How we decide whether a Shopee price is actually a deal

Every verdict on this site rests on the same process. It is written out here in full, partly so you can disagree with it, and partly because most of what follows was learned the hard way and is not obvious from the outside.

1. We keep our own price record

Once a day, the price and stock level of roughly nine thousand Shopee Singapore listings are captured from Shopee's own affiliate product feed and written to a dated file in a public repository. There are currently 32 such files, running from 2026-07-30 to 2026-09-01. Nothing here is scraped: the feed is data Shopee publishes to affiliates, and no product page is ever automatically fetched.

This matters because it makes a "was" price checkable. A sale sticker's crossed-out number is chosen by the seller. Ours is a file with a date on it that we cannot revise after the fact.

2. What that record revealed: Shopee's feed publishes two price lists

The single most useful thing we have found is that the feed does not carry one price per product. It alternates between two, and it switches the entire catalogue at once.

Measured across 24 consecutive daily snapshots: on ordinary days between 1.1% and 9.5% of prices changed. But on five of those days, 33.8%, 39.7%, 43.7%, 45.3% and 47.3% of the entire catalogue changed at once — and then changed back at the next switch. There is nothing in between; the distribution has a wide empty middle.

The practical consequence for a shopper: a one-day price drop is frequently not a drop at all, just the other list. It is also why one regime can run for six days in a row, which is long enough for a price to look convincingly "stable" while being nothing of the sort.

So a price event only counts here if it survived a switch — the low price was still there after the whole catalogue flipped. That single rule rejects more candidates than any other.

3. Is the aisle a market, or one company?

"Cheaper than comparable listings" is meaningless if the comparable listings belong to the same company. This is much more common than it looks, because one company sells under many brand names.

Measured examples from our own data: every in-stock powerbank in the feed was Anker, which also sells as Zolo, Nano and MagGo. Every whey protein was Optimum Nutrition. Of 26 oral-care listings, 25 were Listerine. The router aisle looked like three companies until we checked the storefronts — Deco, Tapo and Archer are all TP-Link, and all 228 of those rows ship from one shop.

A peer set therefore has to span at least three independent corporate groups and at least three independent storefronts. The second test exists because the first is not enough: one storefront can carry several brands, and if one seller sets half the prices in a comparison, that comparison is a catalogue, not a market.

4. Comparing like with like

A false comparison is not a small error — it is the misleading-price problem this site exists to call out, committed by us. Things that have been caught doing exactly that, each now blocked by a test:

5. A person opens the listing

Four things decide a verdict and none are in the data feed: the real prices of each variant, the number of ratings behind a score, the lifetime units sold, and whether the photograph matches what ships. Someone opens each candidate and reads them.

This is where most survivors die. In one recent batch, three listings were out of stock despite the feed showing stock; one "robot vacuum" was a 500ml bottle of cleaning fluid for a robot vacuum; and one appliance turned out to be three different models sharing a price. Of 24 candidates that cleared every automated check, 4 were published.

6. The verdict, and what it is not

A verdict is one of three words — worth it, worth a look, skip — and it always appears beside the evidence it rests on, including the uncomfortable parts. If four of eleven comparable listings are cheaper than our pick, the page says so in the same breath as the claim.

Commission is not an input. The code that forms a verdict is never given the commission field, and automated tests assert that the identical listing earns the identical verdict at a 1% and a 99% commission rate. This is checked on every build rather than promised here.

And a verdict is about price only. Nothing here has been tested, used, or reviewed. A star rating shown on a page is Shopee's, labelled as Shopee's.

7. Pages age in public

A static page cannot know today's price, so every price here carries the date it was observed, is re-checked daily against the newest record, and visibly warns you when it has gone unverified. If the price rises above what we recorded, the page withdraws its own claim rather than waiting to be corrected.

Found something wrong? Every deal page has a "Tell us" link, or write to [email protected]. Corrections run through this same process, so the price history stays intact.