OPEN BY DESIGN
Look behind
the numbers.
A useful review score should come with its workings.
What BeautyBrief does
BeautyBrief brings the community perspective to beauty: skincare, makeup, haircare and beauty devices. The current collection covers skincare, makeup and haircare. Beauty devices are planned for a later expansion.
We synthesize public product reviews into short, attributed English summaries. The current research collection contains 2683 review accounts across 111 products. Research and summarization are AI-assisted. We have not independently tested these products, verified every identity, or validated the weights against a ground-truth dataset.
Each product has its own evidence page. Every summary links back to a source. Retailer totals are shown separately and are never added to the number of reviews examined. The sample is manually assembled from accessible pages, not automatically updated or statistically representative.
1. Specific experience comes first
A review receives a documentation-quality factor between 0 and 1. Concrete experience, duration, skin context and limitations help us assess its detail. Praise, follower counts and an “AI-sounding” style are not proof of reliability. Some sources display machine translations.
Brand marketing and a store’s automated review summary are not consumer reviews. Shipping-only comments, untried purchases and unrelated product copy are not treated as evidence of product performance. Reviews about an uncertain or different formula are excluded from the score and labeled when retained for context.
2. Commercial relationships are visible
Self-reported purchase, affiliate links, gifted products, payment, review rewards and unknown relationships are separated. A verified-purchase label does not rule out an incentive. An iHerb review is marked incentivized when its public review data explicitly indicates a reward, not simply because the platform has a reward program.
No disclosure is not proof of independence. Where multiple relationships are known, the stronger one sets the commercial factor. We do not claim that any category is automatically truthful or false.
3. A transparent calculation
Product score = Σ (review rating × weight) ÷ Σ weights
Original 1–5 stars are used when available. Other scales are converted and labeled. A text-only review can receive an estimated rating: 5 for strong praise or stated repurchase, 4 positive, 3 mixed, 2 negative and 1 strongly negative. The label always distinguishes our estimate from the author’s own score.
Products appear in the collection only after at least ten eligible reviewer records support a score. Products below this floor are held back for further research. Review filters can reduce the available sample and hide products from the filtered catalog. This is an editorial floor, not a scientific confidence threshold. Identifiable repeat authors count once per product, with the later eligible record taking precedence. Explicitly conflicting stars and text can be kept as visible context without entering the score.
A large number of reviews on one site can dominate the result. Product pages show the share of weight supplied by each source. Scores describe satisfaction in this sample; they do not establish safety, clinical efficacy or the likelihood of an adverse reaction.
How much evidence is behind a score?
The coverage label describes the size and spread of eligible reviews in our sample. It does not rate the product, prove that reviewers are independent, or express statistical confidence.
- Broader review coverage: at least 40 eligible reviews across at least 3 websites, with no website supplying more than 70% of the reviews.
- Developing review coverage: at least 20 eligible reviews across at least 2 websites, with no website supplying more than 80%.
- Limited review coverage: the sample does not meet either threshold. A small sample or a single dominant source needs more context, even when its score is high.
These are editorial thresholds. Source share here counts eligible reviews, while the source-influence chart uses their weights. Different websites can share an owner or syndicate reviews. When syndication is explicitly identified, we group the account with its original publisher. Filters and custom weights recalculate coverage from the reviews that still contribute. Coverage is not a measure of freshness or clinical evidence.
Shopping links & disclosures
Shopping options include affiliate and non-affiliate links. We may earn a commission when you buy through links marked “Paid link”, at no extra cost to you. As an Amazon Associate I earn from qualifying purchases.
Retailers are linked when the product and version match our review. A purchase link is not a guarantee of current stock, price, seller authenticity or delivery to your country. Where a listing offers several shades or sizes, check the selected option before buying.
Affiliate relationships do not affect product selection, review weights or scores. Research-source links are separate. Some source authors have their own commercial ties, labeled on their review summaries. Product images from brand or retailer pages identify the products; we have not independently tested them.
Found an error? Report it to the editorial inbox. Include the product, exact version and a source link if available. Reports are reviewed before the brief or its score is changed. We do not automatically treat a submitted claim as verified evidence.
Publisher-linked sources
Reviews by Maya is operated by the same owner as BeautyBrief. When one of its product reviews is included, it is labeled as a publisher-linked source. It receives no ownership-based bonus: detail, product-version fit and disclosed commercial relationships are assessed with the same rules as other sources. Cross-posts by the same author count once per product.
Background reading
FTC: featuring online customer reviews
COLING 2025: limits of AI-text detection
These sources support transparency principles, not the numerical weights chosen for BeautyBrief.