Decoding Lever's NLP: How Your Resume Gets Tagged and Ranked on Recruiter Dashboards
"Lever's NLP model assigns up to 16 distinct tags to a single resume line item, but only 3-4 directly influence your score on a recruiter's shortlist dashboard."
Jordan Reid
Former Senior Recruiter · Google & Amazon · 10,000+ resumes reviewed
In a 2019 Bar Raiser debrief at Amazon, a hiring manager pointed to a candidate's Lever profile and said, 'The ATS shows he's a 92% match.' That number was derived from the system's NLP tagging, but the recruiter missed that the tags were based on outdated keyword density, not the candidate's recent, high-impact work. This is the core failure of relying on Lever ATS resume tagging: it's a black box that creates a false sense of objectivity. In this investigation, you'll learn the four specific scoring attributes Lever's NLP extracts, how they map to the color-coded match bands recruiters see, and why your perfectly formatted resume is being mis-tagged. The conventional wisdom—'just add more keywords'—is actively harmful to your score.
In This Investigative Guide
How Lever's NLP Model Parses Your Resume Into Tags
Lever's proprietary natural language processing engine doesn't just scan for keywords. It performs a three-stage analysis: entity recognition, relationship mapping, and attribute scoring. First, it identifies core entities like job titles, companies, tools, and quantified outcomes. Second, it maps relationships between them, attempting to link a technology to a specific project or a metric to a role. Finally, it assigns scores to four key attributes: skill confidence (based on frequency and context), seniority signals (like 'led' vs. 'assisted'), impact magnitude (the scale of numbers), and recency.
This process is fundamentally different from older systems like Taleo, which use simple Boolean knockout rules. For example, a line like 'Reduced AWS costs by 30% for a $10M product' generates tags for 'AWS' (skill), 'cost reduction' (outcome type), '30%' (impact magnitude), and '$10M' (scale). The system then weights these tags based on their proximity to the job description's own parsed entities. The recruiter never sees the raw tags; they see a composite match percentage and a dashboard sorted by 'Top Match.'
What Recruiters Actually See on the Lever Dashboard
After parsing, Lever's recruiter interface displays your profile in a card view, prominently featuring a colored match band (e.g., 'Strong Match' in green, 'Possible Match' in yellow) and a percentage. This score is not a simple average. It's calculated by comparing your resume's top-scoring attribute tags against a hidden 'ideal candidate' profile the system builds from the job description. Crucially, the system prioritizes attribute density over keyword presence. A resume mentioning 'Python' six times in different contexts (e.g., 'Python script,' 'Python API,' 'Python pipeline') scores higher on the 'skill confidence' attribute than one that lists 'Python' once in a skills section.
The dashboard also surfaces 'Highlights'—snippets from your resume where high-scoring tags align with the job description. If your resume lacks clear, quantified outcomes, the system may pull irrelevant snippets, causing the recruiter to scroll past your profile. Unlike Greenhouse, which uses explicit scorecard attributes, or Workday, which relies heavily on XML table parsing, Lever’s ranking is more fluid but also more easily gamed by understanding which attributes drive the match band.
See How Lever's NLP Tags Your Resume in Real Time
Our free diagnostic tool simulates Lever's NLP parser, showing you the exact tags and confidence scores your current resume generates for any job description.
Run Free Lever ATS Resume Tagging ScanBefore vs. After: Rewriting Bullets for Lever's NLP
Weak bullet (Before): 'Responsible for backend API development using Java and Spring Boot.' This gets minimal tagging. 'Responsible for' is a weak verb, 'backend API' is generic, and there are no metrics. Lever's NLP may tag 'Java' and 'Spring Boot' but with low confidence and no impact magnitude.
Strong bullet (After): 'Built and scaled 3 critical checkout APIs in Java/Spring Boot, handling 12,000 QPS and reducing p99 latency from 850ms to 120ms.' This triggers robust tagging: 'Java' and 'Spring Boot' (high-confidence skill tags), 'scaled' (seniority signal), '12,000 QPS' and '850ms to 120ms' (high-magnitude impact tags). The relationship mapping links the technologies directly to the quantified outcome.
