Job Market Signal — July 2026: The same employer, two answers on pay

Take one German employer in July. On its own career site, it shows pay in four postings out of ten. On the job boards, the same employer, the same month, the same country: one out of ten. We ran that comparison across 46 German companies posting in both places at once, and the pattern held — 41% salary disclosure on their own career sites, 10% on boards. Four to one.

The pay information a job seeker never saw wasn’t missing. It existed the whole time, one click deeper than where most people search.

Across the full experiment — 108 companies in several countries, each posting on both their own site and boards within the same country and month — career sites disclosed pay in 43% of postings against 25% on boards. In the United States the gap ran the same direction on a small sample. Some of the difference is boards stripping salary fields in syndication, some is what employers choose to push to each channel, and a part is how well any crawler can read each format. For the person searching, the practical truth is the same either way: you see materially less pay information on boards than the same employers publish on their own sites.

One market, four doors

"The job market" is really several ponds wearing one name. What you believe about pay, remote work, and entry-level jobs depends on which one you fish in.

The same-employer test is the sharp edge of something broader. Cut July’s 552,000 postings by acquisition channel — employer career sites, aggregators, national employment services, commercial job boards, tech-niche boards — and they barely resemble each other:

Remote work splits even harder. At the national employment services, 0.4% of postings are remote. On tech-niche boards, 21.4% — a fifty-fold spread across channels of the same market in the same month:

Whatever a seeker believes about “the market” — how rare remote is, whether anyone shows pay, whether entry-level roles exist — is largely an artifact of which door they walk through. And the doors differ in one more way that rarely gets measured: depth. The average posting on an employer career site runs about 3,000 characters of description. On commercial boards, about 400. The channel we crawl directly for exactly this reason — thousands of individual career sites, the part of the market hardest to search by hand — is also the one with the richest postings. The data makes that argument on its own.

Our own headline number was wrong

Raw July numbers say EU pay transparency jumped twelve points in a month. It didn't. Here's how we know — and what's actually true.

If we ran this report the easy way, July’s lead would write itself: EU-weighted salary disclosure leapt from 28.7% to 40.6%. A twelve-point jump, right after the Pay Transparency Directive’s transposition deadline. Irresistible — and false.

Three things happened to our pipeline in July, not to the market. A wave of extraction upgrades landed mid-month, and the weekly numbers show a cliff, not a slope — the fingerprint of software, not of employers changing behaviour. A geo-coding fix on the French employment-service feed added tens of thousands of correctly-located, high-disclosure French postings to the EU average. And our direct career-site crawl grew eight-fold, changing the mix of what we see.

Recomputed the honest way — same sources, similar volumes, both months, with the affected feeds excluded — EU disclosure moved from 22.7% to 23.9%. Flat, for the third month since the June 7 deadline passed. The count of member states that have actually transposed the directive into national law is still zero.

Most market reports would have shipped the twelve points. We’d rather tell you our microscope got better — because it genuinely did, and that part is worth having. The disclosure levels we publish are now closer to the truth than in any previous edition: the UK reads 49%, squarely inside the 43–51% range external studies report, where our earlier editions read implausibly low.

The real July signals

Two numbers survived every cross-check we threw at the month.

The entry-level wall got a third brick — and it’s slightly tighter. On the same sources at similar volumes, entry-level roles fell from 2.6% to 2.35% of postings. That direction holds on every cut we ran, which none of the flashier numbers this month could claim. May, June, July: one posting in forty, and now a touch fewer. And the channel view closes a loop: entry-level roles are about three times better represented on employer career sites (3.5%) than at public employment services (1.2%). The seekers with the least experience are best served by the channel that is hardest to search by hand.

AI wrote the headlines, not the job titles. Postings naming AI explicitly — in the title or in a machine-learning, LLM or GenAI skill — fell from 1.77% to 1.37% on the same-sources cut. A decline of about a quarter, in the month AI dominated every front page. One honest footnote: we widened this measure in July to include AI named in titles, so it isn’t comparable in level to the stricter skills-only figure earlier editions carried — but within July’s own like-for-like comparison, the direction is down. Three months in, the pattern is consistent: whatever AI is doing to work, employers are barely writing it into what they hire for, and in July they wrote it in less.

We said remote was shrinking. It stopped.

Two of last month's storylines did not survive contact with July. We're telling you which ones.

June’s edition reported remote work measurably declining, and said a second month down would make it a trend. July is in: on the same sources at similar volumes, remote went from 3.8% to 3.9%. Flat. The decline did not continue, so it was a data point, not a trend — and the better description of remote is now stuck: rare, going nowhere, and (see above) so unevenly distributed across channels that a single market-wide number borders on meaningless. From here on, this series reports remote by channel.

Second correction: June’s skills rotation — “less coordination, more analysis” — partly reverted. Operations-management skills bounced back up 18% after June’s 27% drop, and the analysis cluster drifted back down. A blip, not a rotation; we said a repeat would make it the story of the summer, and it didn’t repeat. One piece survives at a coarser level: project and program management roles fell from 3.45% to 2.93% of classified postings, a second consecutive monthly decline. The role trend stands; the skills story is withdrawn.

One month is a data point. Two months is a candidate. We’ll keep telling you which is which.

France, at street level

A fixed data feed turns out to be the most human story of the month.

The French geo-coding fix that complicated our EU average did something better for the picture of France itself: tens of thousands of postings from the national employment service now land on the map. And the France they show is not tech-board France. The most over-indexed demand in French postings this month, against the global baseline:

Bus drivers, farm crew leads, pâtissiers, paint-shop finishers. Most job data over-represents laptop work, because laptop work is what gets posted where crawlers look. This month, our picture of France got closer to the whole economy — which is the quiet point of building coverage in the first place. One consequence worth naming plainly: because the feed changed so much, we publish no France month-over-month comparisons this edition. France gets a fresh baseline from July.

And one new country on the map: Czechia enters at #8, with 20,000 postings and a 54.1% pay-disclosure rate — higher than the UK, Germany, or Austria on its first appearance.

Into August, three questions carry over: whether the entry wall makes it four months; whether any member state finally transposes the directive and the honest EU number reacts; and whether the same-employer pay gap holds as the pilot grows from 108 companies toward the full career-site registry. The numbers will be here either 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 552,046 unique postings collected across July 2026 from 56 sources in 108 countries; 86.1% carry an identified location. July saw six new sources, major extraction upgrades, and a French geo-coding fix, so every month-over-month figure is computed on the same sources at similar volumes in both months, with pipeline-affected feeds excluded where noted; single-month figures use the full footprint and say so. The same-employer comparison covers 108 companies posting on both their own career site and job boards within the same country and month — a pilot sample. Channel cross-sections are July-only portraits, not causal claims, and confound with country and role mix. Salary disclosure is pattern-detected; July’s extraction improvements make current levels more reliable than earlier editions, and movements more reliable than absolute levels. The AI measure was widened this month to include AI named in job titles; compare its direction, not its level, against earlier editions.

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