As of November 8, 2026, a listing audit published on November 6 found one knife set on Flipkart priced at Rs. 288 and listed eight separate times, every copy carrying the same 1,406 ratings. Four three-piece sets sitting within Rs. 15 of each other wore discount badges of 83, 82, 52 and 34 percent, on stated list prices ranging from Rs. 299 to Rs. 3,499. One seller stated both Rs. 1,999 and Rs. 1,099 as the list price for its own set, and the cheaper tile cost Rs. 24 more. For marketplace and brand teams, the finding is not about knives. It is that ratings attach to listings rather than to products, so any review analysis has to begin with an identity check.
Eight listings, one rating pool
The duplicated listing matters because it breaks the assumption behind every review score. A rating is an average over a pool, and the pool is supposed to represent the population of buyers of that product. When one product occupies eight listings, the same 1,406 ratings are counted eight times in any category-level average, so the category looks better reviewed than it is. Buyers comparing tiles see eight options with strong social proof, which is the opposite of what the data actually says.
A parallel case shows the same mechanism from the other direction. A snapshot exercise across 62 ASINs taken five days apart found ratings unchanged on 55 of them while review volume rose on 48. Two ASINs that started at an identical 6,909 reviews ended at 2,250 and 58. Read naively, that looks like a collapse in satisfaction. Read with an identity check, both look like variant merges: the reviews moved to the parent listing rather than disappearing.
What the audit should measure
Three measures separate a real change in sentiment from a change in listing structure. Any review analysis that skips them ends up measuring listing bookkeeping instead of shopper opinion. The first is the number of reviews behind the score, not the score itself. The second is the verified-purchase share, because unverified text is where most manipulation lives. The third is the age of the oldest reviews compared with the date the product first became available, which is the cheapest hijack test there is.
| Signal | What it can mean | Check before concluding | Action |
|---|---|---|---|
| Rating falls while review count climbs | A new-review flood pulling the average down | Read the theme summaries, then raw reviews | Fix the recurring defect |
| Review count drops sharply | Variant merge or split, rarely deletion | Match title and brand across snapshots | Rebuild the baseline, do not alarm |
| Badge far above peer | Stated list price, not transacted price | Compare price history, not badge depth | Report on price gap, not discount |
| Huge rating pool, tiny price | Long-running generic listing | Check whether reviews describe this product | Flag review hijacking risk |
| Theme drift with flat rating | Real change under a stable average | Pull raw review text for the theme | Update product or copy |
This table is designed for a monthly cadence. Five-day windows are useful for proving a pipeline works, but they are too short to call a trend: any week with a launch or a promotion reads loud for reasons that have nothing to do with sentiment. Monthly reads give trend, and a separate tripwire catches structural change.
Badges measure list price, not price
A discount badge is computed against the stated list price, which the seller sets. Four sets within Rs. 15 of each other showing 83, 82, 52 and 34 percent off is not four different levels of value; it is four different reference prices. Price monitoring that sorts by badge depth will rank the wrong product first, and category managers will draw the wrong conclusion about where the market is going.
Identity first, interpretation second
The cheapest protection against false conclusions is to store title and brand alongside every snapshot. If two records share a review count but differ in title, the count is meaningless for comparison. If they match, a sudden move is probably a merge and the baseline needs a restart. That single field decides whether a report goes to the product team or the platform team.
Running the audit
Start with the sample, not the average. For the twenty listings that matter most, record the review count, verified share, the oldest review date and the date first available. Then sort by exposure rather than by score: a listing with 17,000 ratings and no critical reviews is a bigger risk than one with 300 ratings and a well-documented flaw.
Second, split the pool by recency. Recent reviews describe what is shipping now; older reviews describe a product that may no longer exist. When the two disagree, the recent ones win, because they reflect the current batch and the current fulfilment route.
Audit steps
Five fields and one monthly rerun are enough to catch the failure modes that matter: inflated pools, inherited ratings, and shifts hidden under a stable average. None of it requires new tooling. It requires the discipline to check identity before reading sentiment, which is exactly the step most dashboards skip.
FAQ
Why does a duplicate listing matter if the reviews are genuine?
Because the same pool is counted more than once in category averages, which makes the whole shelf look better reviewed than it is.
Is a sharp drop in review count proof of deletion?
Rarely. It usually indicates a variant merge or split, where reviews move to a parent listing. Check title and brand before concluding anything.
Can discount badges be compared across listings?
Only with caution. Badges are calculated against stated list prices set by sellers, so badge depth is not a comparable measure of value.
How many reviews are enough to trust a score?
There is no universal number. What matters more is the verified share and whether the critical reviews describe the same failure twice.
