Claude Code로 부동산 매물 CSV 업데이트 반자동화하기: 게시 전 점검

부동산 매물 업데이트에서 가장 위험한 일은 느린 게시가 아니라 이미 없거나 정보가 틀린 매물을 계속 공개하는 일입니다. 가격은 바뀌었는데 관리비는 예전 그대로이고, 사진 폴더는 있지만 평면도가 빠져 있을 수 있습니다. Claude Code는 소개문 작성보다 먼저 게시 전 점검에 써야 합니다.

Duplicate property pages and copied portal text can weaken trust and search performance. Google’s guide to canonical URLs is the baseline, and people-first content is the editorial direction. For Claude Code permissions, see settings and hooks.

Key Takeaways

• Check CSV fields, publication status, photos, URLs, and inquiry links before writing listing copy.
• Claude Code can flag stale published rows, missing fields, and canonical mismatches.
• Humans verify price, availability, owner data, key notes, and advertising permission.
• Agency pages need local context beyond copied portal text.
• Track stale listings, missed price changes, missing photos, wrong canonicals, and wrong CTA links.

Workflow From CSV To Publication

Stage	Review target	Stop condition
CSV received	columns, listing ID, update date	required column missing
Data review	price, area, address, station	blank, zero, stale date
Image review	exterior, interior, floor plan	missing folder or floor plan
SEO review	title, description, canonical	duplicate title or wrong canonical
Inquiry review	CTA, form, phone	outdated destination

This table gives Claude Code a bounded job. It is not asked to sell the property. It is asked to stop rows that are not ready.

Claude Code And Human Review

Claude Code handles repeatable checks: CSV column inventory, stale published rows, missing image notes, duplicate titles, canonical mismatch, and outdated inquiry links. Keep private phone numbers, owner names, key locations, and negotiation notes out of the input.

Humans keep the publication decisions. Price, area, availability, ad permission, and owner approval cannot be guessed. The same rule appears in permission decision logs and safe autonomy ladder.

Three Use Cases

Use case 1: CSV field review

Split columns into required, optional, and forbidden. Required fields include listing ID, status, price, address, station, update date, inquiry destination, and canonical URL. Forbidden fields include owner information, private phone numbers, key notes, and unpublished negotiation notes.

Use case 2: Stale listing detection

Published rows with old update dates should appear in a daily review. This is not for blaming agents. It gives the team a calm list: unpublish, confirm price, add photos, or keep live.

Use case 3: Duplicate URL and canonical cleanup

The same property may appear in station pages, branch pages, campaign pages, and listing pages. Group URLs by listing ID and mark the preferred URL. That helps SEO and also gives sales staff one URL to send.

Copy-Paste Prompt

Act as a listing-operations editor for a small real estate agency.
Create a pre-publication check from a property CSV. Do not invent price, area, owner data, or availability.

Input:
- CSV column names:
- current property page URL:
- inquiry destination:
- conditions for unpublishing:

Return:
1. required, optional, and forbidden CSV fields
2. stop-before-publish rules
3. duplicate URL and canonical checklist
4. missing photo, floor plan, and neighborhood checks
5. inquiry CTA copy
6. final human review table

Working Check

const listings = [
  {
    id: "A-101",
    title: "南向き2LDK 駅徒歩7分",
    status: "published",
    price: 128000,
    address: "東京都サンプル区",
    station: "サンプル駅",
    updatedAt: "2026-07-19",
    canonicalUrl: "/properties/a-101/",
    sourceUrl: "/properties/a-101/",
  },
  {
    id: "B-204",
    title: "",
    status: "published",
    price: 0,
    address: "東京都サンプル区",
    station: "サンプル駅",
    updatedAt: "2026-07-10",
    canonicalUrl: "/properties/b-204/",
    sourceUrl: "/rent/b-204/",
  },
];

const today = new Date("2026-07-19");
const required = ["id", "title", "status", "price", "address", "station", "updatedAt", "canonicalUrl"];

const report = listings.map((listing) => {
  const missing = required.filter((field) => !listing[field]);
  const daysOld = Math.floor((today - new Date(listing.updatedAt)) / 86400000);
  return {
    id: listing.id,
    missing,
    stale: listing.status === "published" && daysOld > 5,
    canonicalMismatch: listing.canonicalUrl !== listing.sourceUrl,
  };
});

const failures = report.filter((row) => row.missing.length || row.stale || row.canonicalMismatch);

if (failures.length) {
  console.table(failures);
  process.exit(1);
}

console.log("Listing CSV is ready for publication review.");

This script is a small gate. It does not verify the property itself, but it catches missing fields, stale published rows, and canonical mismatch before humans spend time on copy.

Pitfall: Common Mistakes

Pitfall 1: sending internal data to AI. Remove owner names, key notes, private phone numbers, and negotiation memos.

Pitfall 2: writing copy before checking status. A beautiful listing with the wrong price is still a publication failure.

Pitfall 3: copying portal text unchanged. Add local viewing notes, neighborhood context, and inquiry guidance from real agency knowledge.

Pitfall 4: relying on memory for unpublishing. A daily stale listing report is easier than apologizing after inquiry.

ROI Signal

Measure stale listing count, missed price updates, missing photos, canonical mismatches, wrong CTA links, page views, and inquiries. These numbers show whether the publishing workflow is creating trust or creating avoidable support work.

CTA

If your agency has recurring listing update mistakes, bring CSV columns, public URLs, inquiry forms, and unpublish rules to training and consultation. For listing copy, read real estate listing descriptions.

What Happened When I Tried It

I modeled a listing CSV with ID, status, price, address, station, update date, and canonical URL. The Node.js check detected blank title, zero price, stale published rows, and canonical mismatch. The strongest lesson is that Claude Code creates value before copywriting: it stops listings that should not be published yet.