Ryohin Checker structurally screened 122 Amazon products across 21 categories — using only public structural data (star-rating distribution, Verified-Purchase share, review-posting bursts), not review text. After Fakespot and ReviewMeta shut down in 2025, this is an independent look at how often fake-review signals show up.
Held back (fake-review signals, low rating, or thin data)9%
Of the 11 held back, the reasons broke down as:
Fake-review signals (below the trust threshold)100%
Genuine reviews but a low average rating18.2%
Not enough data to judge0%
Share of the 11 held-back products (overlap possible).
⭐ A high star rating doesn't reveal fakes
The products we held back averaged ★4.55 — barely different from the ★4.45 of the products that passed. In other words, a wall of ★4.2–4.3 ratings doesn't tell you the reviews are genuine. That's why we read the structure of the reviews, not the star number. Across the 122 products screened, 13.9% tripped at least one structural fake-review signal.
Trust-score distribution
90–100 (highly trustworthy)84.4%
75–89 (mostly trustworthy)8.2%
50–74 (caution)3.3%
under 50 (strong fake-review signals)4.1%
How often each fake-review signal appeared
Share of the 122 screened products where each structural signal showed up (a product can trip several).
Almost no low ratings — suspiciously clean5.7%
Shown average doesn't match the distribution5.7%
Bimodal (5★ and 1★ both heavy, thin middle)4.1%
Star ratings over-concentrated at 5★3.3%
High rating but few ratings (unstable)3.3%
Low share of Verified-Purchase reviews2.5%
Very few ratings2.5%
Categories where fakes hid most often (lowest pass rate)
Lower pass rate = more products that looked good on the surface but didn't clear our screen. Tap through to each category's fake-free picks.
Same structural screen, different markets: the US set passed more often (91% vs 83.9%) and tripped fake-review signals slightly less (13.9% vs 14.3%). Both are our seed sets, not representative samples of each marketplace. Japanese edition: /report.
Method & limits (transparency)
We fetch each product's public Amazon page and analyze only the structure — star distribution, rating count, Verified-Purchase share, and posting dates — never the review text. A product "passes" only if it clears all three bars: trust score ≥ 75, average ★ ≥ 4.0, and enough data to judge. The denominator is the 122 products whose full review structure we could read (75 fetch failures and 49 partial reads are excluded to avoid bias).
Limits: this is an aggregate of our own screened seed set, not a representative sample of all Amazon. Structural signals are a probabilistic estimate, not proof — mistakes are possible. Figures are a snapshot as of 2026-07-01.
Run the same check on a product you're eyeing
Paste any amazon.com URL and we'll run this structural check instantly, and surface only the products that pass.
As an Amazon Associate, Ryohin Checker earns from qualifying purchases. Verdicts are estimates inferred from public page data (star distribution, number of ratings, posting dates, verified-purchase share) and do not guarantee authenticity (mistakes are possible). We do not store or republish review text. Rankings and recommendations are not influenced by commissions.