AI resume screening ranks on resume text plus the job description: skills, experience, location, CTC, and notice period when they are present, with must-have skills as gates and good-to-haves as a smaller weight. It should not use social media. Masked identity — name, photo, email, phone — stays hidden until unlock and stays out of the score. If a vendor cannot list those inputs, treat the percentage as marketing.
If the model needs a name to rank, it is not ranking fit. If it needs Facebook, it is not screening a resume.
The inputs that actually move the rank
The pipeline in how AI resume screening actually works is parse, extract, score, rank. Extraction is a data question: what fields exist, and what is left blank.
| Input | Used for ranking? | If missing |
|---|---|---|
| Skills and project bullets | Yes — semantic match to JD | Weak score; cannot invent a stack |
| Must-have vs good-to-have tags on the JD | Yes — gates vs extra weight | Equal-weight counting returns; missing must-have should cap at 55% |
| Experience, titles, companies | Yes — relevance and mapping | Thinner evidence, not a guessed career |
| Location, CTC, notice (if written) | Yes, as fields when present | Left blank; recruiter confirms |
| Name, photo, email, phone | No — masked until unlock | Should never substitute for skills |
| Social media, credits, family | No | Out of scope for resume screening |
Location, CTC, and notice improve ranking when they are on the page. They do not become confirmed fitment until a human asks. That is why 325 AI-matched profiles out of 6,355 applications still need a desk. The model used the documents. The documents were not the whole person.
What is not in the score — including identity
Masked until unlock means the recruiter judges fit on code, role, company, location, and match percentage. Identity is a later click, not a feature vector. If a vendor scores names, you have rebuilt the bias problem the mask was meant to delay. Recruiter notes after a call are a different layer: they can justify an override of the score. They should not be silently baked into a black-box percentage the client never sees.
LinkedIn as a sourcing channel is one thing. Scraping likes into a match % is another. Screening is the resume you lawfully hold.
If the PDF is silent, the field is empty. A guessed 30% hike formula is not extraction.
Name and photo stay off the rank. Unlock is for outreach, not for scoring.
Must-have vs good-to-have as structured data
The JD is not a blob. It is a list of gates and a list of extras. That structure is data. Without it, eight skills share a weight and a missing core language still looks like 75%. With it, the missing must-have caps the total at 55%. That is not a second model. It is a second column in the JD the agency forgot to fill.
Show me the features in the score. If the answer is “our proprietary AI,” ask again: resume text, JD, must-haves — yes or no? Name — yes or no? Social — yes or no?
A DPDP-shaped rule of thumb for agencies
This is not legal advice. Under India’s DPDP Act, 2023, candidate data is personal data you collected for hiring. Ranking should use what you need for that purpose: the CV and the requirement. Expanding the score into social media or identity signals is both a quality mistake and a purpose-creep risk. Practical rules for tools sit in data privacy rules for AI hiring tools in India. Keep the rank explainable: these skills, this JD, this cap. That is also what you will say on a client call.
Hold resumes for the req. Score what is on them. Mask identity until unlock. Do not quietly enrich from the open web. If you cannot explain an input, it should not move the percentage.
Data minimisation is not only a compliance slogan. It is how you keep 4–6 hours of triage turning into a ranked list instead of a mystery number. The less mystery in the inputs, the more a recruiter can override with a straight face.
Rank on the resume. Leave identity masked.
TeamOB - SafalHires scores JD fit from the CV you already have — must-have gates included — and keeps name and contact off the score until unlock.
Free trial