Amazon Review Defect Finder
Pricing
from $15.00 / 1,000 defect cluster founds
Paste an Amazon product URL and see which complaints are real patterns. It fetches the critical reviews and returns each recurring fault, how many separate reviewers named it, how many times it was mentioned, the first and last date seen, the worst star rating, and their exact words.
result
Optional
DEFECT_CLUSTER is a finding. The other four are status rows explaining why there is nothing to report, so an empty answer is never silent: NO_INPUT (you passed no reviews), INSUFFICIENT_DATA (too few to judge), NO_DEFECTS_FOUND (no failure language at all), NO_CORROBORATED_DEFECTS (failures named, but none by enough separate people). Present on every row, which is why it is the field you branch on, and REVIEWS_DATASET_UNREADABLE (a dataset ID was given and could not be read, so nothing was analysed — this is not a statement about the product), and REVIEW_FETCH_FAILED (a product URL was given and the upstream review scraper returned nothing usable, so nothing was analysed — this is not a statement about the product).
asin
Optional
The product this cluster belongs to. Reads _all when the reviews carried no ASIN.
product
Optional
The human-readable product label you passed as productName.
defectPhrase
Optional
The dominant phrase across the cluster, in the reviewers own words rather than a category we invented.
distinctReviewers
Optional
How many DIFFERENT people named this failure. The corroboration count, and the number that decides whether a complaint is a defect.
totalMentions
Optional
How many times it was named in total. Can exceed distinctReviewers when one person says it twice, which is exactly why the two are separate fields.
firstSeen
Optional
Earliest review date in the cluster. Read it with lastSeen: eight months is a manufacturing problem, eight days is often one bad batch.
lastSeen
Optional
Latest review date in the cluster.
severityMinRating
Optional
The lowest star rating among the reviews in this cluster.
exampleQuotes
Optional
Verbatim evidence, each with the user id and date behind it, so every claim can be traced back to a review.
[ { "text": "The left earbud stopped charging after about four months", "userId": "R1", "date": "2026-03-02" }]why
Optional
Present on a status row, explaining why nothing was reported.
criticalReviewsAnalysed
Optional
How many critical reviews were read for this product. The denominator: two clusters out of two reviews and two out of four hundred are different findings.
failureSentencesFound
Optional
How many sentences in those reviews actually described a failure. The gap between this and the review count is how much of your input carried no complaint at all.
clustersFound
Optional
How many distinct failures were grouped, before the corroboration threshold was applied.
clustersBelowThreshold
Optional
How many were dropped for having too few separate reviewers. A NEAR MISS, and the number to read before concluding a product is clean - a fault named by two people when you asked for three is not absence.
minCorroboration
Optional
The threshold this run used, echoed back so a saved row says what bar it cleared.
analysedAt
Optional
When the analysis ran. Reviews accumulate, so a saved ledger that cannot say when it was true is not evidence of a current defect.
productUrl
Optional
The product page you supplied, copied onto this row unchanged. Null when you did not supply one. It is never fetched and never inferred from the ASIN: this Actor does not know which marketplace your reviews came from, and a guessed domain would be wrong for every non-.com review set.