Half have been looking ten weeks. A quarter, more than six months.

A rate counts people, not weeks

The unemployment rate answers one question well: what share of the labour force is without work and looking for it. It answers nothing at all about the second question anyone in that position actually has, which is how long this is going to last. Two economies with the same rate can be entirely different places to be unemployed in - one where most people are out for a month, one where most are out for half a year.

The survey that produces the rate does ask. Each month it records, for every person counted as unemployed, how many weeks they have been looking. Those answers are published as two summary numbers, and the reason there are two is that one of them is easy to misread on its own.

In July 2026 the median was 10.5 weeks and the mean was 24.9. The median is the middle: line up all 6,916,000 unemployed Americans by how long they have been searching, and the person standing in the centre is ten and a half weeks in. The mean is the arithmetic average of the whole line. It sits 2.4 times higher, and the only thing that can do that is a tail.

What a mean is doing when it doubles a median

A mean and a median only diverge when a distribution is lopsided. If most searches ended near the middle, the two would sit close together. They sit fourteen and a half weeks apart, which says that a substantial group at the long end is carrying the average away from the experience of the typical person in the queue.

The published counts say how big that group is. In July 2026, 1,771,000 people had been looking for 27 weeks or more. Against 6,916,000 unemployed altogether, that is 25.6% - one in four. Those people are not a rounding error in the average; they are the reason it is where it is.

This is why the median is the number to quote to a person and the mean is the number to quote about a system. Someone who has just lost a job wants to know what the middle of the queue looks like. A policymaker asking how much long-term joblessness a country is carrying wants the measure that the long end moves.

The middle drifted. The tail grew.

Set July 2026 against July 2019, the last comparable summer before the pandemic disturbed every one of these series, and the three measures do not move together at all.

The median went from 8.9 weeks to 10.5 - a week and a half longer. The mean went from 19.6 to 24.9 - more than five weeks longer, three times the movement of the median. And the long-term share went from 20.2% to 25.6%, a rise of five and a half points.

Read together those three say something a single number cannot. The typical search has lengthened, but only modestly. What has changed much more is what happens to people whose search does not end quickly: there are more of them, and they are waiting longer. The distribution did not shift; it stretched.

Against last July the picture is quieter and worth stating plainly, because a desk that only reports movement invents it. The median rose by three tenths of a week, the mean by nine tenths, and the long-term share by half a point. On the year, this is close to flat.

What this measure does not say

It does not say how long a job search takes. This is the distinction that matters most and the one most easily lost. The survey asks people who are unemployed at the time of the interview how long they have been looking so far. A search that is still running in its thirtieth week is recorded at thirty weeks, not at whatever it eventually totals. The series therefore measures the accumulated waiting of the people currently in the queue - which is a real and useful thing to know, and is not the same as the length of a completed search. A sentence beginning "half of job seekers find work within ten weeks" is not supported by this data and does not appear in this piece.

It does not count everyone who is out of work. Somebody who stops looking leaves the unemployment count altogether, and their accumulated weeks leave with them. That is the subject of the previous edition of this desk: between May and July, the labour force shrank by 984,000. Departures of that kind do not lengthen the measured duration - they remove it from the sample. Whether that flattered these numbers is not something this release can settle.

And the long-term unemployed are not discouraged workers. Discouragement is a separate published category, and it means having stopped looking. The 1,771,000 people in the 27-weeks-and-over count are still searching - that is precisely what keeps them in it. Merging the two makes the more sympathetic number the less accurate one.

For anyone actually looking for work, the useful reading is this. The middle of the queue is about ten weeks, and it has been roughly that for a year. The risk that matters is not the median; it is the quarter of the queue at the far end, and the thing that moves a person toward that end is time itself. Nothing in this release says a longer search is a worse candidate. It says the queue has a long tail, and that a search which passes six months is in a group that has grown.