Yellow Pages Indonesia Business & Lead Scraper

Extract Indonesian business listings from Yellow Pages ID — name, phone, email, website, full postal address and GPS coordinates — plus B2B products, per-province market density and a province/city directory. Pure HTTP, no browser, no API key, no login.

---

Modes

Mode	Input	Emits	What it is for
search_businesses	queries (+ region / city)	BUSINESS	The general lead builder. Keyword search across the whole directory.
category_businesses	queries = category slugs (+ city)	BUSINESS	The curated "top suppliers" block of a /places/<slug> page. Small but hand-ranked.
company_details	companyUrls	COMPANY	Full profile: labelled phone numbers, founding year, legal form, categories.
search_products	queries	PRODUCT	B2B product offers with price, seller and the seller's real shop URL.
market_insights	queries	MARKET_INSIGHT	Listing counts per province and city for a keyword, plus related keywords. One request.
geo_directory	provinceCodes	GEO_NODE	Provinces and their cities — the seed list for a systematic nationwide crawl.

Set enrichWithDetails: true on the two business modes to merge each company's profile page into its listing row (one extra request per company).

Why one actor with modes rather than six actors

Every mode hits the same Django application, through the same host choice, with the same pagination ceiling and the same record envelope. The host choice (below) is the fragile part of this scraper — splitting it across six actors would mean six copies to re-verify whenever the site's WAF configuration changes. Modes that emit different shapes stay separable through recordType, which the dataset schema exposes as its own named views.

---

How it gets the data

www.yellowpages.id puts a Cloudflare Managed Challenge on exactly the three paths that carry the directory data:

Path	yellowpages.id	yoys.id
/ , /product-* , /citymap-*	200	200
/profil-* (company profiles)	403 cf-mitigated: challenge	200
/places/* (categories)	403 cf-mitigated: challenge	200
/listing/* (search)	403 cf-mitigated: challenge	200

All 18 curl_cffi TLS profiles — chrome99_android through chrome136, safari15_5–18_0, edge99/edge101, firefox133/firefox135 — return the identical 403 on the gated paths while the homepage returns 200 from the same IP and the same session. A homepage warm-up issues no clearance cookie. That makes it a path-scoped WAF rule, not a JA3 fingerprint gate, and it is not solvable HTTP-only.

www.yoys.id is the same application serving the same Indonesian dataset — YP Media Ltd runs both brands off one backend. Verified by resolving identical IDs on both hosts: product-…_52399 returns company 492034 on each, and profil-551113 is the same company on each, byte-comparable apart from brand strings. yoys.id applies no challenge on any path.

So the actor reads from yoys.id and writes the matching www.yellowpages.id URL into yellowpagesUrl on every record, so results stay citable against the brand you asked for. If yoys.id ever acquires the same rule the client fails over to yellowpages.id automatically and rotates through the TLS ladder before giving up.

No IP-based blocking was observed, so the proxy is off by default — turn on Apify Proxy only for very high volume.

---

Known limits

These are properties of the site, measured rather than assumed. They are the things that will surprise you if you do not know them.

• 250 rows per query, hard. The result header advertises figures like "10000 hasil", but the backend serves only 10 pages of 25. Page 11 answers HTTP 200 with a fully rendered, empty page — not a 404 — so anything that pages on status code alone will loop silently. To go deeper, slice the same keyword by region or city; each slice gets its own 250. Run market_insights first to see where the volume actually is.
• The free-text location box is a decoy. The site's own l= parameter returns zero results for every value while echoing the place name in the page title. This actor ignores it and uses the facet parameters (adm / cty) that the sidebar actually uses.
• Promoted listings ignore the region filter. A sponsored entry from another province can appear in a filtered result set. Filter on address.region downstream if that matters.
• Two kinds of entry share the directory. entryType: "COMPANY" has a claimed profile page and a companyId; entryType: "PHONE_LISTING" is a phone-book record with a number but no profile page, so companyId is null and enrichment skips it.
• Profile pages carry no coordinates. Geo comes from listing pages only, so enrichment preserves the listing's geo rather than overwriting it.
• Emails are rare. Most listings publish none. Where one exists it is behind Cloudflare's email obfuscation, which this actor decodes.
• Province codes are the site's own, not Indonesia's official BPS numbering (02 is Bali, not North Sumatra). Unused codes return a 200 page titled "Aceh" with zero cities; those are reported as hollow_response errors rather than pushed as empty provinces.
• Missing companies return 302 → /410, not 404. Reported as not_found.
• Fields are passed through as the site publishes them, so occasional upstream junk (an email in a website field, foundedYear: "10") appears verbatim rather than being silently cleaned.

---

Output

Every record carries the house envelope — _input, _source, _scrapedAt, recordType — on top of the payload. Failures become ERROR rows with _error / _errorDetail instead of vanishing, and truncated queries carry a _warning.

{
  "_input": "percetakan",
  "_source": "S1-listing-search",
  "_scrapedAt": "2026-09-20T15:23:18Z",
  "recordType": "BUSINESS",
  "entryId": "365207",
  "entryType": "COMPANY",
  "companyId": "365207",
  "name": "Percetakan Raja Setting",
  "description": "Rajasetting merupakan Percetakan Offset dan Digital Printing…",
  "phone": "+62 81572606669",
  "emails": [],
  "website": "https://rajasetting.com",
  "address": {
    "street": "Jl. Babakan H. Tamim No. 25",
    "locality": "Bandung",
    "region": "Jawa Barat",
    "postalCode": "40125",
    "country": "Indonesia"
  },
  "geo": { "latitude": -6.90525, "longitude": 107.647461 },
  "detailUrl": "https://www.yoys.id/profil-365207-percetakan-raja-setting-bandung.html",
  "yellowpagesUrl": "https://www.yellowpages.id/profil-365207-percetakan-raja-setting-bandung.html"
}

Measured completeness on a 20-row percetakan / Jawa Barat run: phone 20/20, website 20/20, coordinates 20/20, postcode 19/20, region 16/20, street 15/20.

---

Recipes

Build a city lead list

{ "mode": "search_businesses", "queries": ["percetakan"], "city": "Bandung",
  "enrichWithDetails": true, "maxItemsPerQuery": 250 }

Find where the volume is before spending requests

{ "mode": "market_insights", "queries": ["bank"] }

Then feed the returned regions[].filterValue back in as region to collect 250 rows per province instead of 250 nationwide.

Nationwide sweep — run geo_directory once for the city list, then iterate search_businesses over city values.

---

Stack

curl_cffi (TLS impersonation, chrome124 primary with a 5-profile ladder) + selectolax for parsing. Extraction follows JSON-LD first, HTML selectors only where JSON-LD does not reach. Listing rows merge both layers per field, because neither is complete on its own.

Retries are exponential with jitter; 429/5xx back off, challenges rotate TLS profile then host. A shape change raises UnexpectedShape and fails the run loudly rather than emptying the dataset.