Refolk
10 min readMarkdown

Reference Checks Are Moving Before the Interview. Cut Unsourced Numbers

Many employers now survey references before the first interview. Check every AI-written number and skill line against a real source, and cut what fails.

Your former manager may now fact-check your resume before anyone at the company has met you. If an AI tool tailored that resume and slipped in a number or a skill you can't back up, that's where it fails - quietly, and before the first interview.

A recent Dice piece on the resume-screening shifts coming in 2027 reports that many companies are moving reference checks into the initial screening phase to catch exaggerated skill sets. That changes when your resume gets fact-checked. It used to happen in a final-round conversation where you could explain yourself. Now it can happen in a survey your former manager fills out. Every number and skill line on your resume needs a source. Anything without one has to go.

Why are reference checks moving to the screening stage?

Employers are running reference checks earlier because AI makes inflated resumes cheap to produce and hard to spot on the page. Nick Nordin, SVP of Recruiting at the staffing firm BridgeView IT, told Dice that many companies now check references during initial screening specifically to catch exaggerated skill sets.

Both sides of the market back this up:

  • Hiring managers: In a February 2026 Express Employment Professionals-Harris Poll survey, 86% of U.S. hiring managers said AI makes it too easy to exaggerate skills on resumes.
  • Candidates: Greenhouse's February 2026 data found 28% of job seekers admit using AI to generate fake work samples, and 32% have claimed AI skills they don't have.
  • Late-stage leakage: In one recent BridgeView technical hiring funnel, roughly 1 in 4 candidates who reached late-stage interviews still showed fraud-risk signals until a deeper screening layer caught them.

Speed is what makes this practical. BridgeView says its automated reference checks average a 26-hour turnaround and include anti-fraud detection to confirm the responses are genuine. Its iOS, SQL and security role pages say every candidate completes two reference checks. When a reference comes back in about a day, running one on every shortlisted candidate is easier than leaving interviewers to find the fakes one at a time.

86%
U.S. hiring managers who say AI makes it too easy to exaggerate skills on resumes

From a February 2026 Express Employment Professionals-Harris Poll survey.

What does a screening-stage reference survey actually compare?

It compares the skills you reported against what your references say in a structured survey, one line at a time. Dice describes the process like this: an AI system asks you for reference contacts, runs the checks, and combines the feedback. Recruiters then filter out candidates whose self-reported skills don't match what their references said.

For a padded resume, that's worse than any interview. In an interview, if someone asks about "increased revenue 34%," you can explain the baseline, the time window, and what was yours versus the team's. A survey gives you none of that. Your old manager sees a question that boils down to "did this person do this, at this level?" and answers from memory. There are no follow-up questions, and you're not there to add context.

So the test for each line on your resume isn't "can I defend this in a conversation?" It's "would my reference confirm this cold, without seeing my resume first?"

The worst part is that you won't find out. A mismatch between your bullets and your reference's answers doesn't come back as feedback. It turns into a filter, and you just stop hearing from that company. You'll never learn whether it was the invented 34% or the skill line you added because the posting mentioned it. From where you sit, it looks exactly like every other application that went nowhere.

Who is actually reading your reference's answers?

Usually it's someone who has done the work you're describing, not a verification specialist. In Refolk's index of professional profiles, 5,278 U.S. profiles list "Reference Checking" as a skill, but only 34 hold a dedicated verification title such as Employment Verification, Background Screening, or Reference Check.

Segment (U.S.)FigureWhat it tells you
Profiles listing "Reference Checking" as a skill5,278Lots of people run checks as part of a wider job
Profiles with a dedicated verification or reference-check title34Very few people do it full time
Skill holders per specialist title~155xThe checker is usually a generalist who knows the work
C-suite share in a sample of 25 "Reference Checking" profiles20% (5 of 25)Small sample, but CEOs and COOs show up

The specialists do exist. The top employers in that small group include DISA Global Solutions, Verified First, GroupOne Background Screening and Merchants Information Solutions. They're outnumbered by a wide margin, though. The sampled titles of people who list the skill include CEOs, co-founders, COOs, program directors and client-services managers. Staffing firms in the sample, such as Aditi Consulting, Impellam Group, TSR Consulting and Search Wizards, run checks themselves instead of handing them off.

155x
People listing "Reference Checking" as a skill for every dedicated verification specialist, U.S.

5,278 skill holders versus 34 specialist titles in Refolk's index.

Here's why that matters. A background-check specialist confirms dates and titles. An operator who has run a migration or owned a quota knows what "led the migration" looks like at your level, and can tell when a reference's lukewarm answer doesn't match your confident bullet. Assume the person comparing the answers knows what the work actually looks like.

Why does AI tailoring invent numbers and skills?

AI writing tools fill gaps with plausible numbers because a bullet with a metric reads better, and the model has no idea what your real metric was. A TechBullion explainer published November 12, 2026 put it plainly: "Increased revenue 34%" reads well until an interviewer asks how you measured it, and recruiters now treat unusually round, unusually impressive, unsourced figures as a yellow flag.

Fake metrics on AI resumes tend to come in three forms:

  1. Invented outcome percentages. The model sees "improved onboarding" and adds "by 40%" because that's what strong bullets look like.
  2. Round numbers. Metrics from real projects tend to look like 23%, 41% or 17%. AI-generated ones default to 25%, 30% or 50%. A simple rule: if every bullet on your resume ends in a multiple of five, at least one of them is probably made up.
  3. Borrowed skills. Tailoring a resume to a posting tempts the tool (and you) to copy the posting's tool list.

That third one is where AI-tailored resumes pick up exaggerated skills. Resume Genius found that 36% of job seekers have considered "skills manifesting," which means listing skills they don't have yet, and AI tools are the most common skill people list this way. Those are the exact lines a reference survey asks about.

The pattern gets stronger the more you hand over. Among people who let AI write most or all of their resume, 74.8% admitted to lying on a resume, compared with 24.6% of people who don't use AI. Part of that is who chooses to use AI that heavily. Part of it is probably how the tool works: it's optimizing for a match with the posting, not for accuracy about your past.

This is the gap Refolk is built around. I write the resume from your own history and then tailor it to each posting, so the tailored version reorders and rewords things you actually did. I don't add the skills the posting wants. When the posting asks for something you don't have, the fit score shows it as a gap instead of papering over it.

How do you audit every number and skill line?

For every number and skill on your resume, name the document or the person who would confirm it. If you can't name one, rewrite the line as something you can source, or delete it. This is the source-or-delete audit. Do it on the final tailored version, because that's where new claims sneak in.

For every number

  1. Pull out every figure. Copy each percentage, dollar amount, count and time frame into a separate list. Seeing them out of context makes the invented ones stand out.
  2. Assign a source. Next to each one, write either a document (dashboard, performance review, launch email, quarterly report) or a person (which reference saw it).
  3. Run the screen-share test. One fact-checking guide suggests asking: "could I screen-share the source of this figure right now?" If not, the number is a guess.
  4. Check the rounding. If the real figure was 23%, write 23%. A suspiciously clean number costs you credibility even when it's true.
  5. Run the cold-confirmation test. Would the reference you'll list for that job confirm this number without seeing your resume? If they'd say "I think it improved, I'm not sure by how much," the line is going to fail the survey.

For every skill line

  1. Name the project. Each tool or skill needs a specific project or deliverable where you used it.
  2. Name the witness. Did the reference you're listing see you use it? A skill your manager never saw is one they can't confirm.
  3. Check the level. "Familiar with" and "led" are different claims. Dice's advice is to be ready to demonstrate every skill you list, and to tie skills to real outcomes.
  4. Compare against the original. Diff the tailored resume against your untailored one. Any skill that shows up only in the tailored version needs to pass steps 1 to 3 again.
If your reference would answer "I'm not sure by how much," that number is already gone. Cut it before they do.

How do you quantify a resume without exact numbers?

