Tailor Your Resume to a Job Description with AI — Without Inventing Numbers

Who this is for / not for

For you if: you have real work experience, you apply to roles with detailed JDs, and you are willing to spend ten minutes supplying actual numbers.

Not for you if: you are a new grad (the model will inflate one internship into three "led" projects), or you want one master resume to blast everywhere — the whole point of tailoring is one version per role.

What to prepare

1. The full JD, pasted verbatim — not your summary of it.
2. A fact sheet, in this shape. The critical part is that you supply the numbers:

Experience: company / title / dates / 3-5 responsibilities
Metrics: the real number behind each bullet (scale, delta, hours saved, revenue) — or "none"
Skills: what you can do, and at what level (independent / used it / aware of it)
Education & links: degrees, certifications, portfolio
Target: company + role

Steps

1. Install the skill

git clone --depth 1 --filter=blob:none --sparse https://github.com/anbeime/skill
cd skill && git sparse-checkout set skills/tailored-resume-generator

Drop the skills/tailored-resume-generator/ directory into your skills folder (~/.claude/skills/, ~/.codebuddy/skills/, or your client's equivalent). No dependencies, no API key.

2. Make it analyze the JD before writing anything

Here is a job description. Do two things only — do NOT start writing the resume yet:

1. Build a table of hard requirements with three columns:
   Requirement | Do I match | Evidence (which experience proves it)
   Write "no evidence" where there is none. Do not infer.
2. List keywords that appear 2+ times in the JD (these are what ATS parses for).

JD:
[paste JD]

My background:
[paste fact sheet]

Read the "no evidence" rows first. They tell you whether to gather more material or skip this role.

3. Generate with an explicit no-invention rule

Using the match table above, write the resume in Markdown:

1. Use only what I supplied. Any number I did not provide becomes [TO FILL: what is needed].
   No estimates, no "significantly improved" as a substitute for a metric.
2. Map JD keywords only to things I actually match. Put unmatched items in a
   separate "Gaps" section instead of padding my skills list.
3. Structure each bullet as action + object + result, with my real numbers on the result.
4. Finish with a "Gaps" list of this resume's 3-5 genuine weaknesses versus this JD.

4. Human pass (mandatory)

• Search for [TO FILL and fill every one; if you cannot, rewrite the bullet instead of guessing
• For every skill listed, you can name where you used it — otherwise cut it
• Dates are continuous and correct (models mis-arithmetic ranges surprisingly often)
• Your job titles match your actual titles, not the model's upgraded versions
• One page for <10 years of experience in the US market; two pages only for research-heavy or academic CVs

What we actually observed

We ran the workflow against a data-analyst JD (retail/e-commerce, required SQL, Python, A/B testing, Tableau or Power BI). With the no-invention rule in place, the output kept three honest placeholders (coverage scope, campaign count, hours saved) and flagged three real gaps: no rigorous A/B experience, no Tableau/Power BI, no metrics supplied.

Without the rule, the same material produced confident fiction. The skill's own bundled example does exactly this: the input says only "Built Python scripts for data automation" and "Worked with marketing team on campaign analysis" — no figures anywhere — and the output claims

reducing manual data processing time by 60% improving campaign ROI by 35%

That is the failure mode to design against, not a bug you can prompt away politely once.

Traps

Invented metrics. See above. Placeholders beat plausible numbers that collapse under interview follow-up.

Inflated scope. In the same bundled example, a one-year internship becomes "Healthcare industry experience" in the summary line. Watch the verbs: "assisted" quietly becomes "led".

Keyword stuffing. Models pad the skills section to raise JD match. ATS may pass you; the interview will not. List matched skills only, and keep unmatched ones in the Gaps section — it reads as self-aware.

Length creep. Models default to comprehensive. Cap the page count explicitly and cut the "Additional" section before cutting quantified results.

Alternatives

Notes for English-language applications

A few conventions differ from Chinese-language resumes and are worth knowing before you send:

• No photo, age, gender, or marital status. Including them is a red flag at US/UK employers.
• One page is the default, including for senior engineers; two pages is acceptable for academic CVs or research-heavy profiles. This is different from the two-page norm in China.
• Standard section headings (Experience, Skills, Education) parse better through ATS than creative ones (What I've done, My toolkit).
• The XYZ formula works well as a bullet pattern: "Accomplished [X] as measured by [Y] by doing [Z]". It forces a metric into every line — which is exactly the thing the model will fake, so supply the numbers.
• Mirror the JD's job title in your summary if it matches your actual level; recruiters search by title.

FAQ

Is using AI on my resume cheating?

Rewriting expression is fine; inventing content is not. Test: can you talk for five minutes about every line on the page? If not, cut it or rewrite it honestly. The placeholder rule enforces exactly this.

Will ATS flag my resume as AI-generated?

ATS parses and ranks keywords; it does not detect authorship. The real risk is keyword stuffing — padding skills you do not have — which passes the machine and fails the interview.

Do I need a separate version for every role?

Layer it: full workflow for the 3-5 roles you want most; for the rest, adjust only the summary and the ordering of skills. One generic resume across twenty directions performs worse than five targeted versions.

One page or two?

For US/UK applications: one page, even at senior level, unless you are in academia or research. State the limit in the prompt; otherwise the model defaults to "comprehensive".