DEV Community

Matvey
Matvey

Posted on

One in Five AliExpress Rows Quoted Exactly $1.09. All Five I Opened Capped It at One Piece

I run a visa agency in Bali. The part of my week that has nothing to do with visas goes into scrapers on Apify, and one of them reads AliExpress — search cards, and on request the product page with every variant, its stock and its per-shopper limit.

The Actor health-checks itself every night with the same query, so instead of a one-off measurement I have a panel: 86 runs of wireless earbuds shipped to US in USD over 28 distinct days, 3,751 product rows, 450 distinct products, plus 232 rows read all the way down to the product page — 20,936 variants between them.

The first thing the panel told me is that the price on an AliExpress card is not a price. It is an offer, and the offer depends on who is asking.

722 rows, one number

Of 3,630 rows that carried a sale price, 722 quoted exactly $1.09 — 19.9% of everything, spread across 184 different products. Next in line: $0.33 on 112 rows, $1.33 on 110, $2.33 on 99, $3.33 on 88. Hundreds of unrelated sellers, the same handful of prices.

A search I ran while writing this was worse than the average: 17 of 40 rows came back at $1.09.

The explanation is printed on the row itself, in a field that looks like decoration. 1,913 of the 3,751 rows (51%) carry a selling-point tag that begins New shoppers save $… — New shoppers save $14.77, New shoppers save $30.99. Of the 834 rows priced at $0.33 or $1.09, 642 have that tag. It is the welcome deal: AliExpress quotes a first-order price to a visitor it does not recognise, and a scraper is always a visitor it does not recognise.

So I took five products that were quoting $1.09 in that search and opened their product pages in the same Actor. All five came back with detailPrice: 1.09 — and all five capped it:

3256812924699567  variants 5   cheapest maxPerOrder 1   "Max. 1 pcs/shopper at this price"
3256807671771690  variants 21  cheapest maxPerOrder 1   "Max. 1 pcs/shopper at this price"
3256807716091171  variants 1   cheapest maxPerOrder 1   "Max. 1 pcs/shopper at this price"
3256812488850828  variants 4   cheapest maxPerOrder 1   "Max. 1 pcs/shopper at this price"
3256812446184514  variants 3   cheapest maxPerOrder 1   "Max. 1 pcs/shopper at this price"
Enter fullscreen mode Exit fullscreen mode

Across all 20,936 variants I have, 6,692 are capped at one piece per shopper. If you are sourcing, that price is a sample price. One unit, once.

Two reads an hour apart do not agree

I had assumed the welcome price was at least stable for the day. It is not even stable for the hour.

Taking every pair of panel runs started between 1 and 60 minutes apart, I get 8,914 cases where the same product appears in both. 2,203 of them (25%) disagree on the sale price, and 899 disagree by more than 1.5x. On the cheap side of those 899 pairs, the single most common value is $1.09 (420 times).

One product, 7.5 minutes apart: $14.82 and $2.82. Another, 14 minutes apart: $1.09 and $6.55.

The day-level series looks the same. One listing I happen to have on 13 separate days:

2026-10-13         3.90   originalPrice 25.12   discount 84
2026-10-14         1.09                 25.12            95
2026-10-16         3.83                 25.12            84
...
2026-11-09         1.09                 23.43            95
2026-11-12         1.09                 23.43            95
2026-11-13         3.06                 23.46            86
Enter fullscreen mode Exit fullscreen mode

originalPrice moved 7% in a month. salePrice flipped between $1.09 and $3–4 depending on the read. Across the 41 products I have on ten or more days, the median high-to-low ratio of the card price is 5.26x, and 32 of 41 moved by more than 2x. That is not a repricing signal, it is a promotion lottery.

The product page disagrees with the card too: of 222 rows where I have both, 74 (33%) differ, and the pattern is almost always the card quoting less — card $0.33, page $1.09.

What to track instead, if you want a price series that means anything: originalPrice, or the cheapest variant whose maxPerOrder is not 1. Both come from the same row.

The one number that held up

orders is the opposite story. AliExpress prints a bucket on the card — 1,000+ sold — but the exact count is in the data the page ships to its own analytics, and that is what this Actor reads. Of 3,691 rows with an order count, only 7 were divisible by 100. The values are real: 1,476; 40,479; 76,759.

The bucket text, for comparison, is useless at the top end. Rows labelled 10,000+ sold carried counts from 10,751 to 48,585. And the labels come in two vocabularies — 1,000+ sold on 416 rows, 1K+ sold on 64 — so 382 distinct strings describe maybe 30 real rungs. Parse the number, not the sentence.

Because the count is exact, a daily run turns into a sales rate. Over the 101 products I have for at least a week:

orders per day
median 3.8
lower quartile 0.5
upper quartile 13.9
the one at the top 422 (13,402 → 26,906 in 32 days)

That last one is the whole point of scraping a marketplace: a single AI-translation earbud listing selling four hundred units a day while the median listing on page one of the same query sells four.

