Review & Ratings Data Scraping

No, categorically. Review text, rating, date and verified flag are collected; reviewer names, profile links and review histories are not. Every record carries a reviewer_identity field with the constant value not_collected so the exclusion is auditable.

The reasoning is practical as much as ethical: reviewer identity adds almost nothing to product analysis while creating personal data at scale that attracts subject access and deletion obligations the moment it lands in your systems.

We flag anomalies and we will not label reviews as fake. We surface unusual velocity, distribution anomalies, verified-share anomalies, sudden rating shifts and text homogeneity, naming the specific pattern.

Every one of those patterns has legitimate explanations — promotions, press coverage, prompted review campaigns, platform flagging behaviour. Reliable detection of coordinated inauthentic behaviour needs account-level data we deliberately do not collect. A false accusation is serious, so we deliver signals for your team to investigate rather than conclusions.

Because it conceals the two things that matter. Distribution shape distinguishes a product with a specific failure mode from a merely mediocre one at the same average. And a product with thousands of reviews barely moves its average even when recent reviews turn negative, so deterioration is invisible for months.

We deliver full distribution, trailing rating trend and review velocity against prior periods. The average is included; it is just not where the signal is.

About 89% on English reviews, lower on other languages and on very short reviews. Themes come with mention counts, sentiment and trend direction rather than a single sentiment score.

We publish the rate and report per-language performance rather than a single headline figure, because extraction quality varies materially and a team building product priorities on themes should know where the data is weaker.

Because review populations and prompting differ systematically. A retailer emailing purchasers gets a different sample than a marketplace where reviewing is unprompted, and app store review prompts differ again. Gaps of 0.3 to 0.5 stars for the same product are common.

We match items across platforms and compute the gap explicitly, since a single-platform average is not the product's rating. Theme differences by platform are often more informative than the rating gap itself.

As far back as each platform publicly exposes it, which varies considerably. Some platforms show full history; others paginate to a limit or surface only recent and highest-voted reviews.

We state per platform how far back coverage extends rather than implying full history. For products where deep history matters, we report what is retrievable during the pilot so you can judge before committing.

Where the platform separates them, yes. Many platforms pool reviews across a variation family, which means a review of a different colour or size appears on the item you are analysing.

We capture variant references where shown and flag where reviews are pooled, because pooled reviews on a poorly performing variant can drag an otherwise strong product — and that is a listing structure problem rather than a product problem.

We collect publicly posted review content without accounts or credentials. Review text about a product is product feedback; the sensitivity comes from reviewer identity, which we exclude categorically.

Platform terms often restrict automated access and we state that plainly. You receive a written methodology document per platform and a DPA before signature. In our experience the reviewer identity exclusion is what makes this category straightforward for a DPO to approve.

We quote individually. The drivers are platform count, item count, whether theme extraction is required, and refresh frequency. Theme extraction is the labour-intensive component and the main cost differentiator.

Rating metrics and velocity across a defined item set at daily refresh sits at the lighter end. Multi-platform coverage with theme extraction across large catalogues sits higher. One scoping call, a free pilot on your own items within 48 hours, then a fixed monthly quote. Request a quote.