Automatiser la revue d'offres d'emploi avec Claude Code

Les agences de recrutement perdent des candidats quand l’offre attire mais ne répond pas au salaire, lieu, télétravail, contrat, évaluation et tâches. Le risque d’affichage compte aussi. Depuis avril 2024 au Japon, certains éléments de conditions doivent être indiqués lors du recrutement. La revue doit donc lier conversion et contrôle.

Key points

• Review labor conditions before persuasive copy.
• Use Claude Code to table missing fields, candidate questions, risky phrases, and JobPosting mismatches.
• Humans keep legal, salary, age, gender, contract, and client approval decisions.
• Concrete work details beat vague culture words.
• Measure revision loops, client returns, application rate, interview rate, and decline reasons.

Workflow from intake to application

Place the client job post, recruiter notes, hiring background, selection flow, decline reasons, and candidate questions in one table. Job post review is not proofreading; it is information control.

Remove candidate names, current employer names, interview evaluations, salary negotiation notes, and confidential client information before using AI.

Input	Output	Human review
Client post	role, salary, location	labor conditions
Recruiter note	appeal and background	client confirmed
Candidate questions	anxiety and comparison	no personal data
Board copy	title and tags	risky wording
JSON-LD	title, salary, location	visible text match

What Claude Code handles and what humans review

Claude Code extracts missing fields, vague phrases, candidate questions, JobPosting mismatches, and client confirmation questions. It turns culture claims into work, onboarding, required skills, and evaluation criteria.

Humans review law and fact: age exception, gender wording, salary range, overtime, location-change scope, renewal criteria, and client approval.

Official references: MHLW April 2024 display additions, MHLW age restrictions, MHLW worker recruitment ads, Google JobPosting.

3 Use case

Use case 1: Replace vague appeal

Input job text, questions, and decline reasons. Output vague phrase, candidate question, and concrete replacement.

Use case 2: Flag age and gender wording

Input title, requirements, preferred skills, and media tags. Output red flags, reasons, alternatives, and human review items.

Use case 3: Align JobPosting

Input page text, JSON-LD, and board copy. Output differences in title, location, salary, employment type, and dates.

Copy-paste prompt

Act as a recruitment agency job posting reviewer. Find missing fields, vague phrases, age/gender/misleading red flags, candidate-facing information to move higher, JobPosting mismatches, and client confirmation questions. Do not make legal conclusions or guess salary, location, age, gender, or contract terms.

Working check code

const jobPostDraft = {
  clientName: "サンプルSaaS株式会社",
  roleTitle: "カスタマーサクセス",
  businessScope: "導入支援、利用状況の確認、更新提案",
  workLocation: "東京都渋谷区 / リモート週2日",
  locationChangeScope: "本社、会社が定める拠点、在宅勤務場所",
  workChangeScope: "カスタマーサクセス、営業企画、サポート業務",
  salaryRange: "年収 4,800,000円〜6,500,000円",
  contractRenewal: "期間の定めなし",
  workingHours: "9:30〜18:30 / 休憩60分",
  overtime: "月平均20時間",
  requiredSkills: ["法人顧客対応", "SaaSまたはITサービスの利用経験"],
  preferredSkills: ["SQLの基礎", "Salesforce利用経験"],
  reviewText: "B2B customer success role. The person handles onboarding, usage review, renewal proposals, and customer questions.",
  ctas: ["candidateApply"],
};

const riskyPhraseExamples = ["20s welcome", "women only", "men only", "young team only", "guaranteed growth"];

const requiredFields = [
  "clientName",
  "roleTitle",
  "businessScope",
  "workLocation",
  "locationChangeScope",
  "workChangeScope",
  "salaryRange",
  "contractRenewal",
  "workingHours",
  "overtime",
];

const missing = requiredFields.filter((field) => !jobPostDraft[field]);
const draftText = jobPostDraft.reviewText;
const riskyHits = riskyPhraseExamples.filter((phrase) => draftText.includes(phrase));
const hasSalary = /円|年収|月給|時給/.test(jobPostDraft.salaryRange);
const hasOneCta = jobPostDraft.ctas.length === 1 && jobPostDraft.ctas[0] === "candidateApply";

if (missing.length || riskyHits.length || !hasSalary || !hasOneCta) {
  console.error("求人票レビューで人が見るべき項目があります。");
  console.error({ missing, riskyHits, hasSalary, hasOneCta });
  process.exit(1);
}

console.log("Recruitment agency job posting review checklist is ready for human review.");

The code catches missing fields, risky phrases, salary visibility, and CTA focus before publication.

Pitfall: common mistakes and fixes

Only chasing applications
Split appeal, conditions, risk, and structured data.

Keeping young-team wording
Replace age flavor with tasks, onboarding, and required skills.

Showing salary top only
Confirm the experience and role behind the top range.

JSON-LD newer than text
Check board copy, visible text, and JSON-LD together.

ROI metrics

Track revision count, client return loops, candidate follow-up questions, application rate, interview rate, decline reasons, and days to publish. If applications rise but interview rate falls, the post attracts too broadly.

FAQ

Q. Remove all age or gender wording?
No. Flag it and let humans review official guidance and exceptions.

Q. Can salary be emphasized?
Only with conditions for the upper range.

Q. Do agencies need JobPosting?
If they host job pages, visible text and JSON-LD should match.

Consultation path

Bring job posts, decline reasons, candidate questions, JSON-LD, and client review comments to 研修・相談. See also care recruitment page.

Checked result

Official MHLW guidance, age limits, recruitment ad display guidance, Google JobPosting, and the Node.js code were checked. Start by splitting one job post into labor conditions, candidate anxiety, red flags, and JobPosting differences.

Français readers should start with one job post, one risk table, one client confirmation list, and one CTA. Keep legal and salary judgment with humans.