Blibli Store Review & Sentiment Analyzer

Collect buyer reviews from any Blibli store — your own or a competitor's — as clean text ready to feed straight into an LLM for sentiment analysis, complaint clustering, or product-strength mining.

Point it at a store and it returns the written reviews plus a summary of the store's entire review base: rating histogram, negative share, and Blibli's own topic breakdown (quality, price, shipping, packaging, service, authenticity).

No login, no API key. HTTP-only — no browser.

What makes it useful

It captures the complaints. Blibli only exposes ~600 reviews per query, and these stores run ~98% positive — so a naive scrape returns a wall of 5-star praise. This actor sweeps each star rating as its own query, which on the store used for testing returned 292 of the 294 one-star reviews that exist, out of a base of 68,842. Negative feedback is the scarce, valuable part, and it is what you actually get.

It only returns reviews with words. About half of Blibli's reviews are star-only ratings with no text — useless for sentiment work. Those are filtered out server-side, so your run (and your bill) is spent on reviews an AI can actually read.

The summary row alone may answer the question. One cheap record per store carries the topic counts across the whole review base — "what do customers keep bringing up" without spending a single AI token.

Input

{
  "storeUrls": [
    "https://www.blibli.com/merchant/colgate-palmolive-indonesia-flagship-store/COP-60023",
    "COP-60023",
    "https://www.blibli.com/p/sandisk-ultra-microsdxc-64gb/ps--BLL-70058-00122"
  ],
  "coverage": "deep",
  "sortBy": "LATEST",
  "onlyWithText": true
}

storeUrls accepts a store URL, a bare store code (COP-60023), a product URL, a bare product SKU, or brand:<name> — mixed freely, so your store and your competitors' go in one run.

Field	Description
coverage	standard (one sweep, ~600), deep (per-rating sweeps — captures nearly all negatives), maximum (adds per-topic sweeps)
sortBy	LATEST, POPULAR, OLDEST, HIGHEST, LOWEST
onlyWithText	Skip star-only ratings. On by default
onlyWithMedia	Only reviews with buyer photos/videos
minRating	Keep only reviews at or above N stars (0 = off)
maxReviewsPerTarget	Cap per store/product (0 = no limit)
includeSummary	Emit the store summary row. On by default
productBreakdown	Also sweep each product separately — each has its own ~600-review window
maxProductsPerStore	Cap for productBreakdown
proxyConfiguration	Apify Residential + country ID recommended

Output

Two record shapes, discriminated by recordType.

REVIEW — one per review, Blibli's own fields passed through verbatim plus flattened fields for AI pipelines:

{
  "recordType": "REVIEW",
  "id": "74f5b34c-6db2-41c0-b22e-bcb4de44ab56",
  "rating": 2,
  "reviewText": "Gak dapet kantongan tas IU nya",
  "ratingPolarity": "negative",
  "reviewLength": 32,
  "hasText": true,
  "verifiedBuyer": true,
  "productSku": "COP-60023-01124",
  "merchantName": "Colgate Palmolive Indonesia Flagship Store",
  "createdAtIso": "2026-07-27T14:39:13Z",
  "imageCount": 0,
  "videoCount": 0,
  "matchedTopic": null,
  "matchedRatingFilter": "1"
}

STORE_SUMMARY — one per store, describing the entire review base:

{
  "recordType": "STORE_SUMMARY",
  "merchantName": "Colgate Palmolive Indonesia Flagship Store",
  "totalReviews": 68842,
  "commentCount": 32843,
  "averageRating": 4.9,
  "negativeSharePercent": 0.72,
  "positiveSharePercent": 98.48,
  "ratingCounts": { "1": 294, "2": 203, "3": 549, "4": 2817, "5": 64979 },
  "topicCounts": {
    "QUALITY": 1181, "PRICE": 1006, "SHIPMENT": 935,
    "PACKAGING": 627, "SERVICE": 220, "ORIGINALITY": 213, "GENERIC": 776
  }
}

ratingPolarity is derived directly from the star rating (negative for 1–2, neutral 3, positive 4–5). It is not natural-language sentiment analysis — it is a cheap pre-filter so you can spend LLM tokens only on the reviews that matter.

matchedTopic is set when a review was surfaced by a topic sweep, meaning Blibli itself classifies it under that topic.

Feeding it to an AI

Filter to ratingPolarity == "negative" and hasText == true, then group by productSku, and pass reviewText to your model. The topicCounts in the summary row tell you which themes to expect before you start.

Known limits

• ~600 reviews per query slice. A store with 68k reviews cannot be extracted in full. deep and maximum coverage widen the sample, and productBreakdown opens a separate window per product — but the summary row is what describes the complete base.
• Reviews are in Indonesian. Use a multilingual sentiment model.
• Reviewer names arrive masked by Blibli (M***a). This actor does not attempt to unmask them, and returns only reviews Blibli shows publicly.
• brand:<name> targets are accepted but returned no results for every brand tried during recon.
• Roughly half of all reviews have no text; with onlyWithText on (the default) those never reach your dataset.

Reliability

Blibli sits behind Cloudflare. The actor clears it HTTP-only with TLS impersonation plus the Referer header Cloudflare's rule requires — no browser, no CAPTCHA solver. On a block it retries with exponential backoff while rotating both the TLS fingerprint and, when a proxy is configured, the egress IP.

A failing target never sinks the run: it emits an ERROR record with _error and _errorDetail and the remaining targets continue.

See CRAWLING_METHOD.md for the full endpoint analysis, including the decoy endpoint that reports zero reviews for products that have hundreds.