The Zeltru Report · Edition No. 1 · Full Edition

How is the job market right now?

What 503,503 job postings — read daily from 1,124 employers’ own career boards — actually show. Not survey answers. The boards themselves, measured.

Published September 1, 2026·Data as of August 30, 2026·Methodology·Brief version

§1The instrument: what this report measures, and from where

Most descriptions of the job market come from surveys. The Bureau of Labor Statistics asks a sample of employers how many openings they have; job sites report how many listings they carry; recruiters describe what they see. This report takes a different route. Zeltru reads employer career boards directly, every day, on the five applicant tracking systems that host most large-employer hiring: Workday, iCIMS, Greenhouse, Lever, and Ashby. There are no aggregators in the chain, no job-search sites, and no self-reported figures. A posting counts because it is on the employer’s own board, and its date is the date the employer published.

Reading a board daily turns a list of jobs into a time series. A posting that is present today and absent tomorrow has, by our definition, come down. A posting published on June 3 and still up on August 30, 2026 has been listed 88 days, and we can say that because the employer said so, not because we guessed from when we first saw it. Everything in this report follows from those two facts: presence over time, and the employer’s own date.

Two numbers above deserve a sentence each. The 95.6% date coverage is high because most applicant tracking systems publish a posting date as a matter of course, and it is not 100% because some employers configure their boards not to. The gap between 1,124 employers monitored and 1,037 in the age statistics is the employers whose boards carry no dated, non-evergreen postings; they are counted in the total and shown on their own pages, but they cannot contribute to a distribution of ages.

Job postings are not job openings

A reader comparing this report with the federal Job Openings and Labor Turnover Survey should keep the two apart. JOLTS estimates openings, a survey answer about positions an employer intends to fill within 30 days. This report counts postings, public listings on employer boards, whatever the employer intends. One opening can be advertised under several postings; one posting can stand for many hires, or for none. The measures rhyme, and they are placed side by side in §6, but they are different instruments and this report never combines them.

§2How long job postings stay open: the age of 476,618 active postings

The age of a posting is the number of days between the date the employer published it and August 30, 2026. Figure 2 sorts the 476,618 dated, non-evergreen active postings into the seven age buckets the dataset uses.

Half of the standing stock is young. A third of all dated postings are between eight and thirty days old, and one in six went up within the last week. The other half is where the market’s character shows. The buckets past 30 days do not fall off a cliff; they taper, and 4.3% of the postings a job seeker can apply to today were published more than a year before the data date.

Because job seekers and employers tend to think in thresholds rather than buckets, Figure 3 restates the same population three ways. Each bar is the share of dated active postings listed at or beyond a threshold, so a posting listed exactly 30 days is inside the 30-day figure.

Thresholds read “N days or longer”: a posting at exactly N is inside the figure. Shares are rounded to one decimal and may not sum to exactly 100.

What a snapshot of ages can and cannot say

Two cautions travel with these figures. First, the age of a posting is not the time it takes to fill a job. A posting can be filled and left up, or taken down unfilled, and this report cannot see hires. What it can see is how long an advertisement has been in front of applicants, which is the number an applicant needs.

Second, a snapshot of what is open today leans toward postings that stay up a long time. A posting that lasts 200 days has 200 chances to be counted on any given day; a posting that lasts 10 days has ten. That is why the median age of active postings, 28 days, is higher than the median age at which postings are removed, 21 days (§4). Neither number is wrong. They answer different questions: the first describes what an applicant sees, the second describes what employers do.

§3The longest-listed postings, and what a long-listed posting can mean

The tail of Figure 2 ends in postings that have been listed for years. Before the table, a word about what such a posting can be, because the number alone does not say. A requisition can stay open because it describes continuous hiring, such as a nursing pool or a store-associate posting that fills over and over. It can stay open because the role is a specialty that takes a year to fill. It can be reused for a series of hires under one posting. Its date can be the day the requisition was created rather than the day it was last advertised, which is how some boards record it. And it can stay open because nobody took it down, with or without an active process behind it, which is the case people mean when they say “ghost job”. This report measures how long a posting has been listed. It does not measure intent, and it does not claim to.

