Most people read two things off a review section: the star rating and the count.
The trouble is that both numbers blend four sources that mean different things. The review endpoint separates them at the field level — every review carries four booleans, and the distance between them is large.
What the four booleans are
| Field | Meaning | What the review tells you |
|---|---|---|
verified | Verified purchase | This person actually paid. The "VP" review |
vine | Invited review | A platform invitation program; the reviewer received the product free |
free | Free review | The reviewer did not pay to obtain the product |
experience | Early experience | A reviewer who received the product early in its life |
These are four independent booleans, not one four-way enum. A review can be both vine and free, because an invitation program supplies the product at no cost by definition. Read them as combinations rather than assuming they are mutually exclusive.
Why blending them produces a meaningless number
Say a product has 200 reviews averaging 4.3. The question that number is meant to answer is "are real buyers satisfied with this."
But if 60 of them are vine or free, those 60 reviewers never paid for it. Their experience is missing the part that matters most: whether it was worth the price. For a $29.90 item, getting it free and buying it are not the same judgment.
Averaging them together produces a 4.3 that is neither the verdict of paying buyers nor the verdict of free recipients — it describes a population that does not exist.
So a baseline rating should be computed from one kind of review only:
verifiedtrue, withvine,freeandexperienceall false.
That is the "paid for it, bought it normally" group. Compare that number against the rating shown on the page, and the gap is itself a signal — the wider it is, the more the page rating is being lifted by non-paying reviewers.
Variations: you think you are reading one color, you are reading the whole family
Every review carries a skus array naming the variation it belongs to.
A parent listing pools reviews across all its variations. So a page for the black model may be showing reviews of the white one, the large one, the higher-wattage one. When quality differs between variations, the pooled rating hides it.
A concrete version of this: one size has a workmanship problem, the complaints concentrate on that SKU, and the parent rating is diluted by the other sizes until it merely looks "a bit low." Group by skus and the problem lands on a specific variation immediately.
This is the same family of trap as parent-versus-child in sales data; What Amazon sales data can and cannot tell you covers the other side of it.
Which reviews actually move buyers
Reviews do not get equal exposure.
| Field | Meaning | Why it matters |
|---|---|---|
likes | Helpful votes | High-vote reviews sort to the top and get read far more |
image | Has an image | Image reviews get expanded and trusted more readily |
video | Has a video | The same, with more weight |
authorLabels | Reviewer labels | Markers on the reviewer |
When you analyse negative reviews, a 47-vote negative review with a photo and a zero-vote text-only one are not the same order of impact on conversion. Sorting by likes and reading the top twenty gets you closer to what buyers actually see than reading all two hundred.
Pull the low-star reviews directly
The endpoint's stars parameter takes an array, so you can fetch specific ratings only. Problem analysis does not require downloading everything:
- 1–2 stars only → the defect list
- 3 stars only → usually the most informative band, because it spells out what works and what does not
- 5 stars only → confirms whether your selling points actually registered with buyers
The three-star band deserves its own note. Five-star reviews are often one line of praise and one-star reviews often carry heat; three-star reviews tend to be the most specific, because the writer wants neither to condemn nor to recommend.
Review endpointPaginated reviews by ASIN with star filtering, returning source flags, variation SKU, helpful votes and image and video markersThree things this data cannot do
It cannot judge whether a review is genuine. The endpoint returns the sourcing the marketplace publishes, not an authenticity verdict. verified means the platform recorded that purchase, and no more.
It cannot give you reviewer identities. Not available, and the wrong direction anyway.
It cannot categorize the complaints for you. The fields tell you which reviews count; whether a batch of negatives is complaining about the same thing is a human judgment — or a model's, with a human confirming.
This post does not cover manipulating reviews in any form. Everything above is read-only analysis of public data.
Next
To get these fields into a sheet you can fill in row by row, see Exporting Amazon reviews into a spreadsheet, which ships a downloadable review analysis sheet. For turning a filled sheet into conclusions you can act on, see How to analyse Amazon reviews.
Questions
What does VP mean?
VP is Verified Purchase — the marketplace confirmed the purchase actually happened. It maps to the verified field.
Should Vine reviews count toward the average? Not for a baseline rating. Those reviewers received the product free and are missing the price judgment. They are worth reading separately though — early Vine reviews are often the most detailed on product specifics.
Are free reviews and Vine the same thing?
Not quite. Vine is the platform's invitation program, while free only states the reviewer did not pay to obtain the item. A review can be both, which is why you read the combination rather than picking one.
Why does the review count on the page differ from the endpoint total?
First check you are comparing the same ASIN level. A parent page pools every variation's reviews, and you may have queried a single child. Reconcile using skus.