Amazon lets sellers create product listings with AI from a URL

On 13 March 2024, Amazon announced a new way to create product listings: a seller could provide the URL of a page on its own brand website and generative AI would extract information to propose content for Amazon. The feature reduced one of the most repetitive marketplace onboarding tasks, while making one principle more visible: automation works well only when its source is trustworthy.

Amazon’s earlier tools could start from a few words or an image. Adding a URL created a direct bridge between an owned ecommerce channel and Amazon. For a brand with hundreds or thousands of references, that bridge could accelerate onboarding and reduce manual transcription.

What the feature did

According to Amazon, the system analysed the information on the supplied page and prepared a draft listing containing titles, descriptions and other product details. The seller remained responsible for reviewing it before submission.

The workflow was straightforward:

1. Enter an existing product page URL.
2. Let the system interpret its text and images.
3. Generate a proposal adapted to Amazon’s format.
4. Review, correct and complete the information.
5. Publish only after validation.

The feature initially began rolling out to US sellers. Availability in other markets therefore had to be checked account by account. The more important signal was strategic: Amazon was turning a brand’s existing content into input for marketplace operations.

A URL is not a master source

A public ecommerce page is designed to sell, not necessarily to hold every attribute required by a marketplace. It may omit materials, dimensions, warnings, compatibility, identifiers or logistics information. It may also contain promotional claims that do not fit Amazon policy.

When the URL contains incomplete, ambiguous or inconsistent data, the generated proposal inherits those limitations. AI can organise and write; it cannot independently certify that a commercial claim is true or that a measurement belongs to the correct variant.

Brands should distinguish three layers:

• Master source: the PIM, ERP or repository governing approved information.
• Owned channel: the selection displayed on the brand’s ecommerce site.
• Marketplace: an adaptation to Amazon taxonomies, attributes, policies and search behaviour.

A URL can be an excellent starting point. It should not replace data governance.

Catalogue quality remains a competitive advantage

An effective listing is more than fluent text. It must identify the product correctly, answer purchase questions and avoid contradictions across title, bullets, images, variations and A+ Content.

Before automating, review:

• Identifiers, brand, category and variation relationships.
• Technical attributes, materials, dimensions and compatibility.
• Demonstrable benefits and usage limits.
• Primary and secondary images and supporting documents.
• Restricted wording, sensitive claims and regulatory requirements.
• Translation and measurement units for each market.

A brand with structured information can use AI to scale. A brand storing it across disconnected documents merely produces inconsistencies faster.

From generation to validation

Teams spend less time beginning from a blank page and more time reviewing. That change needs an explicit quality process covering accuracy, compliance, conversion and consistency.

Human approval matters particularly in health, food, cosmetics, children’s products, electronics and any regulated category. The reviewer should compare the generated content with the approved source rather than judging whether it merely sounds plausible.

PIM and integrations

A PIM separates stable product facts from channel-specific adaptations. Technical information is maintained once, then transformed for each Amazon category, language and rule set. Integrations reduce copying and provide traceability over the version sent.

AI works best in this system as an enrichment layer: proposing bullets, summarising benefits, detecting empty fields or preparing an initial translation. The PIM preserves the source, rules validate, people approve and the connector publishes.

A controlled adoption method

Start with a small product family whose information is complete. Compare Amazon’s proposal with the live listing, record corrections and turn repeated mistakes into editorial rules.

Measure preparation time, accepted-field rate, incidents, conversion and returns linked to inaccurate information. Increase volume only after quality becomes stable.

How Sitelicon helps

Sitelicon treats Amazon as part of a commercial ecosystem. We connect ecommerce management and operations, marketplaces, PIM, integrations and content so every channel receives suitable information without losing central control.

The feature introduced in 2024 anticipated a catalogue increasingly assisted by AI. The opportunity is not simply to create listings faster; it is to build a product foundation capable of feeding new channels, markets and experiences with confidence.

Editorial note: originally published in March 2024 and reviewed on 11 September 2026 to preserve the continuity of the editorial archive.