Case study · Marketplace listings · Home improvement
Copy fixes misjudgment. It doesn't fix invisibility.
A home-improvement parts retailer had listings that sold fine and came back too often. We rewrote the copy around the one measurement that decides whether a part fits. Returns fell hard on the listings people could see. On the listings nobody could see, nothing happened at all. Both results were the lesson.
The problem
Buyers weren't wrong about the product. They were wrong about the fit.
Replacement parts live or die on one number: a diameter, a glass thickness, a frame size. When that number is buried or missing, the buyer guesses. Guess wrong and the part comes back, the return reason says "not as described" or "doesn't fit," and the listing creeps toward the marketplace's "frequently returned" badge. That badge is permanent once applied and suppresses conversion for the life of the listing.
The fix we tested is what I call a fitment gate: publish the measurement that decides fit, where the buyer actually reads it, before they add to cart.
The result
Every listing that got a fitment gate came down. The one we left alone went up.
The large frame-only kit only fell about three points, which looks modest. The more important change was the reason. Its top return reason moved from "performance or quality not adequate" to "damaged item," at about two thirds of returns. The fit returns stopped. What was left was freight damage: a packaging and carrier problem, not a listing problem. That listing had been the closest in the catalog to the "frequently returned" badge, and it stepped back.
The method
How we knew it was the copy and not the calendar.
The test ran into a busy season, when almost everything in the catalog rose. So a "before and after" on its own proves nothing. The program was built to separate our work from demand.
- Leave comparison listings alone on purpose. In each product family, at least one listing got no changes, so it could tell us what the season did by itself.
- Pair listings that should behave the same. Same family, same season, one changed and one not. Only these pairs carry real weight.
- Read returns, not just sales. Return rate and return reason codes from the marketplace's own return reports, 30 days before against 30 days after.
- Say how sure we are. The trim plug result rests on roughly 30 orders per side; the kits on 25 to 90. A single return moves a point. The direction is consistent across every listing we touched and reverses on the one we didn't. That's why it's worth acting on, and why it gets a second month before anyone calls it settled.
The wrong call, corrected
The first read said the rewrites failed. I said so. I was wrong.
On the morning of the 30-day read, I pulled sales first. Sales on the rewritten listings looked flat against a rising season, and I reported that the copy work did nothing and recommended stopping it.
Then the return data came in, and it told the opposite story. Same listings, same window: returns down across the board, and the untouched comparison listing up. The report was corrected the same day, with the original call and the reason it was wrong left on the page.
Sales tell you whether people found the listing. Returns tell you whether the listing told them the truth. Grade copy on the second one.
FRDTLAB · Reporting rule
The A/B pairs
Rewriting a listing nobody sees does nothing. Three pairs say so.
Some rewritten listings had no sales rank at all. For those, we had untouched siblings to compare against. Units sold, same 30-day windows:
| Pair | Rewritten listing(s) | Untouched sibling |
|---|---|---|
| Hinge, single pack | 1 unit | 4 units |
| Door sweep | 0 units across three listings | 5 → 6 units |
| Wood bottom sweep | 1 → 1 unit | 9 → 24 units |
In all three, the untouched listing did better. A family of 36 rewritten listings with no rank and no parent listing also stayed at zero. Better words can't help a page the marketplace never shows.
The rule it left behind
Know which problem you have before you rewrite anything.
Copy fixes misjudgment. It doesn't fix invisibility.
- Listings with traffic and a fit-driven return problem: publish the measurement that decides fit. It works, and it works hard. The plug listing fell 76% in relative terms.
- Listings with no sales rank: nothing in the words helps. The lever is structural. Fold them into a variation family that already has reviews and rank, or retire them.
- Structured attributes: keep filling them, for accuracy rather than growth. One attribute said "1 item" on every three-pack, including the two best listings in the account. A listing with its color corrected still sold nothing. Attributes alone don't restart a dead listing.
What it was worth: roughly six or seven fewer returns a month on the plug family and about two fewer on the frame-only kits. Modest on its own. The larger value is that the mechanism is now proven and can be pointed at the rest of the catalog, starting with the kits where one point of return rate is worth more than an entire small-parts category.
The next read already shows where to point it: frame-only kits convert visitors to buyers at roughly 4.8% to 6.6%, against 7.5% to 9.3% for the matching glass-and-frame kits. Same buyer confusion, showing up twice. Fewer people buy, and more of the ones who do send it back.
Run this on your catalog
Find out which of your listings have a fit problem and which have a visibility problem.
The AI Visibility Audit
A fixed-price, cross-channel look at what's working, what's breaking, and where the highest-value fixes are hiding. Credited in full if you move to advisory.
See the auditListings, feeds, and returns, run with you
Marketplace and storefront listing programs, comparison-listing tests, and return-reason reporting, scoped to your catalog.
Request a quoteClient details, product names, and revenue figures are removed. Percentages, point changes, and unit counts are from the marketplace's own return and business reports.