15:25
Amazon SEO
Citing Andrew Bezos's new guide, the video argues the keyword era of Amazon SEO is ending: winning ASINs now must be optimized around complete, specific noun phrases ("noun phrase optimization," or NPO) so they can be retrieved, interpreted, and selected by Amazon's Rufus/Alexa AI shopping layer sitting atop the classic A9 search engine, not just rank for isolated keywords.
Amazon search now runs on two layers: A9 builds the retrievable candidate pool from listing text, while Rufus/Alexa for shopping interprets shopper intent, context, and budget to pick the winner.
AI shopping agents fan a single prompt (e.g., 'find something for my kids in our home') into dozens of sub-searches, so sellers must cover the whole 'query plan' a shopping mission generates, not one keyword.
Three acronyms define the new discipline: ACO (agentic commerce optimization), QPO (query planning optimization), and NPO (noun phrase optimization).
A noun phrase gets more specific and less ambiguous as modifiers stack onto a head noun (material, use case, audience, constraint, proof), e.g. lamp -> floor lamp -> metal floor lamp -> dimmable metal floor lamp for reading nook.
Amazon's Cosmo system reads relational, common-sense links between products and human intentions, beyond raw attributes and titles, so sellers optimize NPO as a proxy since Cosmo's internal graph isn't visible to sellers.
Ground rules for NPO: use only accurate phrases that truly fit the product, separate parent phrases from child/refinement phrases, use mega-phrases/noun stacks only when they read naturally, and build phrase evidence from market data (autocomplete, ads, reviews, Q&A, catalog) rather than imagination.
Seven concrete NPO moves are given: map the query plan via autocomplete/Alexa follow-ups; mine Search Query Performance (SQP) reports for high-impression/low-conversion phrases; rebuild the title as a noun phrase stack; put a constraint/proof phrase in the first two bullets; add an audience/occasion line; mine reviews and Q&A for customer phrasing; and run a single controlled test changing one variable at a time.
Amazon rolled out a multi-touch attribution toggle in the sponsored products reporting console, letting sellers see credit shared across every ad that influenced a purchase instead of only the last-click ad.
A fresh Market Maze research stat shows AI shopping assistants are strong on product-card discovery but fragmented on checkout: ChatGPT lacks in-chat checkout while Google, Amazon's Rufus, and Perplexity support it; Rufus/Alexa lead on the full package (checkout, loyalty, price tracking, sponsored placements) while Perplexity is the most bare-bones.
ChatGPT's AI market share fell from 76.4% a year ago to 52.7% now, with Gemini and Claude gaining the difference.
A9 — Amazon's legacy search engine layer that reads the words in a listing and builds the candidate pool of products that can be found at all. Apply: Ensure listing text still contains the literal terms and attributes needed to be included in the initial retrievable candidate pool before worrying about ranking.
Rufus and Alexa for Shopping — The AI layer sitting on top of A9 that interprets what a shopper is actually trying to accomplish, applies context and budget, weighs page evidence, and picks the winning product. Apply: Test your product term directly in Alexa for shopping to see the follow-up prompts it offers, revealing the context and constraints it uses to select winners.
Query plan coverage — The idea, per Andrew Bezos, that a single shopping mission fans out into a whole family of related queries, and sellers should aim for eligibility across that family rather than one keyword. Apply: Instead of tracking rank for a single keyword, map and monitor how many of the queries in a mission's full query plan your product can be found and won by.
ACO (Agentic Commerce Optimization) — Engineering an ASIN so it can be retrieved, interpreted, trusted, and selected across both the A9 and Rufus/Alexa layers. Apply: Treat traditional SEO as a subset of a broader ACO effort that also builds trust signals and machine-legible evidence, since great SEO alone is no longer enough.
QPO (Query Planning Optimization) — The practice of predicting all the search paths a shopping mission might generate and assessing how many your product can be found by and how many you can honestly win. Apply: For each mission relevant to your product, list the likely sub-queries and score your findability and win-rate on each before investing in content changes.
NPO (Noun Phrase Optimization) — The language layer of the framework: organizing product content around phrases that define what the product is, who it's for, where it's used, what constraints it meets, and why it can be trusted. Apply: Rebuild titles, bullets, and A+ content around structured noun phrases built from a head noun plus modifiers rather than loose keyword lists.
Noun phrase anatomy — The recommended phrase structure: head noun, then material, use case, and constraint (e.g., outdoor safe, dimmable), plus a proof layer (e.g., UL listed, ETL certified). Apply: Build each key phrase in your listing by stacking these elements in order so it reads naturally while reducing ambiguity for both shoppers and AI agents.
Cosmo — Amazon's research system, cited as reading common-sense relations linking products to human intentions rather than just attributes and titles. Apply: Since Cosmo's internal graph isn't visible to sellers, write NPO content as relational language optimization — phrases that imply relevant relationships — and then monitor whether marketplace signals improve as a proxy for Cosmo alignment.
NPO ground rules — A set of constraints for doing NPO safely: use only accurate phrases matching real product fit, never force irrelevant terms, separate parent phrases from child/refinement phrases, use noun stacks only when they read naturally, and build phrase evidence from market data rather than imagination. Apply: Before adding any new phrase to a listing, verify it's accurate, check whether it belongs at the parent or child phrase level, and confirm it's sourced from autocomplete, ads, reviews, Q&A, or catalog data rather than guesswork.
