How to Tell If a Candidate Actually Built What's on Their Resume

"Led the migration to microservices." "Owned the recommendation engine end-to-end." "Reduced latency by 40%."

Every resume in your pipeline right now has a line like this. The question that actually matters — and the one most hiring processes never answer — is whether the person in front of you did the thing, or just wrote the sentence.

This isn't a few bad actors

The instinct is to assume resume embellishment is rare — a handful of dishonest applicants in an otherwise trustworthy pool. The data says otherwise. 61% of job seekers say they exaggerated their expertise to better match a job posting. 59% inflated the impact or scope of a previous role. 45% adjusted employment dates to paper over gaps.

And it's not just resume copy anymore. 50% of job seekers say they used AI to tailor a resume for a role they didn't fully meet the requirements for. Three in four candidates who used ChatGPT to write their resume report that it got them an interview. The tools that make a resume sound more impressive are now free, fast, and good — and most screening processes have no way to tell the difference between earned experience and well-prompted experience.

The honest candidates know this too. 60% of job seekers who embellished said they wouldn't have been hired if they'd presented their experience exactly as it happened. Which means the pressure to inflate isn't going away — if anything, it's a rational response to a hiring market where everyone assumes everyone else is doing it.

Why the usual checks don't catch it

Reference checks assume the references are real and willing to be honest. Increasingly, neither is guaranteed — candidates who fabricate experience can also arrange friends or accomplices to confirm it when a recruiter calls.

Resume keyword review rewards whoever wrote the most convincing sentence, not whoever has the most relevant experience. It's the exact mechanism AI resume tools are built to exploit.

A single live interview question like "tell me about a project you led" can be rehearsed. Candidates increasingly practice with AI coaching tools until a fabricated story sounds as fluent as a real one — 61% of job seekers report using AI to practice interview answers until they sounded more convincing.

Take-home assignments have the same problem we covered in an earlier post: a polished submission proves someone has access to AI, not that they understand what it produced.

The common failure across every weak verification method is the same: they all evaluate the artifact — the resume line, the reference, the code file — instead of the person's actual understanding of it.

What actually exposes the gap

There's one verification method that's hard to fake, and it's the oldest one in the book: a real-time follow-up question that depends on the answer to the previous one.

If a candidate genuinely led a microservices migration, they can describe what broke first, why they chose that specific approach over alternatives, and what they'd do differently with what they know now. If they didn't, the follow-up exposes it almost immediately — recruiters who've started probing this way report a consistent pattern of candidates who score well on a written or AI-assisted assessment and then go quiet the moment a hiring manager asks "why that approach, specifically?"

This is the mechanism, not a tool. It works whether a senior engineer is asking the questions or a structured AI interview is. The problem has never been a lack of good questions — it's that most screening processes don't have the time, consistency, or depth to ask three or four layers of follow-up on every claim, for every candidate, every time.

What this looks like in practice

Ray is built around exactly this gap. It runs adaptive voice interviews — Screening, DSA, and System Design — that don't move on when an answer is vague. When a candidate says they "optimized database queries to reduce latency," Ray asks which queries, what the baseline and result actually were, and what tradeoff they accepted to get there. A candidate repeating a memorized or AI-generated answer runs out of road within two or three follow-ups. A candidate who actually did the work keeps going, because real experience has texture that a generated sentence doesn't.

The output isn't a pass/fail score. It's a structured debrief that shows exactly where a candidate's answers held up under follow-up and where they didn't — so whoever makes the next call is working from evidence, not a resume line and a gut feeling.

• Every claim gets a follow-up. Not a fixed script — a question shaped by what the candidate just said.
• The gap shows up in the transcript. Vague answers and confident specifics both come through clearly in the debrief, instead of getting smoothed into "seemed strong" feedback.
• It scales the one verification method that actually works — real-time follow-up — across every candidate, not just the ones a senior engineer has time to dig into.

Resume embellishment isn't going to reverse itself — the incentives and the tools both point the wrong way. The fix isn't catching liars. It's building a screening step that surfaces the truth either way, for every candidate, before anyone's calendar gets involved.