The 9-Word AI Tell: Why Workday Cross-Checks Your Resume Against LinkedIn

Your resume is not getting deranked because you wrote "spearheaded." It is getting deranked because your resume says "spearheaded" and your LinkedIn says nothing at all. In 2026, Workday's screening stack treats your LinkedIn profile as ground truth and your resume as the claim to verify against it, and most candidates are still optimizing the wrong document.

The nine Stanford words (spearheaded, leveraged, pivotal, intricate, showcasing, synergy, delve, realm, robust) are real signals. But they are the smoke, not the fire. The fire is a three-way voice and consistency check between your resume, your LinkedIn, and your cover letter, and it fires before a recruiter opens the file.

Why Workday started checking LinkedIn in the first place

Workday moved to external-signal verification because pure resume scoring became legally and technically indefensible. On March 29, 2024 Workday paid roughly $530 million for HiredScore, and that engine now ships as "HiredScore AI for Recruiting," the layer powering the Workday Recruiting Agent. Then in February 2026 the Mobley v. Workday collective action was certified in the Northern District of California, and the pressure to score candidates against verifiable, external signals (not just parsed resume text) got very loud.

The mechanics matter, because they explain why the same resume can sail through one Workday-run posting and die at another:

• HiredScore is licensed separately from core Workday Recruiting. Employers without the license get parse-and-filter only, no true AI screening.
• Enterprise integrations like Brainner check every application against 3.5 billion data points and 20+ fraud patterns, including AI-generated resume signals and proxy candidates.
• Crosschq earned Workday "Design Approved Integration" status in December 2025 for AI Hiring Intelligence and is already deployed at AAA.

So the same posting can be scored by three different stacks depending on what the employer bought. "Beating the ATS" and "beating the AI check" are two different games, and most advice on the internet conflates them.

The 9-word Stanford list is a symptom, not the trigger

The overused words matter as a cluster, not one by one. Recruiters and detection layers flag resumes where every bullet opens with "Spearheaded / Leveraged / Drove / Orchestrated" and where bullets show near-identical length, because LLMs produce parallel structure by default and humans do not.

Here is what actually gets flagged when the Stanford words appear:

1. Symmetry. Five bullets, all 18 to 22 words, all starting with a past-tense power verb. Nobody writes that way twice in a row, let alone across three jobs.
2. Density. Two or more of the nine words in the same bullet ("Spearheaded a robust, intricate migration...") is a giveaway.
3. Em-dash frequency. Human writers use the em dash roughly once every 500 words. ChatGPT drops one every 50 to 80 words. In one test, 92% of 500 ChatGPT-generated resumes contained an em dash, while "delve" and "tapestry" appeared zero times. The em dash is a stronger tell than any single word.
4. Cover letter cross-voice. If your cover letter says "delve" and your resume says "leveraged" and your LinkedIn About section is empty, the triangulation fails.

Fix the rhythm, not the vocabulary. Vary bullet length. Let two of them be seven words. Let one be a full sentence. Kill the em dashes.

The 61%, the 75%, and the 91%: what the 2026 numbers actually say

The consolidated 2026 data is worse than most candidates realize, and it is now globally uniform. A US resume and a Berlin resume face effectively the same detection stack.

The two-point gap between the US (61%) and Europe (59%) is the real headline. AI detection is not a US quirk anymore; it is the global default. And the deception-caught rate jumped roughly 26 points in six months, a greater than 40% relative increase. Whatever worked in mid-2025 does not work now.

Refolk's index: nobody advertises they run this stack

The stealth is the story. In Refolk's index of professional profiles, only 31 US recruiters and talent acquisition professionals publicly mention "Workday" in their profile. Only 10 mention "Greenhouse" by name. Zero surface for the keyword "HiredScore." Zero surface for the phrase "AI screening."

Meanwhile 61% of US hiring managers say they use AI-detection software. That is a stealth adoption pattern with real implications:

• You cannot search LinkedIn for "the recruiter running my resume through HiredScore" and warm them up. They do not brand themselves that way.
• The employers using the heaviest AI screening (top employers among the 31 include Workway, Cohere, Raytheon, Accenture, and PeopleScout) do not publish which layer they are running.
• Advice like "just call the recruiter" assumes a visible human. The screening layer is deliberately invisible.

You cannot game a specific reviewer, because there often is not one until the fifth stage. You have to write for the machine and the eventual human simultaneously.

The mismatch that kills more resumes than any single word

A polished resume paired with a bare LinkedIn is now treated as fabrication evidence. That is the single most important sentence in this article.

Here is the mechanism. When the Workday Recruiting Agent (or Brainner, or Crosschq) scores your application, it does not just parse your PDF. It looks for corroboration:

• Do your job titles on the resume match the titles on LinkedIn?
• Do the dates match, or has the resume smoothed a six-month gap that LinkedIn still shows?
• Does the resume claim "spearheaded a 40-person migration" for a company where your LinkedIn shows two connections and no recommendations?
• Is there a photo, an About section, endorsements, activity? A "ghost" LinkedIn (no photo, low connections, no recommendations) attached to a sophisticated resume is a top fabrication signal per Enhancv's 2026 breakdown.

