A Correction: We Measured Freshness on a Sample of the Two Slowest Feeds

Age measured from when a posting first entered our index. The corrected bars are computed on all 257,534 active postings; the superseded bars are the 30,000-row sample this page originally used.

The first two rows are the feeds that dominated the old sample. workday adds essentially nothing on any given day and oracle_cloud is not much faster, while himalayas, which is 44.4% of the live corpus and was 3.5% of the sample, turns over three times as fast. The three rows showing 0.0% are feeds that have gone quiet rather than feeds that are slow; they are covered below.

Source: Seeker job corpus, 257,534 active job postings, snapshot 2026-08-16. seekerscore.com/insights/how-fresh-are-job-postings-2026

Correction, 2026-08-16

What this page claimed: 'Of 30,000 active postings: 2.6% first appeared in the last 24 hours, 9% within the last 7 days,' and 'The remaining 56.5% have been listed for more than a month,' measured across 'about 250,000 active listings across 47 job boards.' What is true, measured on all 257,534 active postings rather than a sample: 5.2% in the last 24 hours, 22.2% in the last 7 days, 46.1% over 30 days. The corpus draws from 20 sources, not 47. The 7-day figure was understated by 2.5x, which is the difference between a market where roughly one posting in eleven is new this week and one where it is closer to one in four and a half.

The cause, and the arithmetic that reproduces it

The 30,000 rows were not a random draw from the corpus. As our own metro-concentration page records, that cache was 52.5% workday and 29.2% oracle_cloud, 81.7% between them, against 22.0% of the corpus. Those are the two slowest-refreshing large feeds we carry: workday adds 0.1% of its postings in a given 24 hours and 14.7% in a week, and oracle_cloud 5.9% and 14.3%. Meanwhile himalayas, which turns over at 3.6% and 28.5% and is 44.4% of the live corpus, contributed 3.5% of the sample. Weighting just those two feeds' current refresh rates by their sample shares lands in the right neighbourhood rather than on the nose: about 1.8% for 24 hours against 2.6% published, and about 11.9% for 7 days against 9% published. That reconstruction covers 81.7% of the sample and is deliberately not tuned to fit, since the remaining 18.3% is unaccounted for and the sample composition was recorded three weeks after this page was published. It is offered as evidence of mechanism, not as a derivation: two slow feeds at four fifths of the sample is enough to produce a single-digit weekly figure, and no combination of the corpus's actual source mix is. The old numbers were not an arithmetic error. They were an accurate measurement of the wrong population.

The part that got worse, not better

One thing the original page could not have seen, and it cuts against the correction rather than for it. Six of our twenty sources have added nothing in over a week: adzuna, reed, usajobs, findwork, themuse and jooble all read 0.0% at both 24 hours and 7 days. Those six carry 27,046 active postings, 10.5% of the corpus, and they are dark because they exhausted a monthly API quota in early August rather than because the employers stopped hiring. So the corrected 22.2% is measured across a corpus where a tenth of the inventory is frozen and still being served. If those feeds were live the fresh share would be higher still, which makes 22.2% a floor. Readers are entitled to know that the pool contains listings we have not re-checked since the first week of August.

What stands

The page's central point survives and is worth restating: active does not mean new. 46.1% of live postings have been listed for more than a month, so the pool you apply into is still dominated by roles that have been open for weeks. The explanations offered for why old postings persist are unchanged and were never sample-dependent: genuinely hard-to-fill roles, evergreen pipelines, filled listings that are never taken down, and real roles that have already accumulated hundreds of applicants. What changes is the magnitude of the fresh slice, and therefore how much of a job search should be spent hunting for it.

A better measure exists, and it is less flattering

Everything on this page, before and after correction, dates a posting from when it entered our index. That is the board's clock, not the employer's. Most feeds also publish the date the employer says the role opened, and on that clock only 10.3% of active postings are under a week old rather than 22.2%, with a median age of 28 days. Our companion page at /insights/job-posting-age-two-clocks-2026 measures both on the same postings. The honest summary is that this page's corrected 22.2% is the right answer to the question 'how new is this to the board' and the wrong answer to the question a job seeker is actually asking.

How we measured it (and what we did not claim)

Population: all 257,534 active postings on 2026-08-16, queried directly rather than sampled, which is the single change that produced the correction. Age is the interval between now and the posting's first appearance in our index, in the same three bands the original page used, so the old and new figures are directly comparable. Per-source rates in the table are computed inside each source over its own active count. We report first-seen age rather than last-updated for the reason the original page gave and which remains correct: a listing re-scraped daily would otherwise look new every day. We are NOT claiming that a posting older than 30 days is dead or fake, and we are NOT claiming a death rate per board. The original 30,000-row figures from 2026-07-23 are quoted unchanged above so the error stays auditable.

Frequently asked questions

Related Data

Guides

• Why You're Not Getting Interviews →
• How Many Jobs Should You Apply To? →