Figure 4 shows the ten longest-listed postings among dated, non-evergreen postings with plausible dates, one per employer, labeled by industry and the kind of role rather than by name. The named rows, with a link to each live listing, follow in Table 2.

Table 2 is the same idea with names, and with two kinds of posting the figure leaves out. Postings whose employer-published date falls before the board itself existed are shown and flagged as implausible rather than dropped, since the date is the employer’s and a reader can judge it. Talent-community and continuous-hiring postings are shown and badged. That is why the top row of the table, at 6,113 days, is older than the top row of the figure, at 5,002. Every row links to the live listing. Palantir holds the 11 longest-open postings by age. It is shown once here; the full ranking is at /longest-open.

Dates are as published by the employer. Where a date is implausible we say so and show the row anyway. A long-listed posting is a fact about a listing, not a finding about an employer’s hiring; the reasons a posting stays up are described at the top of this section. Employers who believe a date shown here is wrong can write to support@zeltru.com.

§4How fast job postings are removed: the removal lifecycle

Reading boards daily also shows what comes down. A posting is recorded as removed after three consecutive daily reads in which it is absent from a board that was read successfully. A board Zeltru could not read that day produces no removals, so an outage cannot manufacture one. This series began on July 24, 2026, and the figures below cover July 24, 2026 through August 30, 2026: 37 days, or about five and a third weeks. One day inside that window, August 23, 2026, had no fleet read: the collector was upgraded and split into shards that day, and the fleet did not walk. A posting that came down on it was recorded on the next successful read. Three more — 24, 26 and 28 August — are not confirmed as complete reads: a handful of boards went unread on the first two, and on the third every board was read but the run's own record is incomplete. The days are listed here rather than smoothed over; the methodology page carries each with its cause.

In that window, 185,659 postings came down across 1,012 employers. 175,474 of them came down in August. The median posting was 21 days old when it was removed, and the middle half of removed postings were between 10 and 42 days old.

When a posting is going to come down, it tends to come down within about three weeks. That is the ordinary lifecycle of a listing that is filled, withdrawn, or replaced. The postings listed 90 days or longer in §2 are a different population, one that the removal series mostly does not reach, and describing that population over time is the purpose of this series.

A caveat that travels with every removal figure

Removal statistics are measured over a 37-day window, and a window that short cannot see the end of a posting that lasts longer than it. Long-lived postings are therefore under-represented in the removed set by construction, and the ages here are not survival statistics; they describe postings that were removed during the window, not how long a posting can be expected to last. The window is stated rather than corrected for. As the series lengthens, later editions will be able to say more, and will say when they can.

§5Job posting age by industry, applicant tracking system, and geography

The aggregate figures in §2 average across employers whose hiring looks nothing alike. The cuts in this section separate them. Each cut is published only where its denominator is large enough for a share to mean something; where it is not, the count is shown and the share is a dash.

Which industries carry the most postings, and the oldest

Figure 7 shows where the dated postings are. The largest group is the one we have not yet classified: 468 employers carrying 193,526 postings. Those postings are in every total in this report and in no industry row, and closing that gap is on the list for Edition No. 2. Among classified industries, retail, technology, and healthcare each carry more than 55,000 dated postings, and retail does so from only 14 employers, a reminder that a handful of very large boards can dominate an industry cut.

Figure 8 turns from volume to age. For each industry with enough employers to publish a share, the pale marker is the share of postings listed 30 days or longer and the dark marker the share listed 90 days or longer. Industries are ordered by the 30-day share.

The two markers tell different stories about the same industry. Technology has the widest gap between them: 62.3% of its postings are a month old or more, but only 22.6% are a quarter old, which describes boards where postings routinely stay up past 30 days and then come down. Retail’s markers are closer together, 46.7% and 26.5%, which describes boards where a posting that reaches a month tends to reach three. Healthcare, at 43.8% and 12.3%, has among the youngest postings of any large group, consistent with high-volume clinical hiring that turns over quickly. These are descriptions of boards, not judgments about employers; the reasons a posting stays up are listed in §3, and they vary by industry as much as the numbers do.

