Job seekers pay up to $600 a year to beat a filter some employers never switch on

A hiring lead opens an inbox of applications that all read the same and reaches for the rating feature in the applicant tracking system, hoping for fast differentiation. Wired reports that is one common origin of automated ranking: employers who say they receive hundreds of near-identical applications feed them into their ATS's AI rater ^[12]. That is a reaction to volume, not a published policy, and it is invisible from outside the building.

The invisibility is the product problem. A scoring app reads the job description and your document and returns a number ^[2]. What it cannot read is the employer's configuration: which ATS, which modules were bought, whether the ranker is switched on this quarter. Greenhouse CEO Daniel Chait says no two of these systems are alike, the technology changes fast, and whether AI is involved at all depends on the exact product and which features a team pays for and uses ^[6]. Some systems stop at collecting and tracking applications; others go as far as automated AI screening interviews ^[18]. The score, then, measures text similarity and is sold as a forecast of an outcome it cannot see.

Here is what candidates are told they face: an automatic cut where only the top 10 to 20 percent get read ^[4], which leaves 80 to 90 percent of applications in a pile nobody opens ^[17]. Here is what the reporting supports: automated ranking absolutely happens in some organisations, while in others hiring is handled personally by humans from start to finish ^[5].

The price of guessing wrong is quantifiable. Jobscan runs $30 to $50 a month ^[8], so $360 to $600 a year ^[16] for tuning against a setting the buyer cannot verify, from a company Wired notes has no incentive to see you leave the job market ^[8]. The one outcome data point in the piece runs against the score: a recruiter with atrocious Jobscan results got a dozen interviews and an offer ^[9]. That is a single case and proves nothing about the average, which is the point worth sitting with, because there is no measured pass-through rate on the other side of the ledger either.

What one named employer says it screens on is duller than the folklore. Toshiba's Kim Jones describes culling on job requirements, salary expectations, and whether the person would be a rehire ^[11]: fields and facts, not prose texture. The AI use she objects to arrives later, in the interview, where she hears the pause, then the typing, then the verbose answer ^[19].

For anyone who owns a hiring funnel, the grid has two axes: do you auto-rank, and do you say so in the posting. Rank and disclose, and candidates are optimising for a filter that exists, which is fair dealing. Rank and stay quiet, and your process works while the folklore compounds ^[6]. Read by hand and disclose, and you get documents written for a human reader. Read by hand and stay quiet, and you pay the running costs of a machine you never bought, because every applicant writes for the imagined ranker and you read the output yourself, one keyword-padded page at a time. Chait calls the general pattern a doom loop, where each side uses AI on its own problem in ways that make the problem worse ^[7]. Leaving the quiet-and-manual cell costs one sentence in the job posting, and it is the only move on this list that changes what arrives in the inbox.