process 01Smarter ScreeningRead and score every application, not just the top of the pile. 02Better ShortlistingRank on twenty signals, with the evidence behind each one. 03Faster SchedulingNo calendars, no slots. One link, good for fourteen hours. 04Fairer InterviewsQuestions built from the role, answers scored against a written rubric.
Pricing
use cases 01Resume VerificationEvery claim read in context, not lifted out as a keyword. 02AI Cheating PreventionBuilt for the copilot era: divided attention, novel questions. 03Volume ScreeningThe same rubric for applicant one and applicant a thousand. 04Pre-BGV FilterA consistency check before formal verification spend.
For candidates For investors Candidate login Employer login Get access →
what happened

Only 16% of HR Teams Trust AI Screening Alone. Here's Why.

A September 17 survey of 646 recruiters found 98% have caught a candidate lying, but only 16% of HR staff and 11% of staffing recruiters trust AI screening without a human involved.

Prateek Porwal · · 3 min read

on this page
  1. What the survey found
  2. The real story isn't the fraud rate. It's what “trust” means here.
  3. What the survey doesn't settle
  4. Where AgentR fits, and where it doesn't

On September 17, RefAssured, a reference-checking and screening vendor, published results from a survey of 646 jobseekers, staffing recruiters and corporate hiring managers. The topline number is almost boring by now: 98% of HR and hiring managers say they have personally caught a candidate lying about their qualifications. The number sitting next to it is not boring at all. Only 16% of corporate HR staff and 11% of staffing recruiters said they completely trust an AI-powered screening tool that works without a human in the loop.

What the survey found

The report puts numbers on two trends that have mostly been argued about anecdotally until now: how normal resume embellishment has become, and how far practitioner trust in automated screening actually extends. Neither number is new in kind. Both are new in scale.

  • 84% of HR and hiring managers, and 76% of staffing recruiters, say candidate fraud is more common today than it was three years ago.
  • 74% of jobseekers have used generative AI to write or edit a resume, and 88% have tailored a resume with keywords pulled straight from the job posting.
  • 13% of staffing recruiters say they have personally encountered outright identity fraud, a real problem, but a much smaller one than the near-universal resume misrepresentation above it.
  • 61% of jobseekers say they would still rather deal with a human recruiter than a fully automated process.

The real story isn't the fraud rate. It's what “trust” means here.

Read the two headline numbers side by side and the survey isn't describing distrust of AI in general. It's describing a split between two different jobs one tool is being asked to do: checking whether a specific claim is true, and deciding whether a person should be hired. Practitioners only trust automation for the first one.

Resume misrepresentation, the thing 98% of managers say they've caught, is a checkable-claim problem: did this person hold this title, earn this degree, work at this company. It has a right answer, and a machine can find it by comparing the claim to a public record. Identity fraud, the outright fabricated candidate, is rarer and harder, and even there the fix is better verification, not better judgment. Neither number in this survey describes a case where the actual hiring decision needs to be made by an algorithm. AgentR's own look at candidate honesty found a similar split: candidates admit to embellishing far more than employers ever catch, which is itself evidence that the catching, not the deciding, is where the gap sits.

The gap isn't between people who trust AI and people who don't. It's between checking a claim and making a call.

For a team hiring next week, the useful question isn't whether to trust AI in hiring. It's what exactly the tool is checking, and who signs off after. A vendor claiming to “detect fraud” should be asked whether that means checking facts against a record or replacing the human's call. This survey suggests hiring teams already know which one they want.

What the survey doesn't settle

RefAssured sells reference-checking and fraud-prevention software, so a report headlined around rising candidate fraud also serves the company's own pitch. That doesn't make the numbers false, but it's a reason to read the exact percentages as directional rather than final. The published report doesn't break its 646 respondents down by group, give a margin of error, or separate “exaggerated a skill” from “lied about a job title” from “invented an identity” inside its headline catch rates, so the near-universal 98% and 95% figures likely bundle very different severities of dishonesty into one number. It also says nothing about how any of this differs outside the US and Western Europe, where hiring volumes and candidate-market pressure look different — India alone puts roughly ten million freshers into the job market every year, a scale AgentR has written about separately as its own distinct hiring problem.

Where AgentR fits, and where it doesn't

AgentR reads applications in full and checks specific claims, job titles, dates, credentials, against the public record, which is exactly the category this survey shows practitioners already catch constantly and still don't trust software to handle alone. That's by design: AgentR produces a ranked, evidenced shortlist, not a decision, and every advance or rejection stays with a person on the hiring team. Its resume verification layer is built to answer the checkable-claim question the survey's 98% figure is about.

What it doesn't do: replace the judgment call this survey shows nobody wants automated away, or catch the rarer identity-fraud cases, a fabricated person coached through a live interview, that a claims-verification layer isn't built to see. AgentR's own look at the deepfake-candidate problem covers that separate failure mode in more depth.