27.4% of LinkedIn Jobs Are Ghosts. Your Comp Benchmark Is Poisoned.

If your outbound playbook watches competitor job posts to time messages and set comp bands, a ResumeUp.AI study you probably scrolled past is quietly breaking it. The topline: 27.4% of US LinkedIn postings are likely ghost jobs, and 79% of the tech ghosts they identified were still live at the time of the study. That is not a candidate problem. That is a signal-integrity problem for anyone doing sourcing, prospecting, or benchmarking off public reqs.

The 27.4% number, and why sourcers should care more than candidates

Roughly one in four US LinkedIn postings has no hiring intent behind it, and in tech the concentration is worse: 40% of tech companies posted fake jobs in the past year. The candidate-facing coverage has framed this as a morale story. For a sourcer, it is a data-quality story, and it invalidates the two things public JDs are supposed to do: tell you who is scaling, and tell you what they pay.

Here is what the converging surveys actually say:

• 27.4% of US LinkedIn postings are likely ghosts (ResumeUp.AI, 2025).
• 79% of identified tech ghost listings were still active at time of study.
• 40% of tech companies posted fake jobs in the past year (ResumeBuilder).
• 81% of recruiters admit to posting positions with no intent to fill (Clarify Capital, January 2025).
• 45% of HR professionals post ghost jobs "regularly"; another 48% post them "occasionally".

The 27.4% is the headline. The 79% is the one that kills your workflow. If ghosts came down within a week, competitor-JD monitoring would still function as a leading indicator. Because most of them stay up, "Company X posted 12 backend roles, they are scaling, poach the team now" is a false positive most of the time you fire it.

79%

Ghost jobs are a comp-benchmarking weapon, not just a candidate scam

The most under-discussed motive for posting fake reqs is calibrating internal pay bands off applicant expectations. Which means when you read a competitor's posted band to set your own offer, you may be reading a decoy the competitor posted specifically to reverse-engineer bands like yours.

JobIntel's 2026 stats hub lists "benchmarking salaries, using applicant expectations to calibrate internal pay bands" as a documented reason companies keep reqs open. Read that twice. The comp band on a ghost req is not a mistake, it is instrumentation. The company is running a survey with your candidates as the panel.

This has second-order effects for anyone using scraped comp data:

1. Aggregators quoting "median posted comp for Senior Backend in NYC" are averaging real reqs with decoys.
2. Offer negotiations that anchor on "well, competitor X posts $Y" are anchoring on a number the competitor may have posted to see what candidates would accept.
3. Any internal band built by scraping public JDs inherits the noise of whoever is posting the most ghosts in your segment.

You are not benchmarking against reality. You are benchmarking against a hall of mirrors, and the mirrors are angled by the people you are trying to out-compete for talent.

Mid-market (1k to 5k employees) is the noisiest slice

Companies with 1,001 to 5,000 employees are the worst offenders, with ghost jobs at nearly 25% of postings, per the ResumeUp.AI dataset. That is precisely the segment most boutique search firms and internal TA teams at growth-stage startups benchmark against.

Enterprise sourcers watching FAANG postings get cleaner data because those companies get audited harder and the reqs churn faster. Startup sourcers watching Series-D peers get cleaner data because those companies are too small to run comp instrumentation as a side project. The middle band, Series-E through pre-IPO, is where the ghost rate is highest, the comp implications are largest, and the outbound playbooks lean hardest on public JDs.

Geographic distortion inside "West Coast tech"

LA's 30.5% ghost rate versus SF's 26.0% is a 4.5-point spread that any location-weighted comp model absorbs without noticing. Philadelphia (30.1%), Indianapolis (27.8%), and New York (26.7%) round out the top five US metros. If your benchmarking tool treats LA and SF postings as one "West Coast" bucket, you are over-indexed on speculative reqs from a market that is not hiring at the rate its postings claim.

That last row is the macro tell. Since the start of 2024, reported job openings have outpaced actual hires by more than 2 million per month. In February 2026 alone, BLS reported 6.9 million openings against 4.8 million hires. Roughly 30% of reported "demand" never converts to a hire in the month it is posted. Some of that is legitimate pipeline lag. A material share is postings that were never going to close.

The supply-side truth that makes ghost inflation visible

Public postings tell you what companies claim to want. Profile data tells you what actually exists to hire. The gap between the two is the ghost signal without needing to identify individual fake reqs.

Two numbers from Refolk's index of US professional profiles make this vivid:

• 349,377 current "Software Engineer" profiles in the US, concentrated at Google, Figma, Microsoft, LinkedIn, Ashby, and Glean.
• 27,377 US software engineers with AI or ML skills on their profile, concentrated at Meta, Google DeepMind, LanceDB, and Ambience Healthcare.

That second number is roughly 7.8% of the total software engineer population. Now compare it against CompTIA's 2025 finding that 41% of US tech job postings now require or focus on AI skills. Postings claim AI at a rate more than five times higher than the actual supply of engineers who list those skills.

