What a Stanford study of 4 million job applications says about your job search

You send out applications. Most of them go quiet. No rejection, no interview, nothing. It is easy to read that silence as a verdict on you.

A new study from Stanford suggests the silence is often mechanical, and that the way most people apply makes it worse.

The study

Researchers at Stanford, Chapman, and Northeastern got something rare: a dataset of real hiring decisions made by deployed screening algorithms, tracked across many employers at once. The numbers are large. They analyzed 4,197,168 applications from 3,372,132 applicants to 1,746 positions across 156 employers. It was published at FAccT 2026, the main academic venue for this kind of work.

Three things in it matter if you are looking for work.

1. A machine usually screens you first, and a person may never see your file

The study notes that over 90 percent of large employers now use algorithms to screen or rank applicants. When the model returns "do not recommend," the applicant is, in the researchers' words, likely to be rejected without consideration by a human.

So the first reader of your application is frequently not a recruiter. It is software. That is not a conspiracy theory anymore. It is a measured fact with a sample size in the millions.

2. The same few vendors decide for thousands of companies

Most employers do not build their own screening tools. They buy them from a small number of vendors. The study points out that one vendor's tools were used by more than 60 percent of the Fortune 100 and eight of the ten largest US federal agencies.

The researchers call the result an algorithmic monoculture. When many companies lean on the same model, a "no" at one company is correlated with a "no" at the next. You are not getting a fresh, independent read each time you apply. You are often getting the same judgment, repeated.

3. Applying to a few perfect roles is the riskiest thing you can do

This is the finding to act on. Because the same logic screens you again and again, some applicants get filtered out almost everywhere they apply. The researchers' practical conclusion for job seekers is direct: you have to apply widely to have a real chance that a human eventually looks at your file.

Applying narrowly feels disciplined. The data says it concentrates your risk. Breadth is the defense.

Where I am careful, because the product I build refuses to oversell

The study looked at one specific kind of screening: skills games, not resume parsing. The authors say plainly that resume screening may behave differently. So I am not going to tell you this proves resume robots are biased, or quote you a single scary percentage as if it describes everyone. It does not, and a study this good deserves to be read for what it actually says.

What it does establish is solid: algorithmic screening is now the default, a small set of vendors hold the gate, and applying broadly beats applying narrowly. That is enough to change how you job hunt.

For readers in Sweden and the Nordics

This is not only a US story. The EU AI Act now classifies hiring algorithms as high-risk systems, with compliance obligations for the companies that build and deploy them starting 2 August 2026. The machines screening European applicants are real, and from this summer they are formally regulated. The topic is about to get a lot more public.

What to do about it

The honest takeaway is uncomfortable for anyone who hates busywork. You need to apply to more roles, not fewer. The problem is that doing that by hand, while still tailoring each application so it is not generic filler, takes hours you do not have. That is the exact bind.

It is the reason I built CVFriend. It holds your full history once, surfaces roles that fit, and writes a tailored CV and cover letter for each one in your own voice, scored so you can see it will pass the screen. Breadth without the grind, and without the machine prose that gets you filtered for a different reason.

The silence is not always about you. Sometimes it is about how the system is built, and how you are playing it. You can change the second part today.

Source: Bommasani, R., Bana, S. H., Creel, K. A., Jurafsky, D., and Liang, P. (2026). Algorithmic Monocultures in Hiring. ACM Conference on Fairness, Accountability, and Transparency (FAccT '26). arXiv:2605.27371. DOI 10.1145/3805689.3812400.