Ranking better by deleting two thirds of the keywords

Every quarter someone had added keywords. Competitor brand names, adjacent product types, misspellings of things the brand doesn't sell, seasonal terms from a promotion two years ago — all of it accumulating in backend fields, titles and bullets because removing a keyword feels like removing an opportunity.

It isn't. A listing that appears in searches it cannot satisfy collects impressions without clicks and clicks without orders, and both of those tell the ranking algorithm that this listing is a poor answer. The catalog was slowly teaching Amazon to distrust it, and the visible symptom — falling sessions while the ad account looked fine — read to the team as a competitive problem.

We measured each term's contribution and deleted the ones that were costing relevance.

• Per-term audit — impressions, click-through and conversion for every indexed term, per ASIN, over twelve months.
• Deletion criteria agreed in advance, because the argument “but what if someone searches it” has to be settled by data rather than by seniority.
• Titles and bullets rewritten around what remained, which also made them readable for the first time in years.
• Advertising negatives aligned with the pruned set, so paid traffic stopped reinforcing the same irrelevance.

Cutting indexed terms from 1,900 to 620 per listing raised sessions 10% — far fewer searches entered, and a much larger share of them won.

Conversion rose 38% for the same reason: the traffic arriving now matches what the products are. Revenue grew 52% in six months with no new products, no price changes and a slightly smaller advertising budget. The counter-intuitive part is the whole lesson — in a relevance-ranked marketplace, appearing in a search you lose is not free.