A dash means too few employers for a share to mean anything; the count is still shown. Shares are computed within each industry over its dated, non-evergreen postings. These rows sum to 476,598 postings across 1,036 employers, against the 476,618 and 1,037 of the age-statistics population — a shortfall of 20 postings and 1 employer. The platform table below sums exactly. The gap is in the published industry cut, not in the totals, and it is stated here rather than reconciled.

Job postings by applicant tracking system

Every employer in this dataset runs its board on one of five platforms, and the platform turns out to be one of the strongest predictors of posting age. Figure 9 shows why that has to be read carefully.

Platform is a fact about an employer’s software, not about the employer. The profiles differ sharply because the populations do: Workday and iCIMS host large enterprises and hospital systems with high-volume, fast-turning boards, while Greenhouse, Lever, and Ashby host smaller, mostly technology, employers whose postings stay up longer. A share computed over Lever’s 2,429 postings from 21 employers moves easily and should be read as a description of those boards rather than of a market. The cut is published because a reader comparing two employers on different platforms deserves to know that the boards themselves behave differently.

Where the postings are: United States share

Country scoping is conservative. It follows the location the employer published on each posting; a posting whose location is ambiguous stays unscoped rather than being guessed, and smaller cells are reported as floors or not at all. On that basis, 382,112 tracked postings are in the United States, 75.9% of the total, and 68,069 of those have been listed 90 days or longer.

§6The wider labor market: federal series for context

The figures above describe 1,124 employers’ boards. The federal statistics below describe the whole United States labor market. They are placed side by side so a reader can hold both in view, and for no other reason: this report does not infer that one explains the other, and it does not combine them.

These are different objects, measured by different instruments on different populations. The federal series count openings an employer reports intending to fill, hires, separations and people filing for unemployment insurance; this report counts public listings on employer boards and how long each has been listed. A posting is not an opening and neither number is a version of the other, so they are never compared as levels, never differenced, and never combined. They can move in opposite directions in the same month with neither being wrong. They are placed side by side so that a reader can hold both, and the reading is the reader’s to do.

This section carries the same three series in every edition, so that over time the two pictures can be read against each other by the reader rather than reconciled by us.

§7Methodology and definitions: how Zeltru measures job posting age

Sources and cadence

Zeltru reads employer career boards on five applicant tracking systems: Workday, iCIMS, Greenhouse, Lever, and Ashby. Each board is read once a day. A read collects every posting the board publishes, together with the employer-published posting date, the title, the location, and a link to the listing. There are no aggregators, no job-search sites, and no self-reported data anywhere in the chain. Every posting in every figure links to its source.

Definitions

Employer

The organization whose board is read. Where one organization runs several boards, they are counted as one employer. Where one hosting tenant serves several legal employers, each is counted separately, on the basis of the hiring organization the board itself asserts.

Active posting

A posting present on the employer's board at the most recent complete read.

Removed posting

A posting absent from its board on three consecutive daily reads in which the board was read successfully. A board that could not be read produces no removals.

Reappeared posting

A removed posting that is later present again on the same employer's board.

Dated posting

A posting carrying a posting date published by the employer. 95.6% of active postings are dated. Undated postings are counted in totals and excluded from every age statistic; no date is ever estimated.

Continuous-hiring posting

A posting that does not describe one vacancy: talent communities, standing requisitions, pooled clinical or store-associate postings, and similar. These are excluded from age distributions and shown with a badge wherever they appear in a ranking.

Age

Days between the employer-published posting date and the data date. The date Zeltru first observed a posting is never reported as its posting date.

Threshold shares

“Listed N days or longer” means an age of N or more; a posting at exactly N is inside the figure. Counts and shares at each threshold are both published by the dataset and neither is derived from the other.

Implausible date

An employer-published date that falls before the board itself existed. Such postings are shown, flagged, and excluded from the record figure in §3; they remain in Table 2 because the date is the employer's and a reader can judge it.

Industry

Assigned per employer. Postings on a board carry that employer's industry. Employers not yet assigned an industry are reported as Unclassified.

