Can you trust AI screening when you stay in charge?
A tool that says “candidate 14 is a no” and gives no reason is hard to trust. You would be right to doubt it.
5 min read
HireSalt Articles
Research and practice notes from the HireSalt team: how we score resumes, how we test the scoring, and what we have learned building it.
A tool that says “candidate 14 is a no” and gives no reason is hard to trust. You would be right to doubt it.
5 min read
A client asks why candidate four ranks above candidate seven. You do not want to answer from memory. With HireSalt, every score has a written reason, and the client can read it for themselves. You…
Resumes arrive by email, chat, forwards and spreadsheets. After a week, four people hold copies and nobody knows which one is current.
A candidate interviews well. Then you hear nothing from the hiring manager, and they hear nothing from you. A week later they have another offer.
You post one role. Applicants arrive from ten cities. The resumes all look different, and so do the colleges and employers.
Every resume looks good now. The summaries are tidy. The bullets are sharp. You read ten and cannot tell who is strong.
You open the inbox and there are hundreds of resumes. Most of your week is about to disappear into them.
A solo recruiter can screen resumes without adding another monthly bill. With HireSalt, you pay for each resume scored, not for a seat. One credit scores one resume. Your credits never expire, so a…
The client sends two lines and calls it a brief. You do the work anyway. Then the shortlist comes back with a no.
You search for “customer success” and get a short list. But the best person for the job wrote “client onboarding” on their resume. They never show up.
In our test on 16 October 2026, adding a page of key phrases from the job ad to 8 resumes that did not fit the job moved their scores by 0.25 points on average, between -2.5 and +1.5 points on a 0…
When HireSalt reads a job description, it drafts a short list of job-specific criteria to score resumes against. We tested whether those criteria describe the job’s real work. Across 30 job…
In both counts we have made, yes. In 237 test screenings from our 7 October 2026 benchmark run, all 9,314 criterion scores came with a written reason. A second count on 16 October 2026, on the model…
In October 2026, most of them on 16 October, we ran a set of tests on HireSalt’s scoring using sample resumes and job descriptions and no customer data. New runs used our production model. Two…
In three timed runs on 16 October 2026, HireSalt scored every resume in piles of 25, 50 and 100, with no failures. The rates were 271, 252 and 392 resumes per hour. The rate varied from run to run…
In our test, 5 of 5 resume files HireSalt could not read cost 0 credits once processing finished. That is a small test: 5 files, one run, on 16 October 2026. It shows what happened to these five…
When you upload a resume to HireSalt, the file and its text are stored on Railway in the United States. The text is sent to third-party AI providers, which read it and return the scoring. Before…
For a handful of resumes, you can. A chat window will read them, summarise them and tell you which look strongest, and for one role with a few applicants that is often enough. It starts to fail when…
A recruiter with an open role and a full inbox has three jobs to finish before a shortlist reaches the hiring manager. They have to get through the applications without missing good people. They…
A prior run’s calibration numbers were meaningless because heuristic-labeled pairs and pairs carrying the dataset’s own labels were mixed. The repository’s notes report a 2.6-point separation on the…
Our latest repeatability check, on 16 October 2026, missed the bar we set for it: 21 of 32 resume and job pairs stayed within 5 points across three runs, against a rule of 80%. We do not claim…
HireSalt gives every resume a score from 0 to 100. This article explains how AI resume scoring works in HireSalt, using one candidate from start to finish. The score is a weighted average of…
Put Candidate A and Candidate B side by side and the skills lists match. Both list Python, SQL, experimentation, dashboarding, stakeholder reporting and mentoring. Both have six years of work, and…
Three models scored each of 143 resume-job pairs. On 14 pairs—about one in ten—the same pair received scores at least 20 points apart. Across the seven models tested against the baseline, pooled…
Every criterion on the Scoring Health tab has its own slider, and each one moves ten points either side of its baseline weight. Drag one and the top 10 list updates at once. No scoring run starts…