What happened
- A paper published on arXiv on September 24 models what happens in a labor market when AI tools make it easy to generate and tailor application materials.
- The formal argument: the easier it is to produce a cover letter or résumé tailored to the job posting, the less that document says about whether the person fits the role.
- The company’s response, according to the model, is rational and predictable: if the materials lose signal, it relies more on the only thing that remains observable and cheap to verify, which is prior experience.
- Among the four types of applicant the authors define, the most exposed is the one who has no experience but does fit: they lose precisely the individual information that set them apart from the rest of the inexperienced applicants.
- When reviewing applications is costly, the model predicts selection failures: the company reviews only the experienced applicants, or reviews no one at all.
Why it matters
- It’s the theoretical counterpart to something hiring teams in Chile already report: more applications are coming in, all similar, all correct. The paper names the mechanism and explains why the filter tightens on its own.
- The effect falls on people changing industries, people returning to the market after a break and people coming out of a reskilling program. Those are exactly the profiles public employment policies try to bring back into work.
- There’s a consequence the paper doesn’t spell out: if the company replaces reading with a cheap intermediate test (which is what the model predicts), the cost of applying shifts to the applicant in the form of unpaid time. The friction doesn’t disappear; it changes pockets.
The number
The model defines four types of applicant, crossing prior experience with actual fit. Only one loses in every scenario evaluated.
Context
LinkedIn’s data on the AI labor market already showed a gap between the postings published and the profiles recruiters manage to review. This paper describes the mechanism that widens it.
What’s next
- No timelines announced for empirical validation of the model with real market data.
- The paper is in version 1 and doesn’t state peer review.
Bottom line
The same platform that found a gender gap in AI jobs now distributes the tools that write the applications. The model doesn’t say who wins from that; it says who stops being seen.
Sources
Edited by Rodrigo Cornejo. How we select and verify: who writes these notes.
