AI resume screening parses the file, extracts skills, experience, location, CTC, and notice period when those fields exist, then scores the profile against the job description using must-have gates plus semantic match. The output is a ranked list, not a hire. A recruiter still calls to confirm intent, notice, and salary before anything goes to a client.
The model reads what the resume says. It does not know what the candidate will confirm on a Tuesday morning call.
The four stages between a PDF and a ranked list
Indian staffing desks still live in PDFs, Naukri downloads, and forwarded mail. AI screening does not skip that mess. It turns it into a sequence a recruiter can audit.
The file is converted into text: headings, bullets, tables, the messy footer with CTC. Layout is discarded. What remains is the raw language the scorer can read.
The model pulls structured fields when they are present — current city, years in role, tech stack, last drawn CTC, notice period. If a field is blank, it stays blank. Nothing is invented to make the score look complete.
Must-have skills act as gates. Semantic match then compares meaning, not only exact keywords, so “SSE” can sit near “Software Engineer.” Missing a must-have caps the score — in a gate-based model, at 55% — so a glossy AWS list cannot hide a missing core language. See why most ATS systems get must-have vs good-to-have wrong.
The list is sorted by match. That is the whole product of screening AI: order. The recruiter still phones, still checks notice and CTC, still decides what the client sees.
What the extractor can read — and what it will not invent
Agencies get into trouble when they treat a match percentage as confirmed fitment. Extraction is only as good as the page.
| On the resume | What screening does | What it does not do |
|---|---|---|
| Skills and project bullets | Extract and match to JD language | Verify the candidate can still do that work |
| Location, CTC, notice (if written) | Capture as fields for ranking | Ask whether they will relocate or take 9 LPA |
| Job titles | Map variants (SSE, Consultant, Developer) | Prove the title matches the actual work |
| Name, photo, email, phone | Can be masked until unlock | Must not drive the match score |
For a fuller list of ranking inputs, see what data AI resume screening actually uses. For how meaning-based match differs from Naukri Boolean, see keyword matching vs semantic matching.
A 78% score means the text aligned. It does not mean the person in Pune will join a Hyderabad office on a 30-day notice. That gap is documented in why so many AI-matched candidates fail recruiter screening in India.
Scoring is a gate, then a rank — not a verdict
Equal-weight skill counting is how a candidate missing Java still looks “strong.” A working pipeline does the opposite: check must-haves first, cap the score if a gate fails, then rank the rest on semantic overlap, experience, and good-to-haves. The 55% cap is a signal to the desk, not a delete key. Edge cases stay visible. They just do not sit at the top of a 200-resume pile.
Without ranking, recruiters burn 4–6 hours a day triaging Naukri, email, and LinkedIn in arrival order. Ranked screening is built to cut that by 70–90%, with a first strong match typically in about 2 hours after posting — because the queue is ordered by fit, not by who applied at 9:04 a.m.
Where the recruiter still has to pick up the phone
In one agency funnel, 6,355 applications produced 325 AI-matched profiles. Ranking got the desk out of the raw pile. It did not place anyone. Notice, CTC, location willingness, and whether the person is actually looking still sit on the call. That is not a failure of the model. That is the design: AI orders the work; humans confirm the work.
If your team treats the percentage as a client-ready stamp, you will resend the same screening-fail pattern — high paper match, low confirmed fitment. Use the score to decide whom to call first. Use the call to decide whom to submit.
See ranked screening without the black box
TeamOB - SafalHires parses, extracts, gates on must-haves, and ranks across Naukri, email, and LinkedIn — then leaves the call with your recruiter.
Free trial