AI-Skills Hiring Is Up 16%. The Resume Still Can’t Prove Anyone Has Them.

AI-related hiring in India rose 16% year-on-year in June 2026, even as overall IT job postings fell 3% over the same period, according to Naukri’s monthly JobSpeak report, which tracks listings across more than 150,000 companies. Read that pair of numbers together and the shape of the problem changes. It’s not that IT hiring is shrinking. It’s that the shrinking pool of open roles is increasingly gated on one specific claim: that the candidate can actually work with AI and ML tools. And a resume is the worst possible instrument for checking whether that claim is true.

Everyone learned to write the same line

A resume that lists “AI/ML proficient” or “hands-on with LLMs” costs a candidate nothing to write and, until recently, cost a recruiter almost nothing to accept at face value, because there was no fast way to check it. That was tolerable when AI/ML fluency was a nice-to-have buried in a long skills section. It’s a different problem when it’s the specific line item a growing share of open roles are now screening on, because the gap between “wrote it” and “can do it” now sits directly on the path to an interview slot instead of somewhere further down the funnel.

This isn’t the same problem as a generically inflated resume. We’ve written before about what happens when every resume looks equally polished because AI wrote all of them, a signal-collapse problem, where the document stops differentiating candidates at all. This is narrower and, in a way, more dangerous: the document still differentiates candidates, it just differentiates them on a claim nobody is checking. A recruiter scanning for “TensorFlow” or “prompt engineering” as keywords is running exactly the search a candidate optimized their resume to pass, whether or not they’ve ever debugged a model that didn’t work the first time.

What a keyword match actually verifies

A keyword match verifies that a word appears in a document. It does not verify that the candidate can explain what a learning rate does, describe a time a model’s output surprised them, or walk through why they chose one approach over another on a project they claim to have built. Those are the questions that separate “used it in a tutorial once” from “can be handed a real problem,” and none of them show up in an ATS keyword filter. At the volume Naukri’s data implies, with AI-tagged roles growing while everything else contracts, a growing share of every campus and lateral drive in India is now running a first-pass filter that can’t tell the difference between those two candidates, because the filter was built to check for words, not understanding.

Where our pilot data actually applies here

We’re not going to pretend voice screening is a novel fix for a skills-verification problem specifically. That’s not what our pilot data measured. What our pilots do show is the throughput side of the same constraint: across 1,099 structured voice interviews run over 14 hiring campaigns, a structured conversation auto-classified 60%+ of candidates as clear no’s before a recruiter opened a file, and one drive screened 3,000 candidates in a single evening. A conversation format that already handles that kind of volume is the right shape of tool to extend into asking a candidate to explain a technical claim in their own words, rather than trusting the word on the page. The constraint was never whether a conversation-based screen could operate at scale, it’s whether the question set inside that conversation is built to probe a specific skill claim instead of general communication clarity.

The contrarian take

Most TA teams responding to the AI-skills surge are doing the obvious thing: adding “AI/ML experience” as another keyword filter on top of the existing ones. That makes the funnel narrower without making it more accurate, because a keyword filter and a false claim are the same document doing the same thing they always did. The actual fix isn’t a sharper filter on the same weak instrument. It’s moving the verification question, can this person actually explain and defend the skill they claimed, earlier in the funnel, to before a human interviewer’s time gets spent finding out the hard way that the resume oversold the candidate.

If your next drive is screening specifically for AI/ML-tagged roles and you’re still relying on a resume keyword match to get there, the honest test is simple: pick ten shortlisted resumes and ask each candidate, live, to explain one technical decision behind the project they listed. Talk to us about building that question into a structured first-pass conversation instead of finding out the gap exists in round three.