AI ghost job detection is the application of algorithmic analysis to identify job listings that are fake, stale, or not genuinely intended to result in a hire. With 27% of job listings estimated to be ghost jobs and 45% of HR professionals admitting to posting them, the scale of the problem demands a technology-driven solution. Manual detection — reading every listing carefully and making a gut judgment — doesn't scale when you're sending dozens of applications per week.
At Sovia, we've built a ghost job detection system that scores every listing using 19 weighted rules. This article explains the methodology — the types of signals we analyze, how they're combined, and why algorithmic detection catches patterns that humans miss.
Why Ghost Jobs Exist (And Why They're Hard to Spot)
Before diving into detection technology, it's worth understanding why ghost jobs are inherently difficult to identify. A ghost job looks exactly like a real one. It has a title, a description, requirements, and a company name. It's posted on a legitimate job board. It has an "Apply" button that works.
The difference is intent — and intent is invisible. A company posting a ghost job doesn't put a label on it saying "we're not actually hiring." The listing itself is often indistinguishable from a genuine opening at the surface level.
This is why detection requires looking at patterns rather than individual features. No single signal definitively identifies a ghost job. But when multiple signals align — a listing that's been posted for 60 days, with a vague description, from a company that's had layoffs, with no salary range in a transparency-required state — the probability that it's a ghost job becomes very high.
The Signal Categories
Ghost job detection signals fall into six broad categories. Each captures a different dimension of listing authenticity.
1. Temporal Signals: How Long Has This Been Posted?
Time is the most intuitive ghost job signal. A real job listing is urgent — the company needs someone. They post it, screen candidates, interview, and make an offer. The typical hiring timeline for a real opening is 4-8 weeks. Listings that persist beyond this window are increasingly suspicious.
Temporal analysis tracks:
- Posting age — how many days since the listing first appeared. Listings over 30 days get flagged; over 60 days is a strong negative signal
- Repost detection — some companies close and repost the same listing to reset the "days posted" counter. We track this by matching job titles, descriptions, and company pairs. A listing reposted 3 times in 6 months is almost certainly not a genuine opening
- Last modification date — real listings get updated as requirements evolve. Ghost jobs are posted once and forgotten
Temporal signals alone aren't definitive — some legitimate roles (very senior positions, niche specialties) do take months to fill. That's why this is one category among six, not the sole determination.
2. Description Analysis: What Does the Listing Actually Say?
The language of a job description reveals more than most people realize. Real listings tend to be specific because the hiring manager knows exactly what they need. Ghost jobs tend to be vague because nobody is actually planning to evaluate candidates against these requirements.
Description analysis looks at:
- Specificity score — does the listing mention specific tools, technologies, or methodologies? Or does it use generic phrases like "excellent communication skills" and "team player" without concrete context?
- Requirements realism — are the requirements internally consistent? A "junior" role requiring 7+ years of experience is a signal. A role requiring proficiency in 15 different technologies is a signal
- Copy-paste detection — generic descriptions that appear across multiple companies or roles suggest templated listings not written for a specific opening
- Responsibility clarity — real listings describe what you'll actually do day-to-day. Ghost jobs tend to describe the team or department without specifying your role within it
We use natural language processing to quantify these signals rather than relying on keyword matching. A listing that says "you will build data pipelines using Apache Spark and dbt, processing 2TB daily from our PostgreSQL warehouse" scores very differently from one that says "you will work with data to help drive business decisions."
3. Salary Transparency Signals
Salary transparency laws have created a powerful new signal for ghost job detection. As of 2026, more than a dozen US states and several major cities require salary ranges on job listings. Companies that comply with this requirement are signaling organizational maturity and genuine hiring intent. Companies that don't comply are either careless about compliance (bad sign) or deliberately opaque (worse sign).
Salary analysis tracks:
- Presence of salary range — is a range listed at all? In jurisdictions where it's required, absence is a compliance failure and a negative signal
- Range width — a $50,000-$180,000 range is technically compliant but functionally meaningless. Excessively wide ranges suggest the company hasn't actually scoped the role, which correlates with ghost job behavior
- Market alignment — does the stated range match market rates for similar roles in the same location? Significantly below-market ranges may indicate a performative listing rather than a genuine one
4. Company Health Signals
The company behind the listing provides context that the listing itself can't. A company's recent behavior — hiring patterns, layoffs, growth trajectory — tells you something about whether their open listings are genuine.
- Recent layoffs — companies that laid off 15% of their workforce last month but still have 200 open listings are probably not filling all of those positions. Some may be maintained for investor optics
- Hiring velocity — is the company actually making hires? If they have 50 open listings but their LinkedIn shows zero new employee announcements in 3 months, something doesn't add up
- Listing volume relative to company size — a 50-person startup with 120 open listings is suspicious. That ratio suggests listing accumulation rather than active hiring
- Glassdoor and review signals — employee reviews mentioning "fake postings," "roles that don't exist," or "interviewed but they never planned to hire" are direct evidence of ghost job behavior
5. Application Response Tracking
This is where aggregate data becomes powerful. When thousands of users apply to jobs through a platform, the platform can track which listings generate responses and which don't. Over time, patterns emerge.
- Response rate by listing — if 200 people applied to a listing and zero received any response (not even a rejection), that's a strong ghost signal
- Response rate by company — companies with historically low response rates across all their listings may have systematic ghost job behavior
- Time-to-response — real listings generate responses within 1-3 weeks. Listings with no responses after 4+ weeks are increasingly suspect
This category improves with scale — the more applications tracked, the more accurate the response patterns become.
