1688 Sourcing Intelligence

Search 1688.com and get back what an experienced sourcing team actually looks at before committing money: whether the category is worth entering, how each supplier compares to its own category's norms, and where the weak spot is that a headline score hides.

No login. No invented scores. Everything in English.

Most 1688 scrapers answer "what's for sale". This one answers "what should I do".

A list of products with 50 fields still leaves you the hard part. A ¥3 phone case tells you nothing on its own — ¥3 when the category median is ¥8, from a supplier whose repeat-buyer rate is 14 points below its peers, whose logistics score is 1.6 while its headline is 4.0 is a decision.

That comparison requires seeing the whole category at once. It cannot be produced by scraping one product at a time.

What you get that a field dump can't give you

Field	Why it changes a decision
categoryPriceIqrRatio	Price spread of the category. Near 1–3: everyone competes on price, no room to differentiate. Above ~5: cheap and premium coexist, so you can position.
priceVsCategoryPct	This product against its own category's median — not an absolute price you have no reference for.
pricePercentile	Exactly where it sits in the price distribution.
repeatRateVsCategoryPts	Repeat business relative to category norms. 40% is strong in one category, average in another.
weakestServiceArea + weakestServiceScore	Names the collapsed dimension. A 4.0 overall can hide logistics at 1.6.
supplierKeywordSpread	How many of your searched categories this seller appears in. A seller across unrelated categories is a reseller, not a specialist.
legalEntityHint	What the registered company name implies — factory, trading company, e-commerce reseller, sole trader.
bizTypeMatchesLegalName	false when a supplier calls itself a manufacturer but is registered as a trading entity.
shopNameMatchesLegalEntity	false when the shopfront name and the registered business differ.
customizationOffered	Whether the supplier offers customisation — the starting point for any private-label plan.

Real example from a live run: a supplier appearing in both data cables and screen protectors, self-declared Manufacturer, registered as 深圳市阚及贸易有限公司 — "trading company" written into its own legal name.

We search in Chinese, so you don't get the tourist price

Type "usb cable" into 1688 and you land in the slice built for cross-border resellers. Type 数据线 and you're in the domestic wholesale market where the factories are. Same platform, different world — and nothing tells an English-speaking buyer which one they're looking at.

Measured on the same three products, English term vs. Chinese term, 40 listings each:

	phone case → 手机壳	usb cable → 数据线	screen protector → 钢化膜
Median price	¥3.9 → ¥3.9	¥8 → ¥2.3	¥4.6 → ¥1.82
Share that are factories	13% → 65%	80% → 85%	60% → 55%
Listings with sales data	85% → 90%	80% → 100%	60% → 100%
Listings with inspection badges	1 → 5	11 → 19	4 → 12

So English keywords are converted to the term Chinese wholesalers actually type before the search runs. keyword keeps what you entered; searchedAs shows what was actually searched, so you can always check the conversion. Turn it off with searchInChinese: false.

This needs an API key (the same free-tier key used for translation). Without one, your English term is searched as given and the run log says so — we don't guess at translations, because a machine-translated keyword once turned "yoga mat" into "wet wipes", and you would never have known.

The supplier's official 1688 record

Alongside every listing, in English:

Field	What it is
supplierYears	Years holding a paid, identity-verified 1688 membership (诚信通)
repeatBuyerRate	Share of buyers who ordered again within 90 days (回头率)
serviceScoreOverall + 5 dimensions	1688's official merchant service rating (新灯塔), out of 5
businessTypeEn	Manufacturer, trading company, agency…
certificationsEn	Deep factory / business inspection, Super Factory
serviceCommitmentsEn	Return shipping covered, buy-now-pay-later, counterfeit compensation
inspectionReportUrl	Opens the actual third-party verification report

We deliberately do not compute a trust score or rank the results. A vendor-invented 0–100 number is unauditable. You get the facts 1688 itself publishes, normalised against the category so they mean something, and the benchmarks to read them.

How to use it — the playbook

Every run writes a SOURCING_PLAYBOOK record naming the exact fields for each step:

1. Pick the category first. Compare category* columns across keywords before opening a single product. Stop if categoryKeywordMatchPct is near zero — 1688 didn't recognise your keyword and returned unrelated filler.
2. Shortlist by filter, not by reading. repeatRateVsCategoryPts > 0, supplierYears >= 3, businessTypeEn = Manufacturer, MOQ within your test budget.
3. Hunt the hidden problem. weakestServiceArea, shopNameMatchesLegalEntity, bizTypeMatchesLegalName, supplierKeywordSpread.
4. Verify before contacting. Open inspectionReportUrl and shopUrl.
5. Know the price breaks before you message. priceTiersEn, minOrderQuantity.
6. Run daily to see movement. trend, salesGrowthPct, daysTracked.

Open START_HERE in the key-value store for the same thing as a formatted page: all six steps, plus every field with its official 1688 scale and the market benchmark to read it against — so "67%" becomes "67%, against a platform average of 20–30%". (SOURCING_PLAYBOOK and FIELD_GUIDE hold the same content as JSON if you're consuming this via the API.)

You don't have to open anything, though: the run log ends with what the data actually says about the categories you searched, and which view to switch to next.

English output

Titles, locations, units, tags, certifications and price tiers all come back in English. Company and shop names stay in Chinese on purpose — you need the real name to verify a business or search it. Set includeOriginalChinese to get the Chinese alongside.

Translation works with no API key (a free service is used). For unlimited translation and better wording on marketplace jargon, add any OpenAI-compatible key — free-tier keys work (Google Gemini, Groq, OpenRouter). Translations are cached and reused across runs.

Input

{
  "keywords": ["phone case", "screen protector"],
  "maxResultsPerKeyword": 40,
  "translateToEnglish": true,
  "trackTrend": true
}

Input	Notes
keywords	Chinese or English. Search related keywords together — that's what makes supplierKeywordSpread meaningful.
maxResultsPerKeyword	Broad keywords fill 100+; long-tail keywords stop early on their own
maxConcurrency	Keywords processed in parallel (default 4)
includeAds	Off by default — organic listings only
includeOriginalChinese	Off by default; on = Chinese originals alongside the English
trackTrend	Keeps snapshot history so trends can build up
proxyConfiguration	Residential proxies strongly recommended — 1688 rate-limits hard

Reliability

1688 does not fail loudly, so the usual failure mode of a scraper here is silently returning less than it should:

• Detects silent throttling — a success response carrying no products — and retries instead of mistaking it for the end of the results. Fixing this alone took one test keyword from 49 results to the full 100.
• Recovers from anti-bot blocks by rotating residential IP and re-establishing the session from a shallow page before resuming a deep one.
• Drops the empty placeholder records 1688 mixes into fallback results — they are never billed to you.
• De-duplicates by offerId; 1688's deep pagination returns overlapping pages.
• Survives flaky proxy connections without abandoning the rest of the keyword.

Notes

• Only publicly visible search data is collected — no login, no personal data.
• All supplier metrics are reported by 1688, not computed by us. Relative measures (*VsCategory*, percentiles) are arithmetic on those facts. None of it is a guarantee — it is where due diligence starts, not ends.
• Supply-side only. Demand, resale price and competition in your own market are not on 1688 and must come from your destination marketplace.