Greenhouse's 63% Exaggeration Flag: The Talent-Proof AI Resume

If you are applying to HubSpot, Anthropic, the NFL, Coinbase, or Duolingo right now, your resume is being scored by a fraud layer, not just an ATS. Greenhouse's Real Talent, its CLEAR-powered identity and deception layer, sits in front of 7,500-plus enterprise deployments, and the number that should reshape how you write your next resume is 63%: the share of caught candidate deception that recruiters now label "resume exaggeration."

What Real Talent actually flags, and what it does not

Real Talent runs two checks in parallel: a technical one on your application metadata, and a human one where a recruiter compares your resume to everything else you have said. Neither check verifies your employment history directly. Both punish inconsistency.

The technical check uses IPQualityScore to score 26 signals across your phone number, email address, IP address, and location, bucketed into high-risk, minor-risk, and authenticity markers. Greenhouse then cross-checks the application against CLEAR-returned identity data (last name, email, phone) and, per its own docs, treats even small mismatches like an unexpected email or phone number as potential fraud.

Here is the part most job seekers miss. The 63% "exaggeration" figure from the 2026 AI in Hiring Report is recruiter self-report, not a database lookup. Recruiters flag exaggeration when a resume reads inflated: over-optimized verbs, quantified impact that does not match tenure, LLM parallelism where every bullet has the same shape. The mechanism is a human eye comparing your resume to your LinkedIn and your interview answers, and finding daylight.

The four numbers that reshape how you should write a 2026 resume

Generic AI resumes now underperform hand-written ones at Greenhouse shops because of a squeeze visible in four figures. The table below is the whole argument.

Applications per role tripled. Recruiters are handling nearly double what they did in 2021, on teams 14% smaller. Roughly one in three candidates now uses AI to write the resume, so the pile is more uniform. The response, from Greenhouse and the 117,612 U.S. recruiters in Refolk's index, is a fraud layer that punishes anything that reads generic, mismatched, or overpolished.

Why "exaggeration" is a language problem, not a lying problem

Most resumes that get flagged as exaggerated are not lying. They are overwritten by a model that defaults to consulting-deck grammar. If every bullet starts with a strong verb, ends with a percentage, and covers a scope the recruiter cannot match to your tenure on LinkedIn, you look inflated even when you are not.

The specific patterns recruiters cross-check against your LinkedIn and interview transcript:

• Scope inflation. "Led a team of 12" when your LinkedIn title is Senior Analyst, not Manager.
• Timeline compression. "Delivered a $4M program" inside an eight-month tenure at a company where onboarding takes a quarter.
• Verb theater. Every bullet using "spearheaded," "architected," "orchestrated." Real jobs contain the word "helped."
• Suspiciously round metrics. 40%, 50%, 2x, 10x stacked across five bullets in a row.
• Parallelism. Every line the same length, the same structure, the same cadence. Humans do not write like that. Models do.

The fix is not to remove numbers. It is to make every number defensible in 20 seconds by the person whose name is on the resume. If you cannot explain how you calculated the 42% in the third bullet without opening a spreadsheet, delete it.

The identity check catches auto-appliers, not liars

The technical half of Real Talent is aimed less at exaggeration and more at bots like AIHawk, the open-source auto-apply tool with roughly 30,100 GitHub stars that fires off around 17 applications an hour. If you get flagged here, it is usually a hygiene problem, not an integrity one.

The most common self-inflicted flags:

1. VPN on during application. Your IP resolves to a country your resume does not.
2. Gmail alias or plus-addressing. firstname+coinbase@gmail.com scores worse than a plain address.
3. Phone number from a different country than your listed address. Common for recent movers.
4. Application submitted from a mobile hotspot in a high-risk ASN.
5. Last-name mismatch between resume, LinkedIn, and government ID. Married-name and transliteration cases.
6. CLEAR profile mismatch. Different email or phone on CLEAR than on the application.

Greenhouse explicitly says a failed identity verification does not automatically disqualify a candidate; it just routes you into a slower, higher-scrutiny queue. With Gartner projecting up to 25% of applicants could be fraudulent by 2028, that slower queue is going to get slower.

Get ahead of it. Over 31 million existing CLEAR users can verify with a selfie inside MyGreenhouse. If you are applying to any of the 7,500-plus Greenhouse employers, doing CLEAR verification before you submit puts you in the small share of the pile that skips the investigation lane.

Why AI resumes now hurt referrals more than cold apps

A referred candidate whose AI-written resume does not match how the referrer described them creates a worse signal than a cold applicant. This is the most expensive AI mistake in the current market.

Referrals skip the spam and identity filters, but they do not skip the matching layer or the recruiter's eye. When your referrer tells the hiring manager "she ran onboarding at Stripe for a year," and your resume says "Architected end-to-end lifecycle programs driving 47% activation lift across the North American segment," the referrer looks wrong and you look inflated. Referral candidates are roughly 4x more likely to get hired. Burning that with a generic LLM pass is the worst trade in the market.

