The Hiring Game: Who Lies Better?.

LinkedIn is being investigated over alleged “ghost jobs”. Workday is defending a major case involving automated hiring systems. Stanford researchers are documenting what they call “algorithmic monoculture” in recruitment.

Different companies, different recruiters, different job descriptions, yet increasingly the same technological layer sits between the candidate and the person who might actually hire them. That matters more than it may seem.

Since September last year, I have submitted over a thousand job applications. From those, I received roughly 12 to 15 actual replies. Around eight progressed far enough for a meeting to be scheduled, yet nobody showed up. The rest were rejections. For a long time, the obvious interpretation of numbers like these was simple: hundreds of companies looked at a candidate and independently concluded that the candidate was not a good fit.

I am no longer convinced that this is a reasonable assumption.

When large numbers of employers use the same ATS platforms, the same screening vendors, similar ranking models and increasingly similar automated recruitment pipelines, hundreds of applications do not necessarily represent hundreds of independent evaluations. They may represent the same assumptions being applied repeatedly by a relatively small number of systems.

That brings us to a deeper problem with recruitment itself.

A CV should not exist to please an employer, a recruiter or an ATS. A CV should describe a person’s professional reality: what they know, what they have done, what they have built, what they have learned, what responsibilities they carried and how their career developed. Some degree of presentation is obviously reasonable. If I am applying for a distributed systems role, highlighting my distributed systems experience makes sense. If I am applying for a networking role, the networking part of my career deserves more visibility. But that is editing. It is not rewriting the person. The expectation that candidates should maintain several heavily “tailored” CVs, rewriting experience and terminology to mirror every individual job description, should already tell us that something has gone wrong. At that point, the CV stops being a professional record and starts becoming SEO for recruitment software. Candidates learn to optimize CVs for ATS systems. Employers optimize job descriptions for search engines and applicant conversion. Algorithms interpret both through keyword matching, ranking and scoring. Recruiters increasingly see only the candidates who survive those filters.

And then companies complain that candidates exaggerate their experience.

The system is actively rewarding exactly that behaviour. This is also where I believe a very clear line needs to exist around the use of AI in hiring. AI can be extremely useful in recruitment. It can extract skills, normalize job titles, summarize experience, identify relevant projects, highlight missing requirements and help recruiters understand how a candidate’s background relates to a role.

That is decision support.

But AI should never become the decision maker, directly or indirectly. If a recruiter only sees candidates that an algorithm has already decided are worth seeing, then the claim that “a human makes the final decision” becomes largely cosmetic. The important decision has already happened upstream.

Consider a simple example.

A job asks for ten years of Java experience.

A candidate has five years of Java, but also twenty years of software engineering experience across distributed systems, C/C++, PHP, networking, architecture, databases, infrastructure and high-load systems.

A competent hiring manager may look at that background and think: five years of Java and two decades of engineering experience? That is probably worth a conversation.

A simplistic screening system may see only five years of Java against a requirement of ten.

Fail.

A badly modelled representation of human capability is enough. And experienced engineers may be especially vulnerable to this kind of simplification. The longer and more varied someone’s career becomes, the harder it becomes to compress that experience into a clean set of keywords and numerical thresholds. A junior candidate may match Java, Spring, AWS, Kafka and Docker perfectly. Five boxes checked. Another candidate may have spent decades designing networks, building distributed systems, migrating monoliths, handling production failures, writing software in several languages, designing protocols, operating infrastructure and leading engineering teams.

And somewhere in the pipeline, the decisive question may still become: “How many years of Kubernetes?”

That is not meaningful engineering evaluation. It is data reduction pretending to be judgment.

The healthier model seems remarkably simple to me.

A CV should tell the truth about the candidate.

A job description should tell the truth about the role.

AI can compare the two, organize the information, identify differences and help people understand where the match is strong or weak.

Then a human should decide whether those differences actually matter.

Instead, we seem to be creating a recruitment ecosystem that increasingly rewards people for being good at applying for jobs rather than being good at doing them.

Those are very different skills.

And for an industry that claims to be constantly searching for “problem solvers”, building a hiring system that teaches candidates how to game a classifier may be one of the more impressive problems we have managed to create for ourselves.