Keyword Search vs Image Search in RizzitGO Product Discovery

The keyword endpoint requires a text query, page number and marketplace source. It can return results from Taobao, 1688, Weidian or Xianyu, and it accepts relevance, price and sales sorting. This works best when the source title contains usable words for a product type, model, material or color.

Its weakness is vocabulary. Marketplace titles can mix brand terms, supplier language, abbreviations and promotional phrases. Two visually similar items may not share the same words, while a broad keyword can retrieve many unrelated listings. Sorting changes the order of those results; it does not correct an ambiguous query or certify relevance.

The image endpoint requires a publicly accessible image URL, a page number and a source. The documented image sources are Taobao, 1688 and Weidian. Unlike keyword search, the current image-search schema does not list Xianyu and does not expose the same optional sort parameter.

The response fields are broadly comparable—listing ID, title, link, image, source, price, sales and discount context—so an approved application can present visual matches in a structured result set. A visual match still means resemblance at the retrieval stage. It is not an authenticity opinion and does not show that the photographed item and the eventual order are identical.

Keyword search lists four source codes, while image search lists three. That difference means the two result sets should not be compared as if they scan an identical universe. A missing visual result may reflect endpoint coverage, the selected source or the query image—not the absence of the product across every marketplace.

The practical editorial lesson is to label how a result was found. A directory that mixes keyword and visual retrieval without attribution can make its coverage look more uniform than it is. Search mode, marketplace and capture time are all part of the result's provenance.

Both endpoints return familiar product fields, but the route into those records is different. Keyword search begins with text and can apply a documented sort state. Image search begins with a public image URL and the current specification does not describe an equivalent sort input. