Shop smarter on
every store. One AI copilot.

Every product page you analyse produces a single number from 0 to 100. That number is built from three independently assessed components, each scored on its own before being combined. We publish the weighting because a score you can't interrogate isn't worth acting on.

Why star ratings are the weakest input

A 4.6-star average tells you almost nothing on its own. It compresses a thousand different experiences into one number, it is trivially inflated by incentivised reviews, and it is heavily weighted toward the earliest buyers — who are typically the most enthusiastic and who bought a version of the product that may since have changed.

So the star average is treated as one weak signal among many. The heavier signals come from the review text: which specific complaints recur, whether they cluster around a particular date (often a sign the manufacturer changed a component), whether positive reviews describe the product in concrete detail or in generic language, and how the ratio of detailed-to-generic reviews compares to the norm for that category.

How review-trust checking works

Fake and incentivised reviews leave patterns. The review-trust check looks for the ones that survive translation across marketplaces:

• Timing clusters. Organic reviews arrive on a long tail. A burst of dozens of five-star reviews inside a few days, especially shortly after listing, is not how real purchase behaviour looks.
• Language uniformity. Genuine reviewers write about their own use case in their own words. Paid reviews converge on the same phrases, the same length, and the same structure.
• Rating shape. A healthy product shows a J-curve — mostly high ratings with a real tail of low ones. A product with almost no middle and no low ratings has usually had its low ratings suppressed or its high ratings manufactured.
• Listing reuse. On some marketplaces a seller can swap the product behind an established listing while keeping its review history. Reviews describing a completely different item than the one on sale are a strong signal that this has happened.
• Verified-purchase ratio. Weighted differently per platform, because the meaning of "verified" differs a great deal between Amazon and, say, AliExpress.

The output is an authenticity percentage plus the specific reasons behind it. You always see why, not just the number.

What each score band means

Score	Verdict	What it means in practice
80–100	Buy	Fairly priced against its own history, quality reports are consistent, and the review base looks genuine. No meaningful reason to wait.
60–79	Buy with conditions	Sound product, but something is off — usually the price is above its normal band, or a specific recurring complaint applies to some buyers. The report names the condition.
40–59	Wait	Typically a price that has been higher than usual for weeks, or a product whose recent reviews are noticeably worse than its older ones. Waiting is usually the better move.
0–39	Avoid	Serious problems: heavily manipulated reviews, a seller with a poor track record, a fake discount, or recurring reports of the item failing quickly.

Why the analysis differs per store

A single scoring model applied identically to every marketplace produces bad results, because the marketplaces are not comparable. On Amazon, verified-purchase status and seller reputation are meaningful and available, so they carry real weight. On AliExpress the same fields exist but mean much less, while shipping time and item-as-described disputes matter far more — so those move to the front. On Shein, sizing is the single largest source of returns, so the model prioritises whether reviewers report the item running small or large and whether the photos match what arrived. On Alibaba the buyer is usually a business, so per-unit price at minimum order quantity, bulk tiers and supplier verification displace review volume almost entirely.

Where the price history comes from

Price history is assembled from observed listing prices over time for the specific product identifier, up to 12 months where available. Coverage is better on large catalogues that have been tracked longer, and thinner on newly listed items — where there is not enough history to be useful, the report says so rather than drawing a confident-looking line through two data points.

The practical use of this is spotting anchored discounts. A "was $199, now $99" claim is only a discount if the item has actually sold at $199 recently. A large share of the permanent-sale listings you see have not.

Limits worth knowing

• The extension reads the visible text of the product page you are on. If a page hides key detail behind interactions the extension can't see, that detail is not in the analysis.
• Brand-new listings with a handful of reviews cannot be assessed for review authenticity with any confidence. The report will tell you the sample is too small rather than invent a percentage.
• Very niche or specialist categories — industrial parts, medical devices, some regulated goods — have thin comparable data, and the score should be treated as a rough prior at best.
• AI analysis can be wrong. Treat the report as a well-informed second opinion, not a verdict.