How to Scrape Amazon Reviews and Ratings at Scale
Amazon reviews are some of the most valuable feedback data in e-commerce, and since late 2024 they have also become some of the hardest to collect at scale. This guide covers what data actually lives on a review, the challenges that are specific to Amazon, how to capture reviews on a schedule without re-pulling everything each time, and an honest look at building it yourself versus having a clean feed delivered.
What data lives on an Amazon review
An Amazon product page carries two layers of feedback. The aggregate sits near the title: the overall star rating, the total number of ratings, the total number of written reviews, and the distribution from five stars down to one. Below that are the individual reviews, and each one holds a reviewer name and profile, a star rating, a title, the review body, the date and country, a Verified Purchase flag, a helpful-vote count, the variant bought such as size or color, any attached photos or videos, and badges like Vine or Top Contributor. It is worth separating ratings from reviews, because many customers leave a star with no text, so the ratings count always runs higher than the review count. One detail people miss is that the headline star rating is a machine-learned score, not a plain average, so you cannot recompute it from the numbers on the page.
The challenges specific to Amazon reviews
In late 2024 Amazon moved most reviews behind a login wall, and the product-reviews endpoint now redirects logged-out traffic to a sign-in page. Without a session you see only a handful of featured reviews plus the aggregate and the star distribution, which is enough to track a rating over time but not to read the full history. Getting deeper means an authenticated session, which puts the work under Amazon's terms that prohibit automated collection by a logged-in account and carries real account risk. Amazon also caps how far it will page even when signed in, changes the review markup often, and serves reviews per variant and per marketplace, so the same ASIN reads differently by country and by size or color. These are the reasons a review scraper that worked last month quietly starts returning blanks.
Skip the build, get the feed
Send us your Amazon target and we will scrape a real sample and send it back, no signup.
Get a free sampleCapturing reviews on a schedule
Reviews are append-only in practice, so you do not re-pull everything on every run. You collect once as a baseline, then on each cycle fetch newest-first and stop as soon as you reach reviews you already have, keyed on the review ID so nothing is double-counted. Retain the aggregate on every run as its own time series, because the overall rating and the review count moving is often the signal you actually care about, separate from the text. Deliver it as two clean tables, one row per review and one row per snapshot of the aggregate, both keyed on the ASIN, so you can run sentiment or theme analysis without cleaning the export first.
Build it yourself or have it delivered
For one product and a one-time pull, a careful script and some patience can get you a sample. The trouble starts at scale and over time, where the login wall, the account risk, the paging caps, and the markup changes turn a weekend script into a standing maintenance job. Past a handful of products, or once you need reviews refreshed on a schedule you can trust, a done-for-you feed is usually cheaper than the upkeep. That is our review monitoring: you name the products, yours or a competitor's, and we deliver clean structured review and rating data on your cadence and keep it running as Amazon changes.
Validating the Numbers and Catching Silent Breakage
The most common surprise with Amazon reviews is that the headline star rating is not the plain average of the reviews you can see. Amazon computes it with a machine-learned model that weights recency, verified purchases, and helpfulness, so your own average of scraped reviews will rarely match the number printed at the top. Treat the two as separate fields and do not try to reverse-engineer one from the other. Two more traps sit right next to it. The total ratings count includes star-only ratings with no text, so it will always exceed the number of written reviews you can pull. And Amazon caps how deep the visible review pages go, so you are always working from a sample, never the full history. Say so in your reporting. The dangerous failures are silent. An A/B layout test, a locale swap, or a review block that quietly moves behind a sign-in can return a healthy 200 response with an empty or half-filled section. Nothing errors. The feed just goes flat. Guard against this with canary products whose values you know, row-count floors, star-distribution checks that should sum to the rating total, and a verified-purchase ratio that should stay stable over time. When we run a feed for a client, those checks fire before the data ever reaches them, so a broken selector becomes our problem to fix rather than a bad decision made on stale numbers.
Staying on the Public Side of the Line
Amazon reviews are a public data source, and that is exactly where a clean collection effort should stay. The practical rule is simple. If a review or a rating is visible to any visitor without signing in, it is fair to collect. The moment content sits behind a login, a sign-in prompt, or any access control, you stop. Creating accounts or defeating a gate to reach more data changes the character of the work entirely, and it is a line we do not cross. A few habits keep a program defensible. Collect facts and aggregate signal, star distributions, sentiment trends, and volume over time, rather than compiling person-level dossiers on individual reviewers whose display names happen to be public. Review text is user-authored content, which is fine to analyze internally but carries real rights questions if you republish it verbatim, so keep your analytics use separate from any public display. Pull at a considerate rate rather than hammering the site, both because it is the right posture and because it keeps your collection stable. None of this is legal advice, and Amazon's terms discourage scraping as a contractual matter, so your own counsel should sign off on use and retention. What we bring is a method that is public-only by design, documented, and easy to explain in a room. We do not sell proxies and we do not bypass logins, so the feed you receive rests on data you can stand behind.
Fields worth capturing from Amazon
- ASIN
- Overall star rating
- Total ratings count
- Total reviews count
- Star distribution (percent per star)
- Review ID
- Reviewer name
- Reviewer profile ID
- Per-review star rating
- Review title
- Review body text
- Review date
- Verified Purchase flag
- Helpful votes count
- Variant bought (size, color, style)
- Attached media (photo or video)
- Reviewer badges (Vine, Top Contributor)
- Marketplace or country
Frequently asked questions
Related service: Review & Rating Monitoring
Want a Amazon feed without the build?
Send us the products or pages you need, and we will deliver a clean feed and maintain it. Start with a free sample.
Get a Free Sample