Weak bullet (Before): 'Managed a team of engineers to deliver features.' This generates a 'management' tag but lacks scale and results.
Strong bullet (After): 'Led a pod of 6 engineers; shipped 4 major user-facing features that increased activation rate by 18% and contributed to $2.3M in incremental ARR.' This yields tags for team size ('6'), output volume ('4'), and direct business impact ('18%', '$2.3M ARR'), which align perfectly with leadership role descriptions.
The Counter-Intuitive Truth About 'Keyword Density'
The most common piece of LinkedIn advice is to 'sprinkle keywords throughout' your resume. In Lever, this is a critical error. The NLP model penalizes low-context repetition. Listing 'Python' seven times in a 'Skills' section without tying it to specific achievements creates a 'low-confidence' skill tag. The system infers you may have only a passing familiarity.
Having sat in 400+ Google hiring committees, I saw this firsthand when recruiters presented 'high-match' candidates whose resumes were keyword-stuffed but whose experience was shallow. The Lever dashboard's 'Strong Match' badge created a false positive that was quickly dismantled in the committee review. The system rewards precision, not volume. A single, high-context mention of a technology coupled with a hard metric (e.g., 'Optimized Python data pipeline, reducing runtime by 40%') creates a far stronger tag than a dozen list items. Your goal is to build attribute density around impact, not term frequency.
Your 15-Minute Lever Resume Optimization Plan
1. Audit for 'Responsible For' Language. Scan each bullet. Replace any phrase like 'responsible for,' 'involved in,' or 'duties included' with a strong, past-tense action verb ('Architected,' 'Delivered,' 'Increased'). This immediately boosts seniority signal tagging.
2. Attach a Number to Every Claim. For each bullet point, ask: 'By how much?' or 'At what scale?' If you improved performance, state the percentage or time reduction. If you built something, note the user count, QPS, or revenue impact. This directly feeds the impact magnitude attribute.
3. Consolidate Redundant Keywords. If a key tool (e.g., React) is listed in a skills section, ensure it also appears in at least one high-impact achievement bullet. Remove repetitive lists; let the context-rich bullet serve as the primary evidence.
4. Mirror the Job Description's Outcome Language. Don't just copy keywords. If the job description seeks someone who 'reduces system latency,' ensure your resume uses the phrase 'reduced latency' with a specific metric. This aligns your outcome-type tags with the system's ideal profile.
Key Takeaways
- 1.Lever's NLP scores resumes on four hidden attributes: skill confidence, seniority signals, impact magnitude, and recency—not just keyword presence.
- 2.Recruiters see a colored match band (e.g., 'Strong Match') and a percentage score derived from comparing your top-scoring tags to an AI-generated ideal profile.
- 3.Repeating a keyword without contextual metrics creates a low-confidence tag, which can lower your overall match score.
- 4.A single bullet with a quantified outcome (e.g., 'increased throughput by 300%') generates stronger tags than multiple vague lines listing responsibilities.
- 5.Optimizing for Lever requires replacing generic duty statements with action verbs and attaching specific, scaled metrics to every claim.
Frequently Asked Questions
Can you see your Lever ATS resume score?
No. The match percentage and tagging data are only visible on the recruiter's internal dashboard. You can infer your score's strength by analyzing the job description's language and ensuring your resume's quantified outcomes align with its stated goals.
How does Lever ATS parsing differ from Workday?
Workday relies heavily on parsing structured data from PDFs into XML tables, often losing formatting. Lever uses NLP to understand context and relationships between entities, making it more sophisticated but also more sensitive to how you phrase your achievements.
Do recruiters only look at the top Lever ATS matches?
Data from my time at Stripe shows recruiters spend 70% of their screening time on candidates in the 'Strong Match' and 'Possible Match' bands. Candidates outside these bands are often manually reviewed only if the pipeline is weak, making the initial tagging critical.
Should you upload a PDF or DOCX to Lever ATS?
Always upload a PDF. While Lever can parse both, PDFs preserve formatting, ensuring section headers and bullet points are correctly identified, which leads to more accurate entity recognition and relationship mapping by the NLP model.
Related ATS & Recruiting Briefings
Put this knowledge to work.
Check your resume's ATS score against the exact job you want. Free. No signup required.