How to get Amazon reviews data via API

A logged-out visitor to an Amazon product page sees a small set of featured reviews Amazon chooses to surface — enough to get a sense of a product, not enough to analyze it. Amazon only pages beyond that featured sample for a signed-in session: click "See all reviews" as a logged-out visitor and you hit the same wall an API call without credentials would hit.

That's the shape of the problem the Amazon reviews API solves: it gives you a programmatic path to the same paginated review set a signed-in shopper can browse, returned as structured JSON instead of a page you'd have to scrape by hand.

The review pager checks two things before it answers: a CSRF token and the session cookies it was issued alongside. Both come from a real, signed-in browser session — there's no way around that requirement at the API level, because Amazon designed the pager to answer a session, not a bare request.

In practice this means you sign into the Amazon marketplace you want reviews from, open any product's review page, and copy the cookie header your browser sent with that request out of your developer tools' Network tab. That header — one long line of semicolon-separated name=value pairs — is what you pass as the cookies parameter on every call.

Start with an ASIN or a product URL — either is enough to identify the listing. Add domain if you're not targeting the default marketplace, and starFilter if you only want positive or critical reviews rather than everything. Attach the cookies parameter, and the call returns one page: ten reviews, plus the total review count and the seller the page header states.

Increment page on each subsequent call to move through the rest. Because Amazon paginates ten reviews at a time, a product with 300 reviews is 30 calls — cheap to do one at a time for a quick check, but not the approach to reach for when you need the whole set.

For a product with many pages of reviews, queue the pull as a batch task instead of looping calls yourself. POST the ASIN, your cookies and a page cap to the tasks endpoint, and the Data Engine walks every page in the background — you can start reading finished pages back before the whole run completes, rather than waiting on the slowest page to hold up the fastest ones.

Billing works the same whether you batch or call live: 10 credits per page, only for pages that actually succeed. A run stops cleanly if it runs out of credits partway through, and the pages it already fetched stay readable rather than being discarded.

The API hands back structured fields, not free-floating HTML, which makes the review text itself easy to route straight into a sentiment classifier, a topic model, or a large language model prompt asking it to summarize what buyers like and dislike. Pairing review text with the star rating gives you a useful sanity check — a model's sentiment score should broadly track the rating it was pulled alongside.

For a fuller picture, join reviews with a call to the product-detail endpoint for the same ASIN: pricing, images and specification data sit alongside the sentiment you've extracted, so a dashboard or report can show both what buyers are saying and what they were looking at when they said it.