29% of Every Posting We Have Ever Seen Is Already Dead. 'Active' Is Mostly an Assumption.

For each source, the number of days between our snapshot and the most recent time ANY of its still-active postings was re-seen. Full corpus, 2026-08-13 03:22 UTC. A feed at 0.3 days is being re-read continuously. A feed at 10 days has told us nothing for 10 days, so every one of its 'active' rows is an assumption.

All four columns are full-corpus counts over the jobs table, not a sample. 'Live postings' is every row flagged active. 'Oldest live posting' is the single oldest first-seen date among that source's still-active rows, so it is an extreme value describing one posting, not a typical lifespan. 'Feed last re-seen' is the newest last-seen timestamp across all of that source's active rows, measured from the snapshot at 2026-08-13 03:22 UTC. reed is highlighted because it carries 10,403 live postings on a feed that has been silent for 10.3 days. The four sources with under 1,000 live postings are omitted from the table; they are included in every corpus-wide figure on this page.

Source: Seeker job corpus, 239,806 active job postings, snapshot 2026-08-12. seekerscore.com/insights/how-long-job-postings-live-2026

The lifespan question has two different answers

There is how long a posting stays listed, and there is how long the job behind it stays real. We can measure the first precisely. Of the 239,806 currently active postings, 119,502 (49.8%) were first seen more than 30 days ago. Only 35,617 (14.9%) first appeared in the last 7 days, and 5,930 (2.5%) in the last 24 hours. So roughly half of what you are applying into has been sitting there for over a month. The second question, whether the job is still real, is one no job board answers directly, and the honest position is that we can only observe how recently the source itself last confirmed the posting.

Some feeds go quiet, and their postings keep looking live

This is the finding we did not expect. Five of the twenty sources in the corpus have not had a single active row re-seen in more than a week: jooble at 11.9 days, reed at 10.3, usajobs at 10.0, themuse at 9.0 and findwork at 8.2. Between them they hold 17,130 active postings, 7.1% of the live corpus. Those rows are not marked stale anywhere a reader would see. They are marked active, because active means 'we have no evidence it is gone', not 'we checked'. Meanwhile workday, himalayas, amazon, apple, google, lever, citi, kaiser, jobicy, smartrecruiters and remotive were all re-read within the last 8 hours. Two postings side by side in the same search result can differ by ten days in how recently anyone confirmed they exist.

The oldest live posting on a board ranges from six weeks to six months

Take the single oldest still-active posting on each feed. On greenhouse it was first seen 187 days before this snapshot. On adzuna 184, usajobs 183, ashby 182, himalayas 181, themuse 180. On oracle_cloud and citi the oldest live posting is 43 days old, on google 44, on lever 61. That is a real structural difference: some boards carry postings for half a year, others carry nothing older than about six weeks. It is a statement about the extreme, not the average, because a minimum over a set of rows describes exactly one row. But a board whose oldest live posting is six months old is a board where a six-month-old posting is possible, and that is worth knowing before you read silence as rejection.

Why we are not publishing a death rate per board

We have per-source deactivation counts and we are deliberately not turning them into survival rates, because our own numbers say they cannot carry that weight. Three checks fail. First, jooble shows 1,994 deactivations against 249 live postings, a ratio of 801%, which is arithmetically impossible for a death rate and tells you the numerator and the denominator are measured over different populations at different times. Second, six sources holding 52,946 live postings between them (greenhouse, oracle_cloud, ashby, google, apple, usajobs) show exactly zero deactivations, which is not a credible finding about the world. Third, 53,612 of the 53,626 deactivations we have ever recorded (99.97%) fall inside the trailing 30-day window, and 43,629 of them (81.4%) inside the trailing 7-day window, while only 657 (1.2%) fall in the last 24 hours. That is the signature of one discrete sweep, not of a steady rate. The underlying cause is known: our verifier fetches a posting's URL, and some boards answer an automated fetch with a rate-limit or a refusal rather than an honest 404. Being told to slow down is not evidence a job is gone. Any 'postings on board X die N% of the time' number built on this data would be measuring our crawler, not the labour market.

What actually died, and what we can say about it

99,483 postings in the corpus are inactive. 53,626 of those (53.9%) were deactivated by the URL verifier described above, so that slice is contaminated and we do not attribute it to the market. The remaining 45,857 went inactive by a different route: they stopped appearing in their source feed. Disappearing from the feed is the stronger signal, because it is the board itself no longer listing the job rather than our probe failing to reach it. We are not reporting a per-board breakdown of those 45,857 because the corpus has no deactivation timestamp column, so we cannot say when they went away or attribute them to a window. That is a gap in our instrumentation, not a finding, and we would rather name it than paper over it.

What to do with this

Apply early, and treat feed freshness as part of the listing. Half of the live corpus has been listed for over a month, so the marginal cost of waiting a week is real: you move from a small pile to a large one. When a posting is old and the board it sits on is one that carries six-month-old listings, silence is weak evidence about you. And if you are tracking applications yourself, record the date you first saw the posting rather than the date the board displays, because a board that re-scrapes daily can make a 90-day-old listing look new every morning.

How we measured it, and where the numbers come from

Every figure on this page except where noted is a full-corpus count taken from a single snapshot of the jobs table at 2026-08-13 03:22 UTC: 339,289 rows, 239,806 flagged active, 99,483 inactive. We established that these are full-corpus rather than sample figures two ways: the per-source active counts sum to exactly 239,806, matching the independent total, and the per-source deactivation counts sum to exactly 53,612, matching the independent 30-day total. Age is measured from first-seen, when the posting entered our index, which tracks but does not equal the employer's posting date. Feed cadence is measured from last-seen, the most recent time we observed the row in its source feed. One caveat on the stale-feed finding: we measure the maximum last-seen across a source's active rows, so a feed reads as fresh if any single row was re-seen recently. That makes this a floor. A feed we call stale is certainly stale; a feed we call fresh may still be partially stale underneath.

Frequently asked questions

Related Data

Guides

• Why You're Not Getting Interviews →
• How Many Jobs Should You Apply To? →
• Hidden Job Markets →
• Job Market Trends →