Move 1 — Map your query plan in 5 minutes — A quick-start tactic: screenshot every Amazon autocomplete suggestion for your main product term and note the follow-up prompts Alexa for shopping offers. Apply: Use the resulting list as your query plan and treat each item as a child phrase you need to become eligible for in your listing content.
Move 2 — SQP gap analysis — Pulling Search Query Performance (SQP) reports from Brand Analytics and sorting by top ASIN to find phrases with high impressions but low click or purchase share. Apply: Prioritize fixing language and evidence on those visible-but-not-winning phrases first, and skip terms you already dominate.
Move 3 — Title as noun phrase stack — Rebuilding the product title as one coherent noun phrase (head noun, product type, material, key attribute, then use case or room) instead of a string of crammed keywords. Apply: Write titles like 'metal wall art, large abstract coastal panels for living room' so both a human and an AI agent can parse a single phrase rather than 200 characters of jammed keywords.
Move 4 — Constraint/proof phrase in first two bullets — Placing a specific constraint or proof phrase (e.g., dimmable, outdoor safe) in the first two bullet points. Apply: Identify the exact filters AI agents apply when narrowing a mission and surface them early in bullets, since missing that phrase gets you filtered out before you reach comparison.
Move 5 — Audience and occasion line — Adding a bullet or A+ content line naming who the product is for or when it's used, e.g., 'gift for horse lovers' or 'for a first apartment.'. Apply: Write at least one line describing the recipient or situation for your product, since agents route missions by recipient and situation and most listings omit this.
Move 6 — Mine reviews and Q&A for real phrases — Searching your own reviews and Q&A for how customers actually name and use the product (e.g., 'used it for my camper,' 'held up in the rain'). Apply: Copy customers' exact noun phrases from reviews and Q&A into your listing content as market-evidence-based language rather than invented phrasing.
Move 7 — Single controlled test — A test methodology of changing one variable at a time, such as adding a use case phrase to a bullet, then observing results. Apply: After each single change, watch that phrase's SQP impressions, click share, and purchase share for two to three weeks before making another change, to isolate whether the language change moved visibility.
Multi-touch attribution toggle — A new Amazon feature in the sponsored products reporting console that shows sales credit shared across every ad that influenced a purchase, next to the old last-click model. Apply: Toggle it on, compare it against your last-click data, and reconsider pausing broad or upper-funnel campaigns that show low last-click credit but may be generating assists picked up by exact-match terms.
The video reframes SEO as necessary but no longer sufficient: 'good ACO is great SEO, but great SEO by itself just isn't enough anymore,' shifting the goal from ranking to being eligible, relevant, persuasive, trusted, and machine-legible across an entire shopping mission.
Because Cosmo's relational graph is invisible to sellers, the guide's workaround is to treat NPO as 'relational-where language optimization' — structure content around phrases that imply the relationships Cosmo cares about, then watch marketplace signals rather than optimizing the graph directly.
The recommended triage method for SQP data is to prioritize phrases with impressions but low click/purchase share (visible-but-losing) over terms already dominated, treating that gap as the place to fix language and evidence first.
Constraint or proof phrases (like 'outdoor safe' or 'UL listed') act as hard filters AI agents apply when narrowing a mission — missing that phrase means being filtered out before a product ever reaches the comparison stage, not just ranking lower.
Most listings describe only the product, not who it's for or when it's used, which is why the guide singles out audience/occasion phrasing (e.g., 'gift for horse lovers') as an underused lever since agents route missions by recipient and situation.
Multi-touch attribution reframes broad and upper-funnel keywords that looked like 'dead weight' under last-click reporting as assist-generators that had been quietly earning credit for warming shoppers up before the final click.
«You're no longer trying to rank for a keyword. You're trying to be eligible across the whole family of queries that one shopping mission generates.»
— 02:28
«And good ACO is great SEO, but great SEO by itself just isn't enough anymore.»
— 02:50
«It's a phrase built around a head noun that gets more and more specific as you stack modifiers on it. So, you go from lamp to floor lamp to metal floor lamp to dimmable metal floor lamp for reading nook.»
— 03:25
«Amazon's Cosmo research shows the system reads way beyond just attributes and tiles. It reads the common sense relations that link products to human intentions.»
— 04:05
«So, the bottom line is this, the winning ASIN isn't the one that ranks for one keyword, it's the one that's eligible, relevant, persuasive, trusted, and uh machine legible across the entire mission.»
— 07:44
«Well, Amazon just rolled out a new attribution toggle in the sponsored products reporting console»
— 13:35
«ChatGPT is down to 52.7% market share. A year ago, it held over 76% and now it's barely above half.»
— 15:03
«You may encounter many defeats, but you must not be defeated. And in fact, it may be necessary to encounter the defeats so you can know who you are, what you can rise from, and how you can still come out of it.»
— 14:49
Reception
No comments are available, so audience reception cannot be assessed.
The episode's core segment is a genuinely technique-dense framework (A9/Rufus layering, ACO/QPO/NPO, seven concrete moves) presented as a faithful relay of Andrew Bezos's guide rather than the host's own research, bookended by several sponsored product plugs and quick news headlines whose shelf life is much shorter than the NPO framework itself.
15:25