LEVER ATS · ANTHROPIC · UPDATED MAY 2026

    How to beat the Lever ATS at Anthropic

    The format, keyword, and structure rules that get past Anthropic's Lever setup, plus a free resume review tailored to any Anthropic JD.

    Updated May 2026·9 min read

    QUICK ANSWER

    Anthropic uses Lever as its ATS for engineering, research, and operations hiring. Lever is more lenient than Workday on format (two-column sometimes parses cleanly, decorative fonts often work) but the recruiter screen is famous for weighting writing quality and concrete-impact signal. To get past it: tight single-column resume, named ML/LLM frameworks if relevant, end-to-end project ownership stories, clear evidence of safety-aligned thinking for research roles. The widget below scores yours against the actual Anthropic JD in 30 seconds.

    SEE THE OUTPUT, THEN SCORE YOUR OWN
    EXAMPLE OUTPUTCMU CS student, Anthropic research engineer JD
    YOUR ATS SCORE
    90/ 100, Anthropic-ready
    +27 from 63 before edits
    CriticalCritical (2)
    EXPERIENCEConcrete-impact rewrite
    Applied
    Worked on a machine learning project at a startup.
    Built an evals harness for a 7B-param customer-support model (Python + transformers + vLLM serving), defined 12 task-specific eval suites including 3 red-team prompts, surfaced a 14% factual-accuracy gap between top-p 0.9 and 0.5 sampling that informed our default config.
    Why: Anthropic weights writing quality and concrete impact. Named frameworks (transformers, vLLM), named techniques (top-p sampling), and a measurable outcome lift this from generic to senior.
    PROJECTSSafety-aligned signal
    Applied
    Built a chatbot for fun.
    Built and red-teamed a Llama-3.2-8B fine-tune for a domain-specific tutoring task; wrote a 240-prompt adversarial test set (jailbreaks, prompt injections, off-topic redirects), measured 89% safe-response rate before and 96% after applying constitutional-AI-style critique loop.
    Why: Surfacing the safety mechanic (red-teaming, adversarial prompts, constitutional AI) is what Anthropic looks for. Same project, different framing.
    NotableWorth fixing (1)
    EDUCATIONOpen-source visibility
    Applied
    CMU, Computer Science · GPA 3.9
    CMU '27, B.S. Computer Science · GPA 3.92 · Expected May 2027 · Coursework: 15-440 Distributed Systems, 11-785 Intro to Deep Learning, 10-708 Probabilistic Graphical Models, 15-840 Foundations of LLMs · Open source: vllm-project/vllm (8 merged PRs), github.com/[name]/evals-harness (1.2K stars)
    Why: Anthropic recruiters scan the Education block for the canonical LLM/systems courses and for verifiable open-source contributions. Naming both lifts first-pass relevance significantly.

    Example output. Your real review uses your own resume and the JD you paste.

    1Upload your resume (PDF)
    Drop PDF here, or click to browse
    10MB max. Text-based PDFs only.
    2Paste a job URL, role name, or the full job description

    ATS BY THE NUMBERS

    Lever

    the ATS Anthropic uses for engineering, research, and operations hiring

    Writing

    the differentiator the Anthropic recruiter screen weights most heavily

    30 sec

    what the widget above takes to score your resume against the JD

    Your resume meets Anthropic's bot before it meets the recruiter.

    Anthropic runs every application through Lever, which scores resumes against the JD for keyword match, format, and section structure before any human sees them. The bot reads top-to-bottom, maps your text into structured fields, and scores. The way to "beat" it is to be one of the resumes the recruiter's filter surfaces, which means matching the JD keywords precisely and being parseable as structured data. The widget above runs that scoring on your resume against any Anthropic JD, in 30 seconds.

    What the widget checks for a Anthropic JD

    Writing quality is the screen

    Anthropic recruiters explicitly weight resume bullet clarity. Vague marketing-toned bullets lose to concrete scoped ones.

    ML / LLM vocabulary signal

    If applying to research or ML engineering: named frameworks (PyTorch, JAX, vLLM, transformers), named training paradigms (RLHF, constitutional AI, sparse autoencoders), and named eval benchmarks.

    Safety-aligned thinking

    For research roles, Anthropic looks for evidence you think about safety as a first-class concern: red-teaming, interpretability, robustness, evals. Surface any of these explicitly.

    End-to-end ownership signal

    Anthropic engineering JDs weight projects where you owned design, implementation, and operation. "Shipped X" beats "contributed to X."

    Lever-safe formatting

    Lever parses two-column reliably and tolerates decorative fonts. Single-column is still safer if you also apply to firms on Workday.

    Open-source weight

    Anthropic recruiters notice public AI/ML contributions (a paper, an evals harness, a vLLM optimization, a popular HuggingFace model). Surface in the header.

    No "passionate about AI" language

    Anthropic recruiters skim past "I am passionate about AI safety" / "drawn to mission-driven companies." Replace with one specific thing you have built or read.

    How the free review works

    1. Upload your resume

    DOCX or text-selectable PDF only. Image-based PDFs cannot be read by any ATS. 10MB max.

    2. Paste the job description

    Full JD text or the URL of the posting. The score is tailored to that exact JD.

    3. Apply the rewrites

    Critical and Notable edits are grouped by severity. Each shows the original, the rewrite, and which keyword or formatting rule it fixes.

    4. Download the new PDF

    The preview rebuilds your resume live as you accept edits. Single-column, Workday-safe, ready to submit.

    NOT APPLYING YET?

    Get the weekly ATS-keyword drop

    Every Monday: the specific ATS keywords showing up in newly posted JDs at Goldman, McKinsey, Google, and 20 other firms. Free, no spam.

    Frequently Asked Questions

    What ATS does Anthropic use?

    Lever for engineering, research, and operations hiring. The same Lever instance is used across all teams.

    Is Lever as strict as Workday?

    No. Lever parses two-column reliably and tolerates decorative fonts. The recruiter screen is the binding constraint at Anthropic, not the parser.

    What does Anthropic look for in a research engineer resume?

    Named ML/LLM frameworks (PyTorch, JAX, transformers, vLLM), named training paradigms, named eval methodologies, and evidence of safety-aligned thinking (red-teaming, interpretability, robustness).

    Do I need a published paper to get a research role?

    Not strictly. Strong open-source ML/eval contributions (a popular HuggingFace model, a well-known evals harness, vLLM optimizations) substitute for papers in many recent hires.

    How long does Anthropic's interview process take?

    ~4 to 8 weeks from application to offer. The take-home (technical screen for engineering, paper-discussion for research) is the highest-leverage round.

    Does Anthropic do referrals?

    Yes, and they matter. A referral moves you to a recruiter-read tier within ~5 days. Use Offerloop's Find feature to identify Anthropic employees from your university.

    Is the widget really free?

    No catch. Upload your resume, paste the Anthropic JD, get the score and rewrites without an account.

    How does this differ from OpenAI?

    Similar bar on technical depth. Anthropic weights safety-aligned thinking more heavily; OpenAI weights distribution and product surface. Tune the resume accordingly.

    WHEN YOU ARE READY FOR THE FULL TOOLKIT

    Score your resume, then reach the alum who already got in

    Once your ATS score clears 80, Offerloop helps you find a USC, NYU, Michigan, or UPenn alum at the firm you applied to, drafts the cold email, and tracks the reply.

    Create your free Offerloop account

    Free tier: 3 contacts per search, 2 interview preps, no credit card.