are read at the same month, which is the newest month both surveys have published. ADP comes out two days earlier, so for a couple of days each month it is already a month ahead and this table waits for the official count to catch up rather than comparing two different months.

Six ways to count a job

Nobody counts every job in the country every month. Each of these measures a different slice of the market, by a different method and on its own schedule, and knowing which slice you are looking at is most of the skill.

• The monthly jobs report

  About 119,000 businesses and government offices fill in a form every month, and between them they cover roughly a third of all the jobs in the country. What they report is filled positions rather than people, so somebody holding two jobs is in there twice.

  A job here means anything outside farming, private households and the armed forces, and those three exclusions are the only reason the word "nonfarm" is in the name. This is the headline. When you hear that the economy added 150,000 jobs, this is the count being quoted.

• ADP’s count

  A private payroll company counts the jobs at the firms whose wages it actually processes, then scales that up. So it is a real count of real payslips rather than a survey, but only of its own client base. Private jobs only: it never touches a government payroll and can tell you nothing at all about the public sector.

  It lands two days early, which makes it the first read of the month. It also disagrees with the official number often and by a lot, so treat it as a second opinion rather than a preview.

• New unemployment claims

  A count of people filing for unemployment pay for the first time. Not a survey and not an estimate, but an administrative record of forms that were actually filed. Only people who qualify for state unemployment pay and bother to claim it turn up in it, which leaves out the self-employed and most gig workers entirely.

  This is the fastest signal there is on the jobs market, weekly and only five days behind, and if layoffs are starting it moves before anything else does. The chart here is monthly, so it plots the average of a month's weeks and waits until every week of that month has printed.

• People still claiming

  The same records, counting people who filed earlier and are still drawing pay week after week. That makes it the stock rather than the flow. New claims tell you how many people just lost a job; this tells you how many are still without one.

  Read the two together and you can separate "people are being let go" from "people cannot find anything new". Since 2024 the second has been the more worrying half.

• Job openings, hires and quits

  A separate monthly survey of about 21,000 workplaces, asking how many roles were unfilled on the last working day, how many people started, and how many left. It is measuring demand for workers rather than the number employed. Hires and leavers are both far bigger than the monthly jobs figure, because that figure is the difference between them.

  Openings and quits both turn down well before layoffs turn up, which is what makes them the early warning. The whole thing arrives about a month later than everything else here.

• The household survey

  About 60,000 households get asked directly whether the people living there have work. This is where the unemployment rate comes from, and it is a completely different survey from the one that produces the jobs number. It counts people rather than positions, so somebody with two jobs counts once, and it picks up the self-employed and farm workers, who are missing from the headline count.

  That is why the jobs number and the unemployment rate can point opposite ways in the same month. They are not the same survey and they are not measuring the same thing.

The two big surveys disagree constantly, and neither one is wrong. One counts jobs, the other counts people. One misses the self-employed while the other includes them. One is far larger, and that is why it is the steadier of the two month to month. When a headline says "jobs up, unemployment up too", that is usually the two surveys talking past each other rather than a contradiction.

Every one of these numbers gets revised. The jobs report is revised twice in the two months after it first appears, and then again once a year against near-complete tax records. Those annual revisions have run to several hundred thousand jobs. This page follows the corrections rather than freezing the first version, so a figure here can move after you first read it.

The sector bands on this page add up exactly, which is unusual. They are counts of people, so the parts simply sum to the whole. Nothing is estimated or weighted, and there is no rounding gap to explain away.

What each one actually means

Short answers in plain words. None of it is needed to read the chart, but it is here for when a name on the list makes no sense.

The numbers

This page shows the last ten years. The full history back to 1990, every sector, all four windows and the machine-readable feed come through the Helious API, and through MCP at api.helious.io/mcp if you want your own tools to read it directly.

How this page is built

FAQ