How AI is reshaping hiring

The story everyone feels in their bones is simple. Job hunting has started to look like a lottery. Postings feel disposable. Applications vanish into silence. Candidates blast out submissions with a few clicks. Teams post roles that sound real but are not tied to a budget or a decision window. The label for this is familiar by now. People call them ghost jobs. What we are seeing is not a moral failure. It is a systems failure. And the culprit is a hiring market that switched from friction constrained to volume optimized the moment modern AI tools met legacy applicant tracking.

Before large language models, employers already had automation on their side. Applicant tracking systems parsed resumes and filtered for keywords. The constraint sat mostly with candidates. Tailoring a resume took meaningful time. Portals were repetitive. That friction throttled supply. The pipeline looked thinner but the signal to noise ratio was higher. When ChatGPT arrived, the constraint moved. Candidates could generate clean cover letters, remix resumes for each role, and auto fill forms at scale. Employers could also produce more postings with less effort. The system started rewarding whoever created the most surface area, not whoever signaled the most intent.

Labor economists describe matching with the Diamond Mortensen Pissarides framework. It is a market with search frictions and asymmetric information. Remove friction without adding verified signal and you do not get better matches. You get more attempts with lower intent. That is exactly what the current market has produced. One recent study found that AI made it easier for employers to post jobs and faster to write descriptions, yet it did not increase matches. The posts trended more generic. The marginal postings came from lower intent employers. If you are a candidate, that feels like AI ghost jobs. If you are an operator, it looks like a pipeline stuffed with false positives.

False positives are the quiet cost center here. Headcount plans assume conversion rates from posted role to qualified pipeline to offer. When a team can publish ten look alike roles in a day, the top of funnel explodes. Recruiters now swim through thousands of near identical submissions that pass keyword screens. Coordinators schedule more screens that go nowhere. Hiring managers burn cycles on interviews that do not progress. The optics show productivity. The match rate does not. That gap erodes trust on both sides. Candidates grow cynical. Operators grow numb.

The early career market gets hit first and worst. New grads and career switchers have fewer hard to game differentiators. They do not have deep networks or shipped work. They are competing in the algorithmic lane with tools that make everyone sound equally diligent and equally generic. That is why a student can submit hundreds of applications and see only a handful of replies. It is not that they are unqualified. It is that the system has taught both sides to optimize the wrong variable. Volume looks like progress. It is not.

The fix is not nostalgia for a pre AI world. The fix is to rebuild the matching layer around intent and verifiable work, then rate limit what does not carry either. Start with employers. Every posting should carry three signals of commitment that are visible and auditable. First, an intent timestamp that states the decision window and the budget owner. Second, a historical hire rate at the company and at the team level that lets candidates price the odds. Third, a response service level that commits to a close of loop within a set number of days, even if the answer is no. When a role closes, the system should mark it closed, not leave it lingering to collect vanity metrics. If you want to deter low commitment posting, pair this with a posting bond that is refunded upon close with a decision, not upon publication.

Next, fix the candidate signal. Replace generic cover letters with lightweight proof of work tied to the role. If the job is front end, ask for a small repo or a code sandbox link. If it is customer success, ask for a mock response to a sample ticket. If it is product, ask for a three slide critique of a feature with a single improvement. Keep it small and scoped. The goal is not free labor. The goal is to make the signal expensive to fake and easy to evaluate. The right work sample reveals judgment, not just style. It also yields a better interview. You talk about decisions rather than biographies.

Then, rebalance the two lane labor market that is already emerging. The algorithmic lane is high volume and low intimacy. The relationship lane is referral heavy and human mediated. Wages and conversion rates will diverge across those lanes. Pretending they are one market only breeds frustration. Give candidates a way to move between them. Verified community referrals should carry structured weight in the funnel. Not an old boys club. A transparent ledger of who has worked with whom and in what capacity, pulled from shipped projects, open source contributions, client deliverables, or internships. Early career talent should not be shut out just because they lack a network. They should be able to earn standing through micro internships, scoped trials, and community validated projects that feed the same ledger.

Policy can help, but not with blunt bans. Outlawing ghost jobs attacks the symptom, not the root. What regulators can do is require disclosure of posting to hire conversion and enforce honest status labeling. Open, paused, closed. Add light penalties for stale or misleading postings. Encourage third party aggregation of verified hiring data so that platforms compete on transparency, not opacity. The goal is not to police intent one role at a time. The goal is to make intent a standard field in the hiring stack that markets can price.

Founders and hiring managers do not need to wait for regulators. They can stop burning cycles by publishing fewer, truer roles and by resisting the reflex to widen the funnel when the signal quality is poor. They can move work samples ahead of first calls for roles where portfolios predict performance. They can close loops even when the answer is negative. They can publish salary bands and decision timelines without waiting for legal to draft a manifesto. None of this is charity. It is brand protection and time protection. Teams that communicate clearly fill roles faster and churn less. Candidates who are treated like adults keep showing up at the next opening.

Candidates also have a play. Stop optimizing for application count. Optimize for evidence. One strong project that shows your thinking is worth more than twenty auto generated cover letters. Write a brief about what you built, why you chose it, and what you would change. Map it to the business problem the role cares about. Ask for a conversation around that artifact, not around your schooling. In a market full of generic noise, the person who can show how they make a decision stands out. If you are early in your career, find structured ways to earn that proof. Micro apprenticeships, scoped volunteer projects, or open source issues can build the track record that the relationship lane respects.

Platforms should align their incentives with match quality. Reward employers that close loops on time and that convert posted roles into actual hires. Down rank posters with low intent histories. Surface candidates who carry verified proof of work and validated referrals, not just keyword dense resumes. The technology exists. What is missing is the decision to optimize for matching rather than top of funnel optics. When a platform claims success based on application volume or registered users, it tells you what it values. Push them to value conversion and retention measured in accepted offers and tenure.

AI is not the villain. AI removed friction that had been masking structural laziness in how we hire. The market responded by turning up the volume. That is predictable. It is also fixable. The phrase AI ghost jobs names a real frustration, but it should point us to the underlying design flaw. The system rewards easy publication and easy application. It does not reward clear intent or credible work. Rebuild those incentives and the lottery feeling fades. Outcomes begin to look like decisions again, not dice rolls.

The next phase of hiring will belong to operators who treat matching like a product. It will belong to teams that measure truth instead of traffic. It will belong to candidates who can show how they think, not just how they format. Ghost jobs will not disappear overnight. They will matter less the moment the market starts to price intent. That is the shift worth building.