One in three tech jobs is still up 60 days after posting

Across 13,434 tech listings posted to Greenhouse, Lever, Ashby, and other major ATSes between February 28 and April 28, 2026, one in three was still live 60 days after first appearing. One in eight was still live after 90 days. We tracked every listing continuously through June 28, 2026.

The problem concentrates in the parts of tech where labor demand is supposedly hottest. Hardware engineering postings stayed live at 69.2% past day 60. Embedded engineering: 67.0%. AI-adjacent roles like ML engineering and data science: 49.5% each. Backend engineering, by contrast, ghosted at 42.0%, and most sales and operations roles cycled out at 20-37%.

Onsite roles ghosted at nearly twice the rate of remote (44.9% vs 26.5% at 60 days). Pay transparency laws made no measurable dent: listings with disclosed salary ranges ghosted at 34.1%, those without at 36.6%.

The full per-role-family breakdown is below. The named company-level dataset ships in the downloadable CSV companion.

The headline number

Of 13,434 tech listings posted to supported ATSes between February 28 and April 28, 2026, 33.4% were still active 60 days later. 12.5% were still active at 90 days. The 60-day mark sits one full cycle past the 2025 average tech time-to-hire of 68 days (Huntly). The 90-day mark is past virtually every published benchmark.

Still live at 60 days

33.4%

Still live at 90 days

12.5%

For comparison, the 2024 Greenhouse State of Job Hunting report put the in-quarter ghost-job rate at 18-22%. Our number is roughly 1.6x higher because we measure listing lifespan across the full posting cycle, not whether a listing is active on a given snapshot day. Both numbers describe the same phenomenon from different angles. Neither is wrong.

What we measured

The dataset is every tech job listing Hotfix ingested from supported ATSes (Greenhouse, Lever, Ashby, and others) with a posted_at between February 28 and April 28, 2026. February 28 is the start of our deactivated_at tracking. April 28 is the latest posting date that still allows a full 60-day observation window before our query cutoff of June 28.

For each listing we computed lifespan_days as the number of days between posted_at and either deactivated_at (the date our re-crawl detected the listing had been removed) or the query date for listings still active. A listing counts as still live at 60 days if it had not deactivated by day 60 of its lifespan.

We then bucketed the cohort by role family (a normalization Hotfix applies during ingestion based on title and description), by work type (remote, hybrid, onsite, as the ATS itself classifies the listing), and by whether the posting disclosed a salary range.

The dataset behind this analysis is available as a CSV. (Casey: link goes here.) It includes per-listing fields, role family assignments, and the company-level long-running poster table.

One thing this methodology does not do. It does not distinguish between a role that was filled and never taken down, and a role that was never going to fill. From a job seeker's perspective the two cases are the same: they applied to a listing that was not actively under consideration. From a hiring team's perspective the implication differs. We disclose this rather than infer intent.

Supporting findings

Hardware and embedded engineering ghost the most. Of the 159 hardware engineering listings we tracked, 69.2% were still live 60 days after posting. Of 112 embedded engineering postings, 67.0% were still live at the same mark. Both are well past every other category in the dataset.

These role types do legitimately take longer to fill than mainstream software roles. Senior hardware engineers are scarce, security clearances and clean-room experience are gating factors for some employers, and the interview loops involve more stages. We are not asserting that every hardware listing past day 60 is a ghost. We are noting that hardware sits at the top of the distribution and that the gap between hardware and the next tier (fullstack engineering at 54.2%, solutions architecture at 50.0%) is large.

For comparison, backend engineering ghosted at 42.0%, frontend engineering at 31.8%, and devops at 49.0%. The engineering bucket as it appears in most labor reports averages over a 30-point spread.

Hardware engineering

69.2%

Embedded engineering

67.0%

Fullstack engineering

54.2%

Solutions architecture

50.0%

Devops engineering

49.0%

Security engineering

44.3%

Backend engineering

42.0%

Frontend engineering

31.8%

AI-adjacent roles cluster near the top of the table. ML engineering listings ghosted at 49.5% past 60 days. Data science ghosted at 49.5% as well. Both sit closer to the hardware-engineering tier than to mainstream software engineering.

This contradicts a common reading of the labor market in 2026. The dominant narrative is that AI talent is in such acute demand that listings flip over fast. Our data says the opposite: roles tagged ML, AI, or data science are among the slowest-cycling postings tech employers are running.

A reasonable hypothesis is that some share of these listings are evergreen pipeline-building for talent the company would hire if the right candidate appeared, rather than commitments to fill a specific role on a specific timeline. Our data does not prove that interpretation. The gap between the AI hiring narrative and the lifespan numbers is wide enough to flag regardless.