Date inconsistencies are the worst offenders because they read as AI gap-smoothing, and gap-smoothing is now the single behavior detection layers are tuned hardest against. If you used AI to "clean up" 2023 to look continuous on the resume but LinkedIn still shows the real end date at your last job, the triangulation fails and the file gets deranked silently.

The order of operations most candidates get backward:

1. Wrong order: Generate resume with AI, send it, forget LinkedIn.
2. Right order: Update LinkedIn first (dates, titles, one paragraph in About, a photo, 50+ connections), then generate the resume so it matches LinkedIn, not the other way around.

Why the AI detectors themselves are not the actual gatekeeper

The dirty secret is that AI-detection tools are unreliable and enterprise buyers know it. Stanford found detectors flagged roughly 61% of non-native-English essays as AI-written when humans wrote them. That false-positive rate is why the industry pivoted from detector-scoring to consistency-checking against external, verifiable data.

Which means the winning move is not to defeat the detector. It is to make the three artifacts (resume, LinkedIn, cover letter) sound like the same person and back each other up. That is what "hiring manager AI detection" actually looks like in 2026: cross-reference, not classifier.

A 45-minute fix that does not require abandoning AI

You do not have to write your resume by hand. You have to make the three documents corroborate each other and break the rhythm signals. Here is the sequence.

1. Fix LinkedIn first (15 minutes)

• Add a photo, a headline, and 3 to 5 sentences in About in your own voice.
• Make every job title and every date match what you plan to put on the resume, exactly.
• Get to 50 connections minimum. Below that, you look like a shell profile.
• Ask for one recommendation. One is enough to move you out of "ghost" territory.

2. Rewrite the resume against a specific posting (15 minutes)

• Vary bullet length deliberately. Aim for a mix of 7-word, 14-word, and 22-word bullets.
• Delete every em dash. Use commas or periods.
• Cap the Stanford nine at two total across the whole resume. "Spearheaded" once is fine; "spearheaded" twice plus "leveraged" plus "robust" is a cluster.
• Do not open more than two bullets in a row with the same verb tense pattern.

This is exactly what Refolk automates: paste the job posting, and Refolk rewrites your resume in a voice tailored to that specific role while scoring how well your history actually fits, so you know before you hit send whether the application is worth the slot.

3. Match the cover letter to the resume, not to the posting (15 minutes)

• The cover letter should sound like the person who wrote the resume. Same rhythm, same vocabulary range.
• If your resume says "built," your cover letter should not say "architected." That voice shift is what the triangulation catches.
• One anecdote that appears on LinkedIn (a project, a talk, a shipped feature) closes the loop across all three artifacts.

Refolk drafts the cover letter alongside the tailored resume, so the two documents come out of the same pass, in the same voice, and the cross-check finds corroboration instead of contradiction.

What to stop doing immediately

Three habits that used to be harmless and are now actively harmful:

• Stop pasting ChatGPT output raw. The em-dash frequency alone will flag you. Run one manual pass and delete them.
• Stop hiding prompt injections in white text. 22% of caught deception cases in Greenhouse's 2026 report involved hidden injections, and 42% of candidates admit trying them. The math says most go undetected, but the reputational damage of the few caught is what is driving vendors to bias against anything templated. You are betting against a rising detection curve.
• Stop applying without updating LinkedIn first. If you only have time for one document, it is not the resume.

Greenhouse CEO Daniel Chait told Fortune in November 2025 this is "the first time I can remember where both sides were unhappy." Candidates feel screened out unfairly; recruiters feel drowned in synthetic applications. LinkedIn now processes 11,000 applications per minute, a 45% surge driven by generative AI. The cross-reference against LinkedIn is not a temporary defense. It is the new default, and it is going to get stricter.

FAQ

Does Workday actually reject my resume, or just rank it lower?

Both, depending on which layer the employer licensed. Core Workday Recruiting does parse-and-filter, which mostly ranks and surfaces. HiredScore AI for Recruiting, licensed separately, does real screening and can effectively remove you from view. 75% of companies (per Resume.org, n=1,399) allow AI to reject candidates without human review, so at many employers "deranked" and "rejected" are functionally the same outcome.

If I use AI at all, will I get caught?

Not if the three artifacts corroborate each other. AI-detection tools are unreliable (Stanford found 61% false positives on non-native-English writing), which is why enterprise buyers moved to consistency-checking against LinkedIn instead. The candidates getting caught are the ones with polished resumes attached to ghost LinkedIn profiles, mismatched dates, or a cover letter in a completely different voice. Fix the corroboration and AI-assisted writing is not the problem.

Which of the nine Stanford words is worst?

None of them individually. The trigger is cluster density and structural symmetry, not any single word. Two of the nine in one bullet is worse than one of them appearing three times across the resume. And frankly, em-dash frequency is a stronger tell than the word list: humans use one every 500 words, ChatGPT drops one every 50 to 80 words. Kill the em dashes first.

Should I delete my LinkedIn to avoid the mismatch problem?

No, and this is a common wrong instinct in 2026. A resume with no LinkedIn to cross-reference is treated worse than one with a mismatched LinkedIn, because it removes the corroboration entirely and reads as evasion. The fix is a real LinkedIn (photo, dates matching the resume, 50+ connections, a few sentences in About) not a deleted one.