Retail and consumer dataset
Sephora US Customer Reviews
4.6 million customer reviews of 11,280 beauty products, from 2008 onwards, with rating, full text, reviewer-stated attributes and helpfulness votes.
- Warehouse-ready Parquet, CSV, SQLite
- One row per review
Ask AI about this
- 4,634,376
- reviews
- 11,280
- products reviewed
- 18 years
- of review dates, from 2008
- 77
- fields per review
Overview
This dataset is a corpus of 4,634,376 customer reviews of beauty products sold in the United States, written from August 2008 onwards. Each review carries its rating, full text, the product and variant it refers to, and the attributes the reviewer chose to state, such as skin type, skin tone and age range.
Those attributes are what separate it from a plain review dump. A rating can be read by skin type or by concern, so the question changes from how a product is rated to who it works for.
Each review also carries provenance flags: verified purchaser, incentivised review and staff. Analysts can include or exclude each group and measure how much the flags move the result.
Coverage
| Reviews | 4,634,376 reviews, of which 4,306,006 are on products currently listed |
|---|---|
| Products | 11,280 products reviewed |
| Period | From August 2008 onwards |
| Reviewer attributes | Skin type, skin tone, eye colour, hair type and colour, age range and concerns, where the reviewer stated them |
| Provenance flags | Verified purchaser, incentivised review and staff flags on each review |
| Engagement | Helpful and unhelpful votes, photos, videos and brand responses |
Data dictionary
| Field | Description |
|---|---|
review_id, reviewed_product_id, reviewed_variant_name
|
The review, the product and the shade or size reviewed |
rating, is_recommended
|
Star rating and whether the reviewer recommends the product |
title, review_text, pros, cons
|
The written review |
submission_time, last_modification_time
|
When the review was written and last edited |
skin_type, skin_tone, eye_color, hair_type, age_range, skin_concerns
|
Reviewer attributes, where stated |
is_verified_purchaser, is_incentivized, is_staff
|
Review provenance flags |
helpful_votes, not_helpful_votes, total_votes
|
Helpfulness votes |
photo_count, video_count
|
Media attached to the review |
client_response_text, client_response_date
|
The brand reply, where there is one |
user_nickname, user_location
|
Reviewer display name and location as shown publicly |
The full field list is sent with every sample.
Who uses it
-
Beauty and consumer brands
What customers praise and criticise by product, shade and skin type, across competitors and over time.
-
Consumer insight and market research
Demand signals by concern, ingredient preference and age range, from reviewer-stated attributes.
-
Machine learning teams
A large labelled corpus for sentiment, aspect extraction and recommendation models, with provenance flags for filtering.
-
Investment research
Review volume and rating trends by brand as an indicator of product momentum.
-
Academic research
Work on review helpfulness, incentivised reviews and consumer language.
Source and definitions
Compiled by Direct Dataset from customer reviews publicly shown on the Sephora US website. Direct Dataset is an independent data provider and is not affiliated with Sephora.
Collection, standardisation and quality checks are described in the methodology.
- Covers the reviews publicly shown on each product page.
- Reviewer attributes are self-reported and are populated wherever the reviewer stated them.
- Delivered with or without the reviewer display fields, as you prefer.
- A fresh edition, with the newest reviews, is produced on request.
Samples and licensing
A sample file in the delivery format, with the full field list, is available on request. Write to hello@directdataset.com and say which areas, categories or dates you want to check.
- Quotation: scoped to the period, geography and fields you need.
- Formats: Parquet, CSV, SQLite. Parquet files load directly into Snowflake, BigQuery, Databricks, DuckDB and pandas.
- Updates: Supplied as an edition. A fresh edition is produced on request.
- Licence: internal use as standard. Research, publication and redistribution terms are on the licensing page.
Questions
Can it be scoped to one brand or category?
Yes. A quotation can cover selected brands, categories, a product list or a range of review dates.
Are reviews deduplicated?
Yes. A review shown on several product pages appears once, linked to the product it was written for.
Can it be used to train models?
The standard licence covers internal use, including model training and evaluation inside your organisation. Redistribution of the review text needs a separate agreement.
How to cite
Direct Dataset. Sephora US Customer Reviews [dataset]. https://directdataset.com/datasets/sephora-us-customer-reviews. Retrieved [date you received the data].
Guidance for publications and newsrooms is on the citation page. This page is also available as Markdown.
This dataset in use
-
Use case
Skincare ingredient data from the Sephora US range
Full ingredient lists turn a product catalogue into formulation data. Across 2,328 Sephora US skincare products, 65.5% contain vitamin E, 46.5% hyaluronic acid and 25.5% niacinamide, and every figure can be cut by category, brand and price.
Related datasets
-
11,682 products
Sephora US Product Catalogue and Price History
The Sephora US range as linked tables: 11,682 products from 355 brands with prices, ratings, ingredient lists and categories, and a dated price and rating record for every capture day since February 2026.
-
4.13M daily prices
Retail Price History, US and Europe
The lowest advertised price on each day on record for 7,063 popular consumer products in six countries, with the shop offering it.
-
864,749 products
Tesco UK Product Catalogue
The full Tesco UK catalogue: 864,749 products with prices, Clubcard offers, nutrition, ingredients, allergens and barcodes, in linked tables.