Amazon listing analyzer · Built on COSMO research

You’re still stuffing keywords. Amazon’s AI reads relationships.

Amazon’s COSMO framework ranks products by 15 semantic relationships, not keywords. Zervox shows what your listing covers and what it’s missing, with specific fixes per section.

Sign upSee how it works
Based on Amazon Science researchSIGMOD 2024 paper15 semantic relationships
What COSMO actually does

Keywords show up. Relationships rank.

The old Amazon matched query strings to listing text. The new Amazon runs on a knowledge graph that links products to 15 relational slots — the ones Amazon’s own research teams identified as the patterns shoppers actually reason with.

used_for_func
used_for_aud
used_for_eve
used_in_loc
used_in_body
used_with
+ 9 more
A listing that mentions “yoga mat” ranks shallowly. A listing that covers used by beginners, used in home practice, used with props, used for hot yogaranks across all the queries those relationships generate. COSMO is how Amazon closes the gap between what shoppers think and what they type. Your listing either speaks that language or it doesn’t.
How it works

From listing to actionable fixes.

No seller account access. No algorithm guesswork. You paste the content you already own and Zervox checks it against the framework Amazon published.

01
Submit your listing
Paste your title, bullets, description, and A+ content. Upload product and A+ images. Or autofill from an ASIN.
02
Zervox analyzes all 15 relationships
Your copy and images are scored for semantic coverage across every COSMO relationship, not just keyword density.
03
Get fixes you can apply today
Quick wins from your images, prioritized recommendations per section, and a coverage map. Most quick wins take five minutes to implement — value the moment the report opens.
Sample output

Not a score. A map of what to fix.

Every Zervox report gives you a coverage map across 15 COSMO relationships, quick wins from your existing images you can apply in five minutes, and prioritized recommendations per section. Below: a preview from a mid-market yoga mat listing analysis.

zervox.app / analysis / Bamboo Yoga Mat — 8mm Non-Slip Natural Cork Surface
COSMO Coverage — 15 Relations
4
Strong
5
Partial
4
No
2
N/A
Quick Win — Image Shows, Text Doesn’t
MAIN
Shows

Cork surface texture clearly visible, suggesting non-slip grip on hard floors.

Add to text: “Natural cork surface for stable grip on wood, tile, and polished studio floors — no slipping during downward dog or warrior poses.”
Prioritized Suggestions
1
R12 · Do people like me use this?

Mention specific user types in BP1 and the description: home practitioners, beginners learning poses, traveling yogis needing portable equipment. Listing currently describes the product but not the people who buy it.

2
R02 · What activities can I use this for?

Beyond yoga, mention adjacent activities the 8mm thickness supports: meditation, pilates, stretching routines, floor exercises, kids’ tumbling. Each activity is a search query you can rank for.

Relation Detail
R09Where can I use this?
PARTIALSHOWN
mentionedspecifictypicalreinforcedclear benefit
Text assessment

Listing mentions home practice but misses other locations where yoga mats are commonly used. Studio practice, outdoor retreats, and travel use cases are absent even though the cork surface and 8mm thickness directly support them.

Found in
BP2“Perfect for home yoga practice and meditation.”
Visual assessment

Main image shows the mat rolled out on a wooden floor (home context). A+ image 2 shows the mat in a studio setting, but this is not reinforced in text.

Suggestion · Priority 2

Add location coverage across BP3 and A+ text: reference studio use (grip on polished floors), travel use (lightweight roll, carry strap compatible), and outdoor practice (cork surface is moisture-resistant). Tie each location to a product attribute you already claim.

What you get

Every analysis delivers.

  • Concrete examples for every gap
    Guidance grounded in what your listing actually says and shows.
  • Quick wins from your existing content
    Payoffs you can capture fast using material you already have.
  • Hidden conflicts, surfaced
    Mismatches between title, bullets, and images that quietly erode buyer trust.
  • Buyer questions your listing can't answer
    The specific objections costing you sales.
  • Wasted bullet slots identified
    Redundant bullets pointed out, with ideas for what belongs there.
Full feature list
Science

Where COSMO came from.

COSMO isn’t an SEO blogger’s theory. It’s production infrastructure Amazon published openly in peer-reviewed venues. Zervox is built on that public research.

2023
FolkScope — the precursor
Amazon researchers publish FolkScope at ACL 2023, a knowledge graph of purchase intentions mined from co-purchase behavior. It covers two product domains and introduces the idea that commonsense knowledge can be extracted from user behavior at scale.
June 2024
COSMO paper released at SIGMOD 2024
Amazon and HKUST researchers publish the COSMO system: 6.3M nodes, 29M edges, 15 relation types across 18 product categories. It becomes Amazon’s #1 most-viewed paper and #5 most-read blog of 2024.
Today
Deployed across Amazon
COSMO is live in Amazon search relevance (+28% F1 in reported experiments), navigation (+8% engagement in A/B tests, +0.7% sales), and session-based recommendations. It sits alongside Rufus, Amazon’s GenAI shopping assistant, as part of the same knowledge infrastructure.
Foundation
Built on 30+ official sources
Beyond the COSMO paper itself, Zervox is grounded in 30+ official Amazon Science publications, peer-reviewed papers, and engineering materials covering product knowledge graphs, attribute extraction, visual search, and conversational shopping. Some are core to COSMO. Others are loosely related but inform how the broader Amazon AI stack reasons about products.
We don’t speculate about algorithms. We apply what Amazon published.
FAQ

Questions, answered.

No. Zervox does not analyze keyword density, search volume, or ranking positions. It measures semantic coverage across COSMO’s 15 relationship types. If you want keyword tooling, use Helium10 or Jungle Scout. Zervox answers a different question: does your listing speak the semantic language Amazon’s AI actually uses?
Writing tools generate copy for you. Zervox evaluates what you already wrote and shows the gaps. You stay in control of voice, positioning, and claims. Think of it as a compliance check against Amazon’s semantic model, not a ghostwriter.
No. Zervox never requests or stores your Amazon seller credentials, never logs into your account, and does not automate any actions inside Seller Central. All listing content comes from you — either pasted manually or fetched by ASIN from a public product data provider.
Your analysis results, stored in our database so you can revisit them. No PDF copies are archived. You can request full anonymization of your data at any time from your account settings — the deletion kicks in after a 7-day cooldown so nothing is lost by accident.
About five minutes end-to-end. You’ll see a progress indicator while the analysis runs, so you can leave the tab and come back to the finished report.
One analysis costs 5 credits. Credits are $1 each, so a full analysis is $5. You can buy credits in packs with added value built in — see the pricing page for details.
The fastest demo is to sign up, top up, and run one real analysis on a listing you already know well. You’ll see more from the report than from any walkthrough.