Why the same brief produces different keywords on each platform
Amazon's A9 weights purchase intent heavily, so product-led keywords dominate. Etsy's algorithm rewards stylistic and gift-oriented long-tail phrases, so the same product surfaces entirely different cluster centers. eBay lives between the two with a strong bias toward condition, compatibility, and price-bucket modifiers. A single keyword extractor can be calibrated for all three if you tell it the platform first.
The mistake most teams make is to run a single keyword pass and paste the result into every channel. The right move is to run the extractor three times with different platform settings and keep the outputs in three separate clusters. The clusters share 30-40% of their terms, but the long tail is platform-specific enough that mixing them weakens the relevance signal.
Three tiers that always work
Tier one is the core product term — the shortest, most direct name for the product. 'Insulated water bottle', 'linen bedsheet set', 'leather laptop sleeve'. Tier two is feature modifiers — size, material, color, count, intended use. Tier three is use-case scenarios — 'for gym', 'for travel', 'for dorm room', 'for new parents'. Every listing benefits from all three tiers, but the platform weighting shifts which tier carries the title.
Run the extractor once for tier one, once for tier two, and once for tier three. The output clusters are easier to align with ad group structure that way. Each ad group gets one tier, and each tier maps to a different match type in Google Ads or Amazon PPC.
Aligning clusters to title, backend, and ads
Use tier-one terms in the listing title. Use tier-two terms across bullet points and feature copy. Use tier-three terms in backend search terms, alt text, and ad campaign targeting. Consistency across these surfaces reinforces the relevance signal to the platform algorithm.
If a tier-three term converts unusually well, promote it to tier-two in the next iteration. Keyword strategy is not a static document; it is a quarterly revision of the last 90 days of conversion data.