Job search is an information problem

For work, I build systems that read the open web and turn it into labor-market data - crawl the pages, extract the structure, enrich the records until they hold together as something you can trust. Spend enough time on that side and the job search stops looking like a test of character. It starts looking like what it is: a market where one side has terrible information and gets blamed for the results.

The standard story about a hard search is personal. You are not motivated enough, not polished enough, not applying enough. A whole advice industry runs on that story, because motivation and polish are things you can sell. But when you watch the raw material of the market go by - the postings themselves, at scale - the story falls apart. The pain is mostly upstream of the seeker, in the information they are handed.

Start with the supply. A real opening does not appear once. It gets posted on the employer's site, then picked up by aggregators, then reposted by boards that scrape the aggregators. By the labor-data industry's own reckoning, up to 80 percent of collected postings are duplicates. Some fraction of listings are ghost jobs - posted with no real intent to hire; platform data puts that at roughly one in five in the US. And reposted copies go stale: the role fills, the copies stay up. A board shows the seeker all of it flattened into one list, every row wearing the same face, as if each one were equally real.

Here is the failure at its smallest. A seeker finds a role on a board and spends an evening on the application. The listing was scraped from another board, which scraped it from the employer weeks earlier. The role is filled. Nothing comes back. The seeker concludes something about themselves. An information failure, billed as a personal one. Multiply that by every listing in the stack and you have most of what the search feels like.

Then the folk explanations. The most repeated one says an algorithm auto-rejects three quarters of resumes before a human ever looks. People have traced that number back; it leads to a resume-optimization vendor that went out of business in 2013 without ever publishing a method, and recruiters surveyed about their own systems overwhelmingly say they do not auto-reject on content. What is actually happening is simpler and worse: too many applications for the humans on the other side to read. The myth matters because it points effort in the wrong direction - tuning keywords to beat a robot that mostly is not there, instead of becoming the application a drowning human notices.

Then the feedback. Rejection carries far less information than it feels like it does. In one study, technical recruiters were checked against measured interview outcomes: they were right a little over half the time, two of them looking at the same resume disagreed by about forty percentage points on average, and candidates they rated as near-certain failures went on to pass about half the time. A seeker reads each rejection as a verdict. Measured, it is closer to noise. Worse, the rejections are correlated - a handful of screening vendors serve many employers, so the same person can be filtered out the same way everywhere at once. That is noise arriving dressed as consensus, and it is corrosive, because the seeker has no way to tell the difference.

Then what never gets asked. Every seeker carries constraints that decide whether a job is even possible: visa status, schedule, salary floor, commute, what they refuse to do again. Boards collect a keyword and a location radius. The constraints stay in the seeker's head, so the list on the screen is hundreds of maybes, and the cost of evaluating every maybe lands on the person with the least spare capacity to pay it. The research on choice is clear that this is the worst configuration: options with no clear winner, high stakes, no articulated ideal. People defer. Scrolling that ends in nothing is not laziness; it is the predictable output of the setup.

The clean evidence that information itself is the lever comes from a field experiment I think about often. Unemployed seekers were given one small thing: the system took their stated occupation and showed adjacent occupations people like them actually move into, as suggestions they could prune. Interviews rose substantially - most for the people searching too narrowly - and the number of applications did not increase at all. Nothing about the seekers changed. No new skills, no extra effort. Only the picture in front of them changed, and the outcomes moved.

So what does better information delivery owe a seeker? I hold my own work to five things.

Verified reality. A listing shown to a seeker should be a real job, at a real employer, live right now - checked, not assumed. Anything less spends the seeker's evenings on the system's data-quality problem.

Constraints asked once, honored visibly. If someone says they need sponsorship, every list after that should reflect it, and they should be able to see that it did. Asking is cheap. Re-asking, or silently ignoring the answer, is how trust dies.

An honest zero. When nothing matches, say so, and name a route - an adjacent title, a wider radius, a standing alert. Padding a thin result with near-misses reads as progress and costs the seeker the one thing they cannot replace.

A reason with every recommendation. "Here is why this one" turns a maybe into a decision the seeker can actually make. A ranked list with no reasons is just the pile again, sorted.

Closure. Some answer, even when the answer is no. The silence is the cruelest part of the current market precisely because people fill it with a story about themselves.

The honest limitation: information does not create jobs. The market right now hires slowly no matter how clear your picture is, and a better-informed search in a frozen market is still a hard search - shorter, saner, better aimed, but hard. And I am not a neutral observer here; I build this kind of system for a living, so discount my enthusiasm accordingly and judge the claims by what you can check.

But the core belief survives the discount. When someone struggles to find work, the first question should not be what is wrong with them. It should be what they were shown, what they were asked, and what was never checked before it reached them. In my experience, that is where most of the answer lives.