Replace outcome percentages you can't baseline with scope numbers a reference can confirm. Scope numbers are factual counts of the work itself: team size, caseload, volume handled, deadlines met, systems owned. Outcome percentages claim a before-and-after change, and they often have no baseline anyone recorded.

Claim typeExampleCan a reference confirm it cold?
Unsourced outcome %"Cut churn 30%"Rarely, unless they saw the report
Sourced outcome %"Cut churn 23% (Q3 retention review)"Yes, if they were in that review
Scope number"Managed a book of accounts across three regions"Yes, they assigned it
Qualitative outcome"Became the escalation owner for enterprise renewals"Yes, they made the call

Here's what the rewrites look like in practice (illustrative examples):

  • Before: "Improved deployment speed by 50%." After: "Moved the team's deploys from a weekly release train to daily releases." The cadence change is verifiable. The 50% was invented.
  • Before: "Increased pipeline 40% through outbound campaigns." After: "Built and ran the outbound sequence for the mid-market segment, owned end to end." Your manager knows you owned it.
  • Before: A skills line that repeats every tool in the posting. After: One tool, tied to the deliverable you built with it, such as "Built the data models behind the finance team's monthly close." One real, witnessed skill beats three copied from the posting.

So if you're wondering whether to put numbers on your resume when you don't know them: no. Use the numbers you do know, which are usually scope numbers, and describe outcomes in words. A resume with three real numbers is safer than one with eight round ones.

Does tailoring still pay off once the fake numbers are gone?

Tailoring still pays off, but a resume written by a person from real history beat both AI versions in the one field test available. Kickresume sent 204 applications using four resume versions and tracked positive responses:

Resume versionPositive response rate
Human-written, single column31%
AI draft, tailored to each posting18%
Raw AI draft, untailored10%

Tailoring nearly doubled the AI draft's response rate, from 10% to 18%. Even so, the human-written version did best. The likely reason is the one this whole article is about: a resume written from what actually happened has specific, uneven, sourceable details, and a generated one doesn't. Recruiters read that difference, and now references confirm it.

The practical workflow is to start from your real history, tailor from there, and audit whatever changed. With Refolk you paste the posting and get your own resume back, rewritten for it, along with a drafted cover letter. Then you run the source-or-delete audit on the lines that changed, which is a much shorter list than the whole document.

FAQ

Will every employer run a reference check before the interview?

No. But BridgeView IT's Nick Nordin told Dice that many companies are moving checks into initial screening, and BridgeView's own process averages a 26-hour turnaround, with two references per candidate on its iOS, SQL and security roles. You can't tell from a posting which employers do this, so write every resume as if references will be checked before the interview.

What if the person who could confirm my metric has left the company?

Then the metric has no witness, and you should treat it as unsourced unless you hold a document that proves it, such as a performance review or a launch announcement. You can also list a different reference who saw the same work, or rewrite the bullet around the scope of what you did instead of the percentage.

Is it okay to list a skill I'm still learning?

Yes, if you label it honestly. "Currently learning" or a named course or side project is a verifiable claim. "Proficient" is not, if no reference has seen you use the skill. Resume Genius found 36% of job seekers have considered listing skills they don't have yet, mostly AI tools, which is exactly why reference surveys focus on skill lines.

The data behind this

Put this to work

Paste your career in once. Every application after that is written for you.

Drop a resume or a LinkedIn URL. I rank the live openings against it, rewrite the resume and write a cover letter for the best of them, and fill in the employer's form when you press the button. You read, you decide what goes out.

  1. 01Drop your resume

    A PDF or a LinkedIn URL. About a minute, once.

  2. 02I rank the openings

    Every weekday morning, the live catalog scored against your history. Up to 20 worth your time, not two hundred links.

  3. 03Each one is written up

    Resume rewritten for the posting, a cover letter, a fit score. Press send, or let me fill in the form.

  • New matches ranked and written before you are up.
  • Every bullet stays inside what your history supports. Nothing invented.
  • Queued, submitted, interviewing, offer: one screen, not a spreadsheet.

Free, no card. Up to 25 applications sent a week, and nothing is sent until you say so.

Keep reading