6. Structural and Metadata Signals
The technical properties of a listing — separate from its content — carry information:
- Duplicate detection — the same listing posted on 10+ job boards simultaneously with identical text may be a spray-and-pray approach common in ghost job posting
- Application mechanism — listings that redirect to a "general interest" form rather than a role-specific application are often talent pipeline collection rather than genuine hiring
- Required information asymmetry — listings that ask for extensive personal information (SSN, date of birth) before even a phone screen may be data collection operations rather than real jobs
The 19-Rule Scoring System
Sovia's ghost score combines 19 distinct rules across these six categories into a single probability score. Each rule is weighted based on its predictive power, determined through analysis of labeled training data (listings with known outcomes — real hires vs. confirmed ghost jobs).
The weighting isn't equal. Temporal signals and response tracking tend to carry more weight than description analysis alone, because they're harder to fake. A company can write a convincing description for a ghost job, but they can't fake a recent hire or manufacture response patterns.
The final score is expressed as a probability: how likely is this listing to be a ghost job? This is deliberate — binary classification (ghost/not ghost) would be overconfident. A listing with a 73% ghost score isn't definitely fake, but it means you should think carefully before spending 45 minutes on that application when there are 27%-score alternatives available.
How Scores Are Presented
Rather than overwhelming users with 19 individual signal values, the Ghost Score is presented as a simple, actionable indicator:
- Low risk (0-30%) — strong indicators of a genuine opening. Apply with confidence
- Medium risk (31-60%) — some concerning signals but not conclusive. Apply if the role is a strong match, but don't spend hours on customization
- High risk (61-100%) — multiple ghost job signals present. Consider skipping unless this is your dream company and role
What AI Detection Catches That Humans Miss
An experienced job seeker develops intuition about ghost jobs over time. They learn to check the posting date, be suspicious of vague descriptions, and avoid companies with bad reputations. But human detection has fundamental limitations that algorithmic detection solves:
Pattern Recognition at Scale
A human can evaluate one listing at a time. An algorithm can compare a listing against thousands of similar listings simultaneously. Is this description unusually vague compared to other Software Engineer listings at similar companies? A human would need to read hundreds of listings to develop that baseline. An algorithm has already processed them.
Cross-Signal Correlation
Humans tend to evaluate signals independently. "The posting date looks fine" + "the description seems okay" + "the company is well-known" = "probably real." But algorithms can detect subtle correlations: this combination of posting age + description vagueness + salary range width + company layoff history creates a pattern that individually benign signals don't reveal.
Historical Memory
An algorithm remembers that this exact listing was posted, closed, and reposted three times in the last year. A human seeing it for the first time has no way to know that. Historical tracking turns a single data point into a time series, which is dramatically more informative.
Consistency
Human judgment varies with mood, fatigue, and desperation. On day 60 of a job search, your standards for "this listing looks legit" drop significantly. An algorithm evaluates with the same rigor on application 1 and application 500.
The Limitations (Honest Assessment)
Ghost job detection isn't perfect, and claiming otherwise would be dishonest. Here are the genuine limitations:
- New companies — without historical data, detection relies more heavily on description and salary signals, which are less reliable
- Well-crafted ghost jobs — a company that deliberately writes a specific, realistic description and posts it with a competitive salary will score low on ghost probability, even if the listing is fake. Intent is hard to detect when the execution is professional
- Niche roles — very specialized positions may legitimately take months to fill, triggering temporal signals that don't actually indicate ghost behavior
- Data freshness — company health signals (layoffs, hiring velocity) lag real-time events. A company that decided yesterday to freeze hiring won't show that signal in the data yet
These limitations are why the system produces probability scores rather than binary verdicts. A 45% ghost score doesn't mean "don't apply" — it means "be aware that there are moderate risk signals, and factor that into your time allocation."
The Future of Ghost Job Detection
The technology is evolving rapidly. Several directions will improve detection accuracy over the next few years:
- Federated response data — as more job seekers use platforms that track application outcomes, the aggregate data on which listings produce responses becomes richer and more reliable
- Real-time company signals — integrating hiring announcements, SEC filings (for public companies), and funding data provides more current company health information
- Regulatory pressure — legislation targeting ghost jobs (several states are considering it) will create new compliance signals and potentially required disclosures
- Cross-platform correlation — linking the same listing across multiple job boards allows for repost detection and application volume analysis that single-platform views can't provide
Why This Matters
Ghost jobs aren't a minor nuisance. When 27% of listings are fake and the average job seeker sends 294 applications, roughly 80 of those applications are wasted on listings that were never real. At 30-45 minutes per application, that's 40-60 hours of wasted labor. For OPT students, it's wasted visa days. For single parents, it's time away from their children spent on a mirage. For anyone experiencing job search burnout, it's the difference between manageable stress and breaking point.
Ghost job detection technology doesn't eliminate the problem — only regulation and employer accountability can do that. But it gives job seekers something they've never had before: information asymmetry in their favor. For the first time, candidates can see what companies don't want them to see — that some of these listings were never real.
The system is still broken. But at least now there's a way to see through it.
Want to see ghost scores in action? Try Sovia — the Chrome extension that scores every listing you see, auto-applies to verified-real jobs, and proves every submission with screenshots. 19 signals. Zero ghost job tolerance.