Amazon & Marketplace SEO

Two Search Engines, Two Optimisation Problems

On Google, authority, freshness and topical depth drive rankings, and the reward is a click. On Amazon, relevance to the query is filtered through conversion performance: click-through rate, add-to-cart rate, purchase rate, return rate and review signals. A listing that ranks well and converts poorly loses position quickly, because the system is optimising for the transaction rather than for the answer.

The practical implication is that marketplace SEO cannot be separated from merchandising. Imagery, price, availability, review volume and the offer itself are ranking inputs. A consultant who only edits keywords is working on a fraction of the problem.

Conversational Queries Change the Target

Shopper queries on marketplaces used to be short keyword strings. Assistants now interpret longer, situational questions: what suits a particular use case, a particular constraint, a particular environment. That shifts weight from keyword-stuffed titles toward content that actually answers the question, including the specification attributes and the review corpus the assistant draws on.

This is where the traditional marketplace playbook breaks. A title optimised for an exact-match keyword string does not help a system deciding whether a product answers a situational question. Structured attributes, honest specification copy and review content do.

Marketplace Versus Owned-Site Demand

A brand selling both direct and on marketplaces faces a cannibalisation question with no clean answer. Marketplace listings frequently outrank the brand site for commercial queries because they carry more trust signals and faster delivery expectations. Fighting that head-on wastes budget.

The workable split assigns roles: the owned site carries brand, education, range depth and the content that AI answers draw from; the marketplace carries transactional convenience. Both need to be accurate about the same product, and both need the same structured attribute data, or the answers diverge.

Style Guides and Listing Suppression

Marketplaces enforce title length, capitalisation, prohibited characters, image standards and attribute completeness through automated checks. A listing that fails them gets suppressed or downgraded, sometimes silently, and the seller discovers it when sales fall rather than when the change lands.

With thousands of SKUs across multiple marketplaces, compliance becomes a data problem rather than an editorial one. Product titles, attributes and images need a single governed source with validation before syndication, because fixing suppression listing by listing does not scale.

Catalogue Data Is the Technical Layer

There is no crawl to run and no Core Web Vitals to fix. The equivalent technical work is feed quality: attribute completeness, parent-child variation mapping, browse node placement, image compliance and consistency of identifiers across systems.

This is where marketplace SEO differs most sharply from website SEO. The practitioner needs to understand product information management, feed pipelines and category taxonomy rather than rendering and indexation. An agency that applies website technical SEO to a marketplace account will produce a report full of findings that cannot be actioned.

Reviews and Content You Cannot Edit

Reviews, customer questions and answers are ranking inputs and assistant inputs that the brand cannot directly control. Negative reviews on a specific attribute will surface in conversational answers about that attribute, accurately and repeatedly.

The honest response is product and packaging work plus responsive seller engagement, not suppression. Any approach built on incentivising, filtering or removing genuine reviews is both a platform policy violation and a consumer law exposure.

Platform Change Is the Constant

Search algorithms, content module formats, advertising interfaces and policy rules all change on the platform's schedule. A strategy built on a specific ranking behaviour has a shelf life measured in months. The durable parts are the ones that survive any algorithm change: accurate attributes, complete specifications, genuine review signals and consistent brand content.

Platform Policy

Marketplaces publish policies covering restricted product categories, prohibited claims, review manipulation, listing content standards and brand representation. Breaches are enforced by the platform, not by a regulator, and the penalty is commercial: suppression, listing removal or account action. There is no appeal to a court, only to the platform.

For a brand, the exposure is that a reseller or distributor can breach policy on a listing carrying your brand name. Brand registry gives some control over who can list and what content appears, which makes it a prerequisite rather than a nice-to-have.

Australian Consumer Law Still Applies

Selling through a marketplace does not remove obligations under the Australian Consumer Law. Misleading claims about a product, false discount framing, and review manipulation all remain the seller's responsibility regardless of which platform hosts the listing.

Country-of-origin, safety and category-specific claims carry the same requirements they would on an owned site. A marketplace listing is a marketing representation in the eyes of the law.

Track One: Marketplace Search

Start with the catalogue. Attribute completeness, correct browse node placement, parent-child variation structure and identifier consistency determine whether a product is even eligible for the queries you want. Keyword work applied on top of a broken catalogue produces nothing.

Then listing content: titles that respect the style guide while carrying the terms that matter, bullet points that answer the questions shoppers actually ask, enhanced content modules that carry specification detail, and imagery that meets the standard. Backend search fields carry the terms that do not fit in visible copy, without duplicating what is already indexed.

Finally the performance loop. Because conversion drives position, the listing work has to be tested rather than assumed. Change one element, observe, keep what works. This is slower than website SEO and more empirical.

Track Two: AI Assistants and Agentic Commerce

Conversational shopping assistants answer from the same product data plus the review corpus. Structured attributes and honest specification copy are what make a product quotable; a title stuffed with keywords is what makes it incomprehensible to a system trying to match a situational query.

Agentic commerce protocols extend this further: assistants that can transact need machine-readable offers, availability and pricing. Most marketplace sellers will not implement protocol support directly, but they will feel the effect through how their catalogue data is surfaced, which makes attribute quality the thing to invest in now.

Platform reporting exposes impressions, clicks, click-through rate, units and revenue by search term for a limited window, alongside business reports covering sales and inventory. That is enough to measure listing performance and not enough to see the whole demand picture.

What Cannot Be Measured Cleanly

The window between a shopper seeing a listing and buying it is not attributable when they return later through a different entry point. Organic marketplace sales and marketplace advertising interact in ways the reporting does not separate, so a listing change and a campaign change in the same week produce an uninterpretable result. And cross-marketplace demand cannot be compared directly because the reporting differs by region.

The discipline that keeps this honest: change one variable at a time where possible, report the window and the definition with the number, and do not claim a result the reporting cannot isolate.

Phase One: Catalogue and Account Audit

Listing health review, suppression and policy issue identification, attribute completeness assessment, variation structure mapping, search term performance baseline and an AI-answer baseline for the core product set. Two to three weeks for a focused catalogue; longer where SKUs run into the thousands across multiple marketplaces.

Phase Two: Strategy and Prioritisation

A ranked plan covering catalogue remediation, listing content, attribute strategy and the marketplace versus owned-site split. Sequenced by revenue impact and by how quickly a change can be verified, because marketplace changes need observation windows.

Phase Three: Implementation and Testing

Listing content specifications, attribute remediation plans for the product information system, variation restructures, and staged changes with defined observation windows. Where a team executes the changes, I write the specifications and review the result against them.

Phase Four: Measurement and Iteration

Monthly reporting against the baseline, quarterly catalogue health review, and a standing watch on platform policy and format changes that affect the listings.

The published retail engagement on this site covered the owned-site side of the same problem: technical remediation, template and category work, and measurement built around revenue rather than rankings. Marketplace work applies the same discipline to a different surface, with catalogue data replacing the crawl and conversion rate replacing indexation as the health metric.

The share-of-voice case study is relevant for a different reason. Marketplaces are contested categories in the same sense banking is: a small number of large sellers absorb most of the demand for a given term, and the way to gain ground is to work the terms where you can genuinely win rather than the head terms you cannot.

In Scope

Out of Scope

Account suspension is worth calling out. It is a specialist legal and platform-relations matter, not an SEO matter, and a consultant who claims otherwise is not helping. If that is the situation you are in, resolve it before commissioning search work, because nothing else will function while the account is restricted.