Job Market Signal — August 2026: The wall gets a fourth brick, and the map that doesn't move

Note, 3 October 2026: two findings in this edition did not hold up under a closer check: the entry-level decline and the shift from customer service to care. Details in the September edition.

In May, one job posting in forty was entry-level. In June, still one in forty. In July, a touch fewer. In August, fewer again — the fourth month in a row that the door into the labour market got a little narrower. It is the steadiest signal we track, and it is going the wrong way.

Most of what makes a market report exciting each month turns out to be noise: a source changes volume, a feed gets fixed, and a headline moves for reasons that have nothing to do with anyone’s job search. So we throw most of it away and keep only what survives being measured the same way twice. Two things survived August, and one thing that looked like a story turned out — on a closer look — to be a single company.

The wall gets a fourth brick

The entry-level wall is the number we trust most, because it is the one that holds up under every way we cut the data. On the same sources at similar volumes in both months, entry-level roles fell from 2.1% of postings to 1.7% — a fifth of them gone in a month, the sharpest single-month drop in the four we have watched. Strip out the two sources that grew in bulk-import batches and the fall is cleaner still, a steady weekly slide with no cliff in it: 2.1, 2.0, 1.6, then holding just under 1.6 through the second half of the month.

May to August, the direction has never once reversed. That is rare in this data. Almost every other line we plot wanders up and down inside its own noise; this one only points one way.

There is a cruel geometry to where entry-level roles do appear. They are best represented on employer career sites — the channel a job seeker reaches by visiting thousands of individual company pages one at a time, the part of the market that is hardest to search by hand. The people with the least experience are the ones the open market serves worst.

Demand moved from the phone to the ward

The second thing that survived August was a rotation in what employers ask for — and it is coherent enough to name. Across the matched set, a whole cluster of customer-facing and coordination skills fell together, while care and clinical skills rose.

Eleven related customer-service, cargo-handling, logistics and project-coordination skills each dropped between 8% and 15%. The care side rose across the board: helping patients with special needs, supervising nursing staff, rehabilitation. And the shift is not just a quirk of how we label skills — it holds at the level of whole industries, where Health & Social Work gained the most ground of any sector and Transportation & Storage and Information & Communication lost it.

Part of the rise on the education side — managing student admissions, assisting with enrolment — is plainly the autumn intake, and we would expect it to fade by October. But the movement away from customer-service and coordination roles is not seasonal in any obvious way, and it is worth watching whether it becomes September’s story too. As always: one month is a data point, two is a candidate, three is a trend.

The map that doesn’t move

Here is the quiet finding underneath the monthly churn. Three editions in, the geography of skill barely changes. Each country has a distinctive demand fingerprint, and month to month it holds almost still.

None of these is a headline. Each is a structure. A job seeker’s sense of “what the market wants” is really a sense of what their country’s market wants, and those are as different from each other as the ports of Portugal are from the payroll desks of Prague. If you are moving countries, you are not entering the same labour market at a different address — you are entering a different economy.

When one employer is a whole country

And then there is the trap the fingerprints set for us, which is worth showing because it is exactly the kind of thing that produces a confident, wrong headline.

Austria’s most over-represented skill in August was “test semiconductors,” at 29 times the global rate. Read quickly, that is a story: Austria is pivoting to chips. Read carefully, it is one company — Lam Research, the American chip-tool maker, building out its Villach site, accounting for 156 of the 164 postings behind that number. Strip Lam Research out and Austria looks like what it has looked like all along: manufacturing and hospitality, with AT&S making circuit boards, Billa hiring for shops, Magna building cars.

That is not a flaw in the data — it is the data being honest. A single large employer really can bend a small country’s numbers, and knowing whose hiring you are looking at is the difference between a trend and a press release.

Remote, still stuck

Two smaller notes to close. Remote work did not move: 5.4% of postings in August against 5.2% in July, flat for the third comparison running. The more useful way to read remote is by channel — it ranges from one posting in five on tech-niche boards to fewer than one in two hundred at national employment services — so a single market-wide number for remote is close to meaningless, and we will keep reporting it split by where you look.

And a note on our own plumbing, in the working-out-loud spirit these editions are written in: August was a month of pipeline repair as much as pipeline reading. Sources we had quietly lost went dark and were brought back; a couple of feeds that looked like they were declining turned out to have been over-counting earlier, not shrinking now. The numbers above are computed to survive exactly that kind of change — which is the whole reason we compute them the hard way.

Into September, three questions carry over: whether the entry-level wall makes it five months or the graduate season reopens the door; whether demand keeps rotating out of customer service and into care; and whether any of the national fingerprints finally shifts on its own, rather than because one employer arrived in town. The numbers will be here either way, measured the same careful way — with the method notes attached.

Methodology: every figure comes from Kitsuno’s production crawl, classified by the Extractor agent using the ESCO skills taxonomy and an internal role-family scheme. The dataset covers 533,842 unique postings collected across August 2026 from 53 sources; 84.8% carry an identified location. Month-over-month figures are computed on a matched-volume set of about 40 sources whose August-to-July volume ratio falls between 0.5 and 2.0, so that sources joining, leaving, or changing scale do not masquerade as market movement. Country skill “fingerprints” are over-index measures — a country’s share of a skill divided by the global share — filtered to exclude single-poster noise and cross-checked against the employers behind each skill; where one employer drives a pattern, we say so. Single-month figures use the full footprint and say so.

For how the pipeline works, read How we measure job-market signal. For the live numbers, visit Market Pulse.