Ranking Signal

A ranking signal is any data point Amazon's search-and-discovery algorithm uses to rank ASINs for a query. The algorithm is undocumented and changes continuously, but enough has been inferred from controlled testing (and confirmed by Amazon publications) that the major signals are well-understood.

Relevance signals

"Does this listing match the query?" These are static or semi-static and can be optimised in the listing copy:

• Title tokens. Highest-weight indexing field. Title presence of a query token gates eligibility for most queries.
• Bullet points. Mid-weight indexing field.
• Backend search terms. Hidden field, ~250 bytes, accepts non-displayed tokens including synonyms, misspellings, and cross-language terms.
• Brand field. Used for brand-restricted queries.
• Category and browse node. Wrong category = invisible for many category-scoped queries.
• A+ content text. Indexed but lower weight than the structured fields.
• Attribute fields. Colour, material, size — used for refinement filtering, which gates query eligibility.

Performance signals

"Do shoppers buy when shown this listing?" These are dynamic and rebuild over time as the listing accumulates trading history:

• CTR for the query (does the listing earn the click when shown?).
• CVR for the query (do clickers buy?).
• Sales velocity in absolute units across the trailing window.
• Repeat purchase rate and Subscribe & Save adoption.
• Return rate — high-velocity sales followed by returns demote rank.
• Review velocity and rating trajectory.
• Add-to-cart rate (the pre-purchase intent signal).

Performance signals are query-scoped: a listing that converts brilliantly on its branded keyword may rank poorly on a generic category keyword where its CVR is mediocre.

The flywheel

The two signal classes interact:

Relevance signals  →  eligible to show for query
Performance signals →  rank within eligibility
Performance signals →  feed forward into more impressions
More impressions   →  more performance signal accumulation

A new listing has only relevance signals — it gets a small "rookie" impression allowance, and what happens during that allowance window determines whether the flywheel spins. This is the launch problem, and the reason PPC during launch is not optional: paid traffic creates the performance signal the listing otherwise can't earn.

Common mistakes

• Optimising relevance and ignoring performance. A perfectly indexed listing with a bad main image never gets out of the rookie window.
• Treating "A9" as fixed law. The algorithm shifts. What worked in 2022 is not what works today. Trust controlled testing over folklore.
• Ignoring return-rate signal. A SKU with great sales velocity and a 22% return rate is being quietly demoted; the velocity papers over the demotion until inventory cycles.