What's New

Shopping Assistant now analyzes product images to improve its recommendations. Until now, the assistant matched products using only text, including product titles, descriptions, variant names, some metadata. This meant it could miss visual details that only appear in images, and get tripped up by abstract or misleading product naming, such as a color called "Dynasty" when a shopper is looking for dark blue.

With this update, Shopping Assistant uses an AI-generated description of each product's main image as an additional signal when finding the best match for a customer's request.

How It Works

Each product's main image is analyzed by an AI model, which generates a structured description of its visual attributes (color, silhouette, style, details, etc.)

These descriptions are added to the product knowledge Shopping Assistant searches at query time

The catalog stays up to date automatically as products are added or images change

Why It Matters

More accurate recommendations for visual queries: customers asking for "a navy crewneck" or "a slim-fit blazer" get better matches, even when the product description doesn't use those exact words

Fewer bad recommendations from incomplete text: colors, silhouettes, and design details that only appear in images are now searchable

It works across your full catalog: no manual tagging or additional setup required