The rule: for a referral, write the resume in the words your referrer would use if they were writing it. Then tailor it to the posting. The order matters.

The talent-proof resume: a checklist you can run in 20 minutes

A talent-proof resume is one where every claim survives three cross-checks: your LinkedIn, your references, and a 20-second verbal explanation on Zoom. Here is the pass.

1. LinkedIn parity

Line up your resume next to your LinkedIn profile. Every company, every date, every title has to match to the month. Titles like "Senior Engineer II" on LinkedIn and "Staff Engineer" on the resume are the single most common exaggeration flag. Pick one and use it in both places.

2. Bullet defensibility

Read each bullet out loud. If you cannot explain, in one sentence, how you got the number and what you personally did, cut the number or rewrite the bullet with a smaller, honest one. "Reduced p95 latency 22% by rewriting the search fanout" beats "Drove 40% performance gains across critical infrastructure" every time.

3. Voice consistency

Rewrite three bullets in the voice you would actually use in a Zoom interview. Now compare to the rest of the resume. If the tone drifts sharply, the whole document reads AI-generated. Bring the polished bullets down to match your voice, not the other way up.

4. Metric variety

Count your percentages. If you have more than four "%" symbols on one page, or three bullets in a row ending in numbers, break the pattern. Real work has qualitative outcomes too.

5. Reference alignment

Text your three references the version of your resume you are about to send. If any of them says "I would not have described it that way," rewrite until they would.

6. Identity hygiene

Turn off your VPN. Use your primary email, not an alias. Apply from the country your address is in. Verify with CLEAR before you hit submit on a Greenhouse posting.

Steps 1 through 4 scale badly if you are applying to more than a handful of roles per week. Rewriting each posting's resume by hand, in your own voice, with tenure-matched claims, is exactly where Refolk earns its keep: it drafts from your actual history, tailors to the posting, writes the cover letter, and scores how well you fit before you send.

The sleeper flag: the 35% AI-interview number

The 63% exaggeration figure gets the headline, but the 35% "AI-assisted interview responses" flag is the one that ends candidacies rather than delays them. Getting flagged as inflated moves you to a slower lane. Getting caught using AI live in a video interview usually removes you.

The tells recruiters cite: unnatural eye movement patterns as the candidate reads off a second screen, response latency that lines up with an LLM's first-token time, and answers that are too structured, always three points, always with a preamble. If a bullet on your resume was written by a model and you cannot narrate the story behind it without hesitation, delete the bullet. This is why the checklist above exists.

What to do this week if you are actively applying

The short version, in order of leverage:

1. Verify with CLEAR before your next Greenhouse-employer application.
2. Reconcile every title and date between your resume and your LinkedIn.
3. Cut one metric per bullet that you cannot defend in 20 seconds.
4. Rewrite three bullets in your voice and align the rest of the document to that tone.
5. Text your references the version they will be asked about.
6. Tailor per posting, either by hand or with a tool that keeps claims tied to your actual history rather than confabulating new ones.

The point is not to write a slower resume. It is to write one that is still fast, still tailored, still AI-assisted, but that reads like a human with a paper trail. That is the resume Real Talent is not looking for.

FAQ

Does Greenhouse's Real Talent actually verify employment history?

No. Real Talent's fraud layer analyzes application metadata (phone, email, IP, location) via IPQualityScore's 26 signals, and cross-checks identity against CLEAR. It does not call your former employers or verify dates against a payroll database. The "resume exaggeration" flag is recruiter judgment, triggered when your resume does not match your LinkedIn, your interview answers, or what your references say. Optimize for cross-source consistency, not for a fictional employment-verification API.

Will using ChatGPT or Claude to write my resume get me flagged?

Not on its own. Roughly 30% of job seekers now use AI to write or customize resumes, up from 17.3% in 2024, and Greenhouse cannot detect that directly on the resume itself. What gets flagged is the pattern AI resumes tend to produce: over-polished verbs, uniform bullet structure, metrics that do not match tenure, and voice that drifts from how you speak on Zoom. Use AI for the draft, then rewrite in your voice and cut any claim you cannot defend verbally.

Should I do CLEAR verification before I apply?

Yes, if you are applying to any of the 7,500-plus Greenhouse-powered employers, which include HubSpot, Anthropic, the NFL, Coinbase, and Duolingo. Over 31 million CLEAR users can verify with a selfie inside MyGreenhouse in under a minute. Verified candidates skip the "investigate further" queue that unverified or mismatched applications get routed into. It is free, fast, and puts you in the small share of the pile that gets fast-tracked.

What is the single most common self-inflicted Real Talent flag?

An email or phone mismatch between your application, your LinkedIn, and your CLEAR profile. Greenhouse's own docs say small mismatches like an unexpected email or phone can signal potential fraud. Second most common: applying with a VPN on so your IP resolves to a country your resume does not mention. Both are hygiene problems that take under five minutes to fix and immediately move you out of the high-scrutiny lane.