Country scoping

Follows the location the employer published on each posting. Ambiguous locations remain unscoped rather than guessed.

Denominators, exclusions, and floors

Every statistic in this report carries its denominator. The age statistics use 476,618 postings: active, dated, and not continuous-hiring, from 1,037 employers. The removal statistics use 185,659 postings removed between July 24, 2026 and August 30, 2026, from 1,012 employers. Boards larger than a single read can enumerate are reported at the count enumerated, with coverage stated, so board-level counts are floors rather than estimates. Where a cut’s denominator is too small for a share to mean anything, the count is shown and the share is not.

Known limitations

Four limitations apply to this edition and are stated rather than corrected for. The removal series is 37 days old, so long-lived postings are under-represented among removals. A snapshot of active postings leans toward postings that stay up a long time, for the reason given in §2. Industry classification is incomplete: 468 employers carrying 193,526 postings are Unclassified, the largest single row in the industry cut. And platform composition differs so much across the five applicant tracking systems that a platform share describes those boards rather than a market. Each of these will narrow as the series lengthens and coverage improves, and each edition will say which have.

Frozen figures and corrections

Every figure in this edition was computed at press time and is frozen. A later refresh of the data becomes Edition No. 2 rather than a silent rewrite of this one, so a number quoted from this page stays quotable. If a figure or a label is corrected after publication, the correction is recorded in the site’s changelog with the date and what changed; nothing here changes without a visible record of the change. The dataset behind the age statistics is available as posting-duration.csv, and the complete method is at /methodology.

What Edition No. 2 will add

Edition No. 2 publishes on October 1, 2026 and will carry the first month-over-month comparisons this series can make, a removal window twice as long, an examination of the 10,123 reappearing postings and the cumulative age of postings that come down and go back up, and wider industry classification. Anything it cannot yet measure, it will say it cannot.

§8Questions this report answers

How long do job postings stay open?

Among 476,618 active postings with employer-published dates, the median has been listed 28 days. 48.5% have been listed 30 days or longer, 28.1% for 60 days or longer, and 18.8% for 90 days or longer (§2).

What percentage of job postings are more than 90 days old?

18.8%, or 89,408 of 476,618 dated postings, had been listed 90 days or longer as of August 30, 2026 (§2, Figure 3).

How long does it take for a job posting to be taken down?

Among 185,659 postings removed between July 24, 2026 and August 30, 2026, the median was 21 days old when it came down; the middle half were between 10 and 42 days old (§4, Figure 5).

Do removed job postings get reposted?

10,123 of the removed postings later reappeared on the same employer's board. Why, and what those postings' cumulative ages look like, is the subject of Edition No. 2 (§4).

Which industries have the oldest job postings?

Among industries with enough employers to publish a share, technology boards carry the largest share of postings listed 30 days or longer (62.3%) and retail boards the largest share listed 90 days or longer (26.5%). Utilities and energy boards carry the smallest 90-day shares, 6.2% and 5.5% (§5, Figure 8).

Does the applicant tracking system matter?

The profiles differ sharply, from 46.4% of Workday postings listed 30 days or longer to 83.2% of Lever postings, but the difference describes the employers who use each platform more than the platform itself (§5, Figure 9).

How many of the tracked job postings are in the United States?

382,112, or 75.9% of the tracked total, of which 68,069 have been listed 90 days or longer (§5, Figure 10).

Is this the same as the BLS job openings number?

No. JOLTS estimates openings from an employer survey; this report counts postings read from employer boards. They are different instruments and this report never combines them (§1).

Are these ghost jobs?

This report measures how long a posting has been listed and does not measure intent. A long-listed posting can be a continuous-hiring requisition, a hard-to-fill specialty, a reused posting, a board's date convention, or a listing nobody took down. §3 describes each; the figures do not distinguish them, and this report does not claim that they do.

Cite this edition

Next edition

Edition No. 2 publishes October 1, 2026. The Brief version of this edition carries the headline figures on one page.

Zeltru — dated, re-read on a daily schedule, every posting links to its source.