A job seeker can apply to ten different companies, get turned down by all ten, and never learn that one automated tool made every one of those decisions. Each employer looks separate, with its own logo, job posting and portal, so the rejections read like a run of bad luck or a weak application.
Stanford researchers measured that pattern in an independent study of AI hiring tools, the largest of its kind when it came out. It tracks more than 4 million job applications from roughly 3 million people. Although the applications went to 156 employers in eleven industries, every one passed through screening software from a single vendor. That let the researchers test whether repeat screening by one tool changes the odds for the same person.
It does. People screened by that vendor for several jobs are rejected by all of them more often than chance would predict. About one in ten people who apply to four such jobs are turned down by every one. Even among people who apply to ten, roughly one in twenty-five is rejected every time, a pattern the researchers call systemic rejection. They ran the same test on 83,000 older applications to Fortune 500 companies, sent before this kind of screening took over. There, rejections lined up almost exactly with what unconnected decisions would produce. The excess appears only once a single vendor makes the calls for many employers.
The vendor’s method explains why. Instead of a resume, candidates complete short, game-like tasks meant to measure how they behave. Employers using the vendor rely on 42 different scoring models, all built from the same data, so they tend to reach similar conclusions from the same answers. The vendor also lets candidates retake the test only once every 330 days. Someone who applies to several of those companies within a year is therefore scored on the same answers each time. One rejection can then become a rejection everywhere the tool is used, for most of a year, with no new evaluation.
Colorado has signed a law that covers decisions like these. It requires companies to disclose when an automated tool shapes a hiring decision and to let applicants have a person review it. The law takes effect in January 2027 and does not change how the screening tools are built, so one score can still reappear under a different employer’s name.
