If you sell on Amazon, research a category, or build product-intelligence tooling, you eventually need the same thing: ratings and review signal across many products at once. Not one ASIN. Dozens or hundreds. A competitor's whole catalog. Every product in a subcategory. The question is never "what do people say about this one item." It's "where is the rating strength, where are the complaints clustering, and how is that moving over time."
That's harder than it looks, for two reasons most tutorials skip.
Why naive scraping fails
Amazon treats automated traffic as hostile by default. Send a plain HTTP request or a default headless browser and you'll usually get the "Robot Check" / CAPTCHA page instead of the product. Default headless fingerprints (the giveaway user-agents, the missing browser signals) get flagged fast.
There's a second, quieter failure that costs people more. Amazon serves different pages to different countries. Hit a .com product from an IP in the wrong country and Amazon often returns a stripped page, sometimes with the reviews section missing entirely. Your scraper "works," returns a 200, and hands you a row with no review data. You don't notice until the dataset is already wrong.
What actually works
Four things, together:
- A realistic browser fingerprint. A real Chrome user-agent and consistent browser signals, not the default headless profile. You're loading the page the way a person's browser does.
-
Residential proxies pinned to the marketplace's country. Scraping
amazon.co.uk? Use UK residential IPs..de? German IPs. This is the single biggest fix for the "empty reviews" problem. It's what makes Amazon serve the full page. - Scroll to trigger lazy loading. Reviews and the rating breakdown load as you scroll. Grab the DOM too early and you miss them.
- Treat empty as failure. If a run finishes with zero reviews and no rating summary, that's a bug, not a result. Fail loudly. A silent empty return is how you end up trusting a broken dataset.
The honest part: what's actually public in 2026
Here's the reality most tools won't tell you. Since late 2024, Amazon only shows roughly the top ~8 reviews per product to logged-out visitors. The dedicated "all reviews" page now sits behind a login wall. Any tool claiming to pull a product's complete review history without an account is either logging in (against Amazon's terms, and fragile) or quietly returning less than it implies.
So be clear-eyed about scope. Publicly, per product, you can reliably get:
- The rating intelligence: overall star score, total number of ratings, and the full star breakdown (% 5★ / 4★ / 3★ / 2★ / 1★).
- Amazon's own "Customers say" AI summary.
- The top ~8 public reviews, each with reviewer, star rating, title, body, date, verified-purchase flag, and helpful votes.
That's a lot of signal when you're pulling it at scale. What it isn't is deep single-product review mining.
A quick workflow
- Collect the ASINs or product URLs you care about (a competitor's catalog, a subcategory).
- For each, load the product page with a real browser fingerprint and a country-matched residential proxy.
- Scroll to force the reviews and rating breakdown to load.
- Parse the rating summary, the star breakdown, the "Customers say" summary, and the visible reviews.
- Validate: any product returning nothing is flagged, not silently dropped.
The actor
I built this into an Apify actor: Amazon Reviews & Rating Intelligence Scraper. Give it a list of ASINs or product URLs, and for each it returns the rating summary, the star breakdown, the "Customers say" summary, and the public reviews, across marketplaces (.com, .co.uk, .de, and others). It uses the fingerprint plus country-matched residential proxy approach above, and it fails loudly on empty results.
When to use it: competitor catalog research, category/subcategory analysis, and rating monitoring over time. Basically anywhere the per-product summary plus top reviews across many products is the point.
When not to: if you need one product's entire multi-thousand review history. Publicly that isn't available (~8 reviews/product), and this tool won't pretend otherwise.
Pricing is pay-per-event: platform usage passed through, plus about $1.50 per 1,000 rows.
Takeaway
Amazon rating data at scale is very doable if you match the proxy country, use a real fingerprint, scroll, and fail loudly on empty. Full public review history per product isn't, because Amazon closed that door in late 2024. Knowing exactly where the public line sits is what keeps your dataset honest.
Top comments (0)