ML engineering

49.5%

Data science

49.5%

Devops engineering

49.0%

Backend engineering

42.0%

Data engineering

38.1%

Frontend engineering

31.8%

Onsite roles ghost at nearly twice the rate of remote. Onsite listings (per ATS classification) stayed live past 60 days at 44.9%. Hybrid: 35.0%. Remote: 26.5%.

The work type most aggressively pushed by return-to-office employers is also the work type whose listings turn over slowest. We do not assert causation. We do note that the asymmetry runs counter to the framing common in 2025-2026 that onsite work is where serious hiring happens and remote postings are noise. In our dataset, remote postings resolved fastest.

Onsite

44.9%

Hybrid

35.0%

Remote

26.5%

Pay transparency laws did not fix the ghost-job problem. Twenty-five US jurisdictions now require employers to disclose salary ranges on at least some job postings, including California, New York, Washington, Illinois, Colorado, and Maryland (GovDocs). The intent is to make hiring more transparent. The data shows transparency on pay has not produced transparency on intent.

Listings with disclosed salary ranges ghosted at 34.1%. Listings without disclosed ranges ghosted at 36.6%. The 2.5-point gap is within rounding distance of no effect.

This is worth naming because it complicates the policy story. The Truth in Job Advertising and Accountability Act, the federal bill introduced by Eric Thompson and profiled by CNBC in 2025, is one of several proposals aimed at making employers commit to filling the roles they post. Pay-disclosure laws were the prior generation's answer to a similar transparency gap. For the ghost-listing problem specifically, they did not move the number.

Salary disclosed

34.1%

Salary not disclosed

36.6%

The long tail is named in the data. A small group of well-funded AI infrastructure, defense-tech, and data-platform companies post listings that disproportionately stay live past 60 days. Several are running more than 20 active postings each with 60%+ still-live rates two months in.

We are not naming them in the post body. The full company-level table ships in the CSV. The decision to keep the named list in the CSV rather than the headline is deliberate: we want readers to look at their own employer or target employer with the same lens, not to absorb a list of 15 villains and stop there. The dataset is in the public release. The pattern is what matters.

What this means

For job seekers, the operational read is straightforward. If a posting is more than 60 days old, the prior probability that it is being actively filled is lower than the prior probability that it is not. That is even more true if the role is hardware, embedded, ML, or onsite. We are not telling people to stop applying. We are saying the math on time spent per application changes when a third of the listings being applied to are not in active hiring loops.

For employers, the implication is a credibility one. Greenhouse's 2024 number put in-quarter ghost-job rates at 18-22%. Our 60-day still-live rate of 33.4% is roughly 1.6x higher because it captures the same population from a different angle: lifespan over the full lifecycle rather than activity at a snapshot. The gap between them is the gap between a listing being worked and a listing simply not being taken down. Job seekers cannot distinguish those two states. They are spending real time on the second one.

For the market, the data argues that ghost listings concentrate where the labor narrative is loudest. The fields tech employers describe as their most acute hiring shortages (AI/ML, hardware, anything that requires onsite presence) are the fields where listings persist longest without resolution. The most defensible explanation is pipeline maintenance: keep the listing up to keep candidates flowing, hire when the right one appears. That is a legitimate posture for an employer. It is also the posture a job seeker has the least visibility into.

The policy debate is moving in the direction of forcing the issue. The Truth in Job Advertising and Accountability Act would require employers to commit to a posting window and a hiring outcome. The Congressional Research Service produced a primer on ghost job postings in 2024 (IF12977) framing the issue as a candidate-protection problem. Pay transparency laws, the most relevant precedent, did not move our number. Whatever the federal bill becomes, disclosure rules by themselves are unlikely to be sufficient.

Counterpoints

Three things worth saying about the limits of this analysis.

The 4-month tracking window is the floor on what we can say. Hotfix's deactivated_at field started populating on February 28, 2026. Anything before that we cannot include because we do not have the deactivation events. The cohort is one quarter long. We are presenting this as a baseline. Trend claims (this is up, this is down) will be possible after Q3 2026 numbers are in.

We cannot distinguish filled-but-not-removed from never-going-to-fill. A listing that closes via an ATS without our re-crawl picking up the deactivation is indistinguishable from a listing that is genuinely still active. For job seekers the two cases are functionally the same, since a hiring team that does not bother to deactivate filled roles is also a hiring team not responding to incoming applications for those roles. For employers the two cases are different. We disclose this.

Some role types take longer to fill. Hardware, executive, and security-clearance roles all have legitimately longer hiring cycles. The per-role-family table separates these out rather than averaging them in. Hardware sitting at the top of the table is partly a real signal about the hiring market and partly a signal about cycle length. Both interpretations are present in the data. We are not picking one for the reader.

Per-role-family table