Some of that gap is legitimate skills shortage. Some of it is job description inflation. And some of it is postings written by people who do not have a specific human in mind, which is another way to describe a ghost.

Postings claim AI at five times the rate engineers exist who can do it. That gap is a ghost signal without needing to catch a single fake req.

This is the exact gap Refolk closes for sourcers who have given up on JD-derived intent. You describe the person you want in plain English ("Series-C infra engineers in NYC with production Rust and Postgres internals experience") and get a ranked shortlist built from the profiles that actually exist, not from postings that claim to want them.

The replacement signal stack

Stop treating LinkedIn job posts as leading indicators. Use signals that are either regulator-mandated or public behavioral exhaust, neither of which can be spoofed to game a benchmark. Here is the stack that survives ghost inflation:

1. WARN Act filings

Legally required layoff notices are the cleanest contra-signal to a posting spike. A company posting 40 new reqs while filing a WARN for 300 people is not scaling, it is reshuffling. WARN filings publish titles, effective dates, and site addresses in most states, giving you numbers you can plan against instead of a careers-page headcount claim.

2. H-1B LCA disclosures

The Labor Condition Application is a public record of a specific role, at a specific worksite, with a specific wage the employer has legally committed to pay. It is the closest thing to an audited comp benchmark that exists in the US, and it is not something a company files to fish for salary data.

3. GitHub commit velocity on public repos

Real engineering headcount leaves fingerprints. A company whose infra repos have not seen a commit from a new author in six months is not building a new infra team, regardless of what their careers page says. This is behavioral exhaust that no comms team can fake without actually hiring the engineers.

4. Insider LinkedIn employee posts

Individual employee posts are harder to fake at scale than corporate JDs. A wave of "we are hiring on my team, DM me" posts from actual engineers at a company is a stronger signal than the same roles appearing on the corporate careers page. It also gives you a warm intro path the JD does not.

What this means for the 112,092 recruiters in the index

The population most directly harmed by corrupted posting data is US recruiters and technical recruiters, of which Refolk's index holds 112,092 profiles, concentrated at Experis, K2 Partnering Solutions, and Robert Half. If you are one of them, three practical changes cost nothing to make this quarter:

1. Stop citing competitor posted bands in offer conversations. Cite LCA-disclosed wages instead. They are legally binding and publicly searchable.
2. Weight WARN filings inversely to posting spikes. A company posting a lot while filing WARNs is a raid target, not a competitor.
3. Track author diversity on public GitHub repos for the 20 companies you care about. New committer names showing up is a real hiring signal. New job posts on their careers page are not.

Dice's 2025 survey found that 47% of tech professionals are actively job hunting, up from 29% the year before. The supply side has never been more liquid. Accurate demand-side signals have never been more valuable, and public JDs have never been a worse proxy for them.

The regulatory clock is running

Ontario's ghost-job legislation takes effect January 2026, requiring employers to inform applicants about candidacy status and effectively banning ghost postings in the province. It is the first jurisdiction to do it. It will not be the last.

US enforcement will lag Ontario by years. That means the poisoning of comp benchmarks and hiring-intent signals on LinkedIn is a durable feature of the US market for the rest of this cycle, not a temporary distortion. The sourcers who adapt fastest are the ones who stop treating public postings as ground truth and start treating them as marketing copy from an adversarial counterparty. In the case of comp-benchmarking ghosts, that is literally what they are.

FAQ

How do I tell if a specific LinkedIn posting is a ghost?

There is no perfect single-post test, but the signal stack rules out most false positives. Cross-reference the posting against the company's LCA filings for the same title in the last 12 months, their WARN status, GitHub commit activity if the role is technical, and whether any current employee has posted about hiring on that team. A req with none of those corroborating signals, sitting live for more than 60 days, is far more likely to be instrumentation than an active search.

If competitor JDs are unreliable for comp, what should I benchmark against?

H-1B LCA disclosures are the strongest public comp source in the US because the wage is a legal commitment tied to a specific worksite and title. For roles not covered by LCAs, offer-stage data from your own recent closes beats any scraped posted band. Anchoring on competitor postings means anchoring on numbers a share of which were placed specifically to move yours.

Does the 27.4% ghost rate apply outside LinkedIn?

The ResumeUp.AI figure is LinkedIn-specific, and no comparable audit exists for Indeed, ZipRecruiter, Glassdoor, or ATS-fed pages on Workday, Greenhouse, Lever, and iCIMS. That is a blind spot, not a clean bill of health. The recruiter incentives driving ghost postings on LinkedIn - comp instrumentation, pipeline-building, appearing to grow - apply equally to those surfaces, so assume the base rate is similar until someone audits it.

Are mid-market companies really the worst offenders?

Yes, per the ResumeUp.AI dataset. Companies with 1,001 to 5,000 employees post ghosts at nearly 25% of their listings, higher than either enterprise or small-company rates. That is exactly the growth-stage band boutique search firms and internal TA teams at Series-E and pre-IPO startups benchmark against most heavily, which is why the practical impact of ghost inflation lands hardest on that segment.