Verification Proves a Record Exists, Not That the Person Did the Work

What does an employment verification confirm?

What an employer or a registry filed, and nothing past it. A verification returns dates of employment and job title, and for education the credential and the date it was conferred, drawn from a payroll database, a school registry or a call to an HR line. That is a genuine control and worth buying. It establishes that a record exists, which is a narrower thing than establishing that the description on the resume was true.

The vendor you inherited is one of many, with different sources behind it. The CFPB's market snapshot cites an industry analysis estimating 1,954 background screening companies in the United States with revenue of $3.2 billion, two firms accounting for fourteen percent of it ². Those are 2019 estimates from an industry analysis rather than a Bureau census, and the report describes a market that was consolidating. What travels is the shape: this is a fragmented supply chain, and two vendors handed the same candidate will not necessarily reach the same records.

The genre of writing about verification is almost entirely produced by the firms selling it, which is why it reads as though the check settles the question. Set against that, notice what is never on the list of confirmable fields. Scope. Ownership. Whether the person led the migration or attended the meetings about it. Whether the title was a real level or one the former employer handed out instead of a raise.

Those are the fields a hiring decision turns on, and none of them is filed anywhere. Which is why the useful sequence is to verify the record, then get capability evidence from somewhere the candidate cannot simply assert it, the same argument behind how you verify an AI-proficiency claim on a resume.

Why does a clean verification miss the inflation?

Because a headline accuracy rate is dominated by the easy cases, and the easy cases are not where the problem is. The clearest published example is the federal employment-eligibility system. USCIS's synopsis of the Westat evaluation reports that E-Verify queries return an accurate response 96 percent of the time, and that primarily because of identity fraud roughly 54 percent of unauthorized workers run through it are found work authorized anyway ¹.

Those two numbers describe the same system and get quoted separately, which is the whole lesson. The 96 percent covers the entire caseload, overwhelmingly authorized workers; the 54 percent applies only to the 6.2 percent of queries that concerned unauthorized workers ¹. Read the second number and the system looks broken. Read the first and it looks solved. The honest reading is that a name-and-number check confirms what documents say and cannot see a borrowed identity behind them. That evaluation is old, its fieldwork runs to 2008, and it describes a government program, so it belongs here as a lesson about the shape of verification and rates no vendor's product.

The circulating figures for how often candidates lie on resumes come from opt-in panel surveys asking people whether they have ever done it, which measures willingness to admit. The one setting where a rate was actually checked against an index is instructive: verifying every publication listed by applicants to one orthopaedic surgery residency, Meeks and colleagues found misrepresentation in 1.18% of 1,100 citations in the 2016-2017 cycle, against 18.0% in the same program's 1999 study and 20.6% in its 2007 study, and attribute the fall to the application form adding a PubMed identifier ³.

One program, one specialty, one uniquely checkable claim, and the authors offer the causal story as a hypothesis they did not test. Still, the mechanism is the one worth carrying into a hiring process: what moved was not a warning against lying, it was making the claim cheap to check. Fields nobody can check stay soft no matter what the policy says.

Read a mismatch as a question, not a finding

Put it to the candidate, because honest work histories generate discrepancies constantly. Acquisitions, payroll providers, contract staffing, internal title changes applied by the former employer and a legal name change all produce a mismatch that means nothing at all. The CFPB's snapshot names the mechanical version of the same problem: mismatches, possibly from reliance on non-unique identifiers, with insufficient, common or erroneous identifiers raising false positives ².

A common name is the risk factor that mechanism implies. So a discrepancy report is an instruction to ask a question, and the question is boring: which entity paid you, what did they call the role internally, and who would the payroll record be under. Most of them resolve in one email.

What this rule protects is not abstract. Rejecting on an unexplained gap or an unmatched record punishes people with messy but entirely ordinary histories hardest: contractors, people who worked through staffing agencies, people whose employer was acquired twice, people who changed their name. None of those are integrity signals and all of them look identical to one on a report.

Where the report came from a screening company there is a legal floor under this. Under 15 U.S.C. 1681b(b)(3), a US employer must give the candidate a copy of the report and a written description of their rights before taking any adverse action based on it ⁵. That is the statute requiring the conversation the process should have had anyway. It says nothing about how to weigh the answer, and which of your checks the provision reaches is a question for counsel.

The education half has its own version. A no-record response is frequently an archive limit, because the depth of history a registry holds varies by school, and older or transferred records are the ones that fall out of range. Find out which registry answers for that institution and how far back it goes before an empty answer goes into a file as a finding.

Verify the employer, not just the contact

Confirm that the employer exists independently of the number the candidate supplied. The weakest link in the chain is a phone line that routes to a friend, and a verification that dials the contact on the form has established only that somebody answered it. A record at a real employer can also belong to somebody who did not do the work, so a clean report and a fabricated history are not mutually exclusive.

One question to your vendor settles most of this: does the check confirm the employer's existence through an independent source, or does it confirm what the contact the candidate named said. Payroll-database verification and a call to a number on the application are sold under the same label and are not the same control. Ask which one you are buying for each employer on a candidate's history, because a vendor can fall back from the first to the second without saying so.

Even a perfect record is thin evidence about performance. A meta-analysis of 81 independent samples found prehire work experience correlates .06 with later job performance and .00 with turnover, with experience in relevant tasks, jobs or occupations doing barely better ⁴. Those are corrected correlations, they concern experience measured at hire rather than tenure in the current job, and they say nothing about licensure or legally required minimums. So the cleanest report you can buy confirms a variable that barely moves with the outcome you are hiring for.

The verification stays where it belongs, as an integrity control, and the capability question moves to an instrument built for it: a work sample, a structured reference episode about a specific piece of work, or a working session you can read back afterwards. When the artifact under discussion is a portfolio project, the same split applies and the interesting question is who actually did the work behind it, which no registry will ever answer.

See a sample report >