When Eric Thompson lost his job last year, he expected to follow a familiar playbook: polish his resume, search job boards, submit applications, and wait for responses. Instead, he applied for countless roles that may have been fake. Thompson was so frustrated with these “ghost jobs” that he’s now pushing federal legislation to ban them.
Thompson isn’t alone. His experience is increasingly common and reflects just one of the ways that artificial intelligence is changing job hunting. AI is transforming job hunting from a manual, labor-intensive process into a high-volume but increasingly impersonal numbers game.
Finally, a reason to check your email.
Sign up for our free newsletter today.
The evidence is everywhere. Software engineering and computer science graduates, once thought to be shoo-ins for stable employment, are struggling to find work. Early-career workers in AI-exposed occupations are facing declining employment prospects. Capturing the broader mood, writer Kelsey Piper described job hunting as feeling increasingly like a lottery.
That sentiment rings true for millions of workers. But the economics tell a more complex story.
A core concept in labor economics is the Diamond-Mortensen-Pissarides framework, which models how unemployed people find work. It is rooted in the idea that finding a job is a matching process, one plagued by frictions, information asymmetries, and search costs for both workers and employers.
We can apply this framework to the time before and after the development of large language models (LLMs). Before the release of ChatGPT, employers had state-of-the-art applicant tracking systems to extract basic information from resumes like contact details, work history, education, and skills. They used this data to screen resumes against job requirements. For employers, this process was largely automatic.
But workers in this pre-LLM world faced considerable constraints. They were forced to tailor resumes manually for each application and upload the same information again and again into clunky portals. They couldn’t easily automate their search or optimize submissions at scale. This situation felt like a lottery, too, where the odds were hidden and the effort rarely paid off.
The release of ChatGPT changed these dynamics by enabling job seekers to file applications en masse. This has had several effects on labor markets.
First, it has become harder to know who is genuinely looking to hire and who is seriously looking for work. AI lets workers file applications in droves and allows employers more easily to post job openings. The growing number of low-commitment employers in the market would explain the apparent proliferation of ghost jobs. From a job seeker’s perspective, there’s little practical difference between ghost jobs and postings from low-commitment employers. Both result in applications disappearing into the void.
AI has also made it harder for workers to differentiate themselves from other applicants. To so do in the future, job seekers will have to use harder-to-game differentiators, such as portfolio projects, networking connections, or referrals.
AI could drive market segmentation and create a bifurcated system in which job postings mediated by LLMs are high-volume and algorithmic, while those mediated by humans are relationship-based. Each segment would develop distinct wage and employment dynamics.
This model may explain the dwindling employment prospects of early-career workers in AI-exposed fields. Students and recent graduates entering the workforce cannot rely on harder-to-game differentiators, like referrals, since these prospective workers often lack the extensive work samples, professional networks, or proven track records that more experienced candidates can leverage.
One of the few papers to study the effects of AI on hiring suggests the dynamics are evolving. In “Generative AI and Labor Market Matching Efficiency,” economists Emma Wiles and John J. Horton found that employers using AI were 19 percent more likely to post a job and spent 44 percent less time writing the job description. The net effect of AI was that “job posts were . . . more generic and less informative to jobseekers.”
Importantly, the authors found no discernible increase in matches, which means AI hasn’t improved hiring efficiency. It has created more noise in the system. They explained, “The lack of match formation was mostly due to marginal jobs being posted by employers with lower hiring intent”—in other words, low-commitment employers are entering the market.
Eric Thompson’s proposed legislation to ban ghost jobs addresses a symptom, not the underlying cause of workers’ distress. Rather than trying to restore the pre-AI status quo, policymakers should focus on improving market transparency for both workers and employers by developing new frameworks that help both sides of the labor market signal genuine intent. While some employers already offer hiring rates and timelines, third-party platforms could standardize and expand this practice by aggregating verified hiring data. The groundwork is underway, but making it a standard practice will take time.
The feeling of playing a “lottery” in the job market isn’t inevitable. It’s the result of a matching system that hasn’t yet adapted to new technologies.
Will Rinehart (@willrinehart) is a senior fellow at the American Enterprise Institute, where he focuses on the political economy of technology and innovation.
Photo: WANAN YOSSINGKUM / iStock / Getty Images Plus