AI recruiting in India: what it is, where it helps, and how to use it responsibly

The term is used loosely, so it helps to separate the jobs AI can do. The first is parsing: turning an unstructured resume or JD into structured fields such as skills, years of experience, employers, projects, certifications, location and notice period. The second is comparison: lining up what a candidate has stated against what a role requires. The third is generation: drafting outreach messages, interview question banks or summary notes. The fourth, and most contested, is ranking or scoring candidates.

Parsing, comparison and generation are assistive — they save time while leaving judgement with the recruiter. Ranking with a single opaque score is where most of the risk sits, because it hides the reasoning behind a number and invites people to trust it. A responsible AI recruiting setup leans heavily on the first three and treats the fourth with a lot of suspicion.

Indian tech hiring has a few structural features that change how AI should be applied. Notice periods of 30 to 90 days are common, and buyouts are frequently negotiated, which means the gap between offer and joining is long and uncertain. Offer drop-offs and counter-offers are a routine part of the process rather than an exception, so any tool that stops at 'shortlisted' misses the part of the funnel where many hires are actually lost.

The supply side is also layered. Global capability centres in Bengaluru, Hyderabad, Pune, Chennai and Gurugram often work with several staffing agencies at once, and some profiles pass through vendor chains before reaching the hiring team. The same candidate can arrive twice with two differently formatted resumes. Campus hiring and lateral hiring need very different signals — projects and fundamentals for freshers, delivered work and depth for laterals. A generic AI tool trained on other markets usually handles none of this well.

The biggest win is reading volume. A recruiter screening for an SAP S/4HANA FICO consultant can spend most of a day just finding out which resumes mention an actual implementation versus a training course. AI that extracts projects, modules and roles into a consistent structure lets the recruiter scan a shortlist in minutes and spend their attention on the handful of profiles that deserve a call.

The second win is consistency. When every resume is compared against the same structured list of requirements, two recruiters looking at the same pool are far more likely to reach similar conclusions. That consistency also makes it easier to explain a decision to a hiring manager: instead of 'I liked this one', the recruiter can point to the specific requirements where the candidate showed evidence and the ones that still need to be probed in interview.

The most common failure is treating keyword overlap as competence. Resumes in India are often written to pass filters, listing every tool the candidate has heard of. A system that rewards keyword density will push exactly those resumes to the top, while a quieter resume from someone who actually ran Spark jobs in production sinks. The second failure is opaque scoring: a '78% fit' tells a recruiter nothing about why, and it cannot be challenged or corrected.

The third failure is inference about people rather than work. Models can pick up proxies for age, gender, religion, caste, marital status or disability from names, photos, graduation years, addresses or gaps. Any system that lets those signals influence ranking — even indirectly — creates legal, ethical and reputational risk. The safest design is not to infer protected attributes at all and to strip photos and similar fields out of what the matching logic sees.

India's Digital Personal Data Protection Act, 2023 sets expectations for how personal data is collected and processed, including notice, purpose limitation and consent. For recruiting teams this means being clear with candidates about what data is being processed, for what purpose, and giving them meaningful control. Running every resume you have ever received through a new AI tool without that clarity is a poor starting point.

Practically, teams should be able to answer a few questions for any AI recruiting tool: what data it stores, where, for how long, whether candidates can see and correct what was extracted, and whether candidates can control who sees their profile. This is not legal advice — your own counsel should interpret the Act for your organisation — but these questions are a sensible baseline for any vendor evaluation.

Guardrails work best when they are written down and built into the product rather than left to individual good intentions. The core principle is simple: AI can organise, compare and suggest, but a human makes every decision that affects a candidate. That includes rejections, which should never be triggered automatically by a model output.

The second principle is explainability at the requirement level. Rather than one number, a recruiter should see, for each requirement, what evidence the candidate provided, where it came from, and whether anyone has checked it. That makes the AI's contribution auditable and lets a recruiter overrule it quickly when it is wrong.

RecruitGPT uses AI to structure resumes and job descriptions and then shows, for each requirement, whether there is Evidence found, Partial evidence or No evidence yet. Every piece of information carries a provenance label — Candidate Provided, AI Extracted, Candidate Confirmed, Verified or Not Verified — so recruiters can see exactly how much weight to put on it. 'Verified' is only used when a human reviewer has actually checked the item; RecruitGPT does not currently connect to third-party verification or background-check providers.

The AI never auto-rejects and never auto-hires, does not produce opaque fit or hireability scores, and does not infer age, gender, religion, caste, marital status, disability or anything from photos. Recruiters make every decision, and candidates control their own profile visibility.

Before rolling out any AI recruiting tool, run it against a real requisition — ideally one you have already closed — and compare its output with what your best recruiter concluded. Look closely at the profiles it ranked highly and the ones it missed. The goal is not to find a tool that is always right, but one whose mistakes are visible and easy to correct.