NPO is the language layer of the ACO framework described in Andrew Bezos's guide to post-keyword Amazon SEO: instead of stuffing loose keywords, sellers structure titles, bullets, and A+ content around noun phrases built from a head noun plus stacked modifiers — material, use case, audience, constraint, and a proof layer (e.g., lamp → floor lamp → metal floor lamp → dimmable metal floor lamp for reading nook, with UL listed / ETL certified as proof). NPO exists because Rufus (Amazon's AI Shopping Assistant) and Alexa for shopping sit atop the classic Amazon A9 Algorithm search layer and must interpret complete, specific phrases to select a winning product, not just match isolated keywords. Because Cosmo (Amazon's Rufus-Powering Algorithm)'s relational graph linking products to human intentions isn't visible to sellers, NPO functions as a proxy: writing phrases that imply the relationships Cosmo cares about, then watching marketplace signals (via Amazon Search Query Performance (SQP) Report) as feedback rather than optimizing the graph directly.
Ground rules: use only phrases that accurately fit the product; never force irrelevant terms; keep parent phrases (what the product fundamentally is) separate from child/refinement phrases; use noun-phrase stacks only when they read naturally; and source phrase evidence from autocomplete, ads, reviews, Q&A, and catalog data rather than inventing language.
Seven concrete moves: (1) map the query plan in 5 minutes via autocomplete screenshots and Alexa follow-up prompts; (2) mine SQP reports for high-impression/low-conversion phrase gaps; (3) rebuild the title as a coherent noun-phrase stack instead of a crammed keyword string; (4) place a constraint/proof phrase in the first two bullets, since these act as hard filters agents apply before a product reaches comparison; (5) add an audience/occasion line (e.g., 'gift for horse lovers'), since most listings omit who the product is for; (6) mine reviews and Q&A for customers' actual phrasing; (7) run single-variable controlled tests, watching SQP impressions, click share, and purchase share for 2-3 weeks per change.
See also QPO (Query Planning Optimization) for the mission-mapping practice NPO content is built to satisfy.
Amazon's algorithm can surface a listing for long-tail variants built on a root term it already ranks well for — e.g. a listing selling well for "book nook kit" can also surface for a variant phrasing like "booknook kit" — even when that exact long-tail phrase never appears anywhere in the listing, provided the listing already ranks and converts well for the root term. This lets a seller skip keyword-stuffing every low-volume long-tail variant into visible copy; see Amazon Generic Keywords (Backend Search Terms Field) for where such terms can go instead.