Two caveats found in the same pass. First, 2 of 45 tracked counters went down — one by 177 units over a month, one by 337. A lifetime total cannot fall, so this counter is a window, not a career total; treat differences as "sales in the last month-ish", not as cumulative truth. Second, the shelf moves: of the 60 products in my first day's top 60, 15 were still there on day 28.

Two of my own bugs, both from this panel

A card with no price at all. 121 of the 3,751 rows (3.2%) had title, image, rating, order count and nothing else: no salePrice, no originalPrice, no skuId, no tags. They were never in the top ten and arrived one or two at a time, which is exactly why my existing repair — re-read the page from another address when more than half the cards look thin — never fired. I tried widening that repair first and measured it doing nothing: 12 price-less cards over four runs, 0 filled, because the same product comes back price-less from every address.

The price was on the card all along. The card's own analytics map carries salePriceAmount: "109" and originPriceAmount: "1075" — minor units of whatever currency you asked for. I verified the scaling on 118 cards that had both a price block and the map: 118 matched to the cent (originals to within two cents). Now the row falls back to it. Before: four runs of 60 rows → 4, 1, 3, 4 empty prices. After: four runs of 60 rows → 0, 0, 0, 0, with 13 cards filled from the map.

A price range set by things you cannot buy. AliExpress lists SKU combinations whose option values are not in the page's own option lists; all 475 of them in my sample are salable: false and 471 have zero stock, and they keep a placeholder price. On one neon-rope listing that placeholder is $100,666.99, and it was deciding priceMax:

latest  priceMin 16.98  priceMax 100666.99
new     priceMin 16.98  priceMax 48.16      (highest price a shopper can actually pay)
Enter fullscreen mode Exit fullscreen mode

Both fixes are live in build 0.2.21 of AliExpress Products Scraper.

Calling it

{
  "keywords": ["wireless earbuds"],
  "maxItemsPerQuery": 40,
  "shipTo": "US",
  "currency": "USD"
}
Enter fullscreen mode Exit fullscreen mode

A real row from that run, trimmed:

{
  "productId": "3256812924699567",
  "title": "TWS True Wireless Stereo Headset Touch Contro With Charging Case...",
  "salePrice": 1.09, "originalPrice": 15.86, "discountPercent": 93,
  "formattedPrice": "US $1.09", "currency": "USD",
  "rating": 4.9, "orders": 1510, "ordersText": "1,000+ sold",
  "isAd": false, "isChoice": true, "shipFrom": "CN", "shipTo": "US",
  "tags": ["New shoppers save $14.77", "Lowest price in 90 days"],
  "listedAt": "2026-09-03 00:00:00", "position": 11,
  "scrapedAt": "2026-11-14T21:24:18.928Z"
}
Enter fullscreen mode Exit fullscreen mode

From Python, run and read in one call, and throw away the prices that are not yours:

import requests
rows = requests.post(
    "https://api.apify.com/v2/acts/lergassy~aliexpress-products-scraper/run-sync-get-dataset-items",
    params={"token": "apify_api_..."},
    json={"keywords": ["wireless earbuds"], "maxItemsPerQuery": 40,
          "shipTo": "US", "currency": "USD"},
    timeout=600,
).json()
real = [r for r in rows if r.get("type") == "product"
        and not any(str(t).startswith("New shoppers save") for t in r.get("tags") or [])]
print(len(rows), "rows,", len(real), "without a welcome-deal price")
Enter fullscreen mode Exit fullscreen mode

A 60-row search run costs me $0.054 and finishes in 10 to 50 seconds, no proxy and no browser — the search page ships its results as JSON, so the work is parsing, not clicking. Variants, stock, seller and shipping come from the product page and are a separate, paid pass.

Two more traps before you build on it. Stock numbers are theatre: the median variant stock in my sample is 999, the maximum is 999,993, and totalInventory is above 100,000 on 78 of 232 product rows. Use stock: 0 as a signal and ignore the rest. And option labels are not a taxonomy: for 2,777 of 5,451 option values (on 151 of 188 products) the seller's display name differs from the catalogue value behind it — a group called Emitting Color whose values are 5m, APP, APP 24Key, a Wattage group whose values are 12V 0.5M. That is why the row carries both name and rawName. You cannot group "black" across sellers; you can only group within one listing.

When this is the wrong tool. If you need the price you will actually pay for 50 units, the card will lie to you by 2–5x and so will a competitor's card scraper — read variants and bulk pricing. If you need lifetime sales, nobody has them; what exists is a trailing counter that can go down. If you need buyer questions, or anything behind a login, this Actor never signs in, so that data does not exist for it. And if you are writing your own: the price is in the card's tracking payload as well as its price block, which is the only reason one in thirty of my rows is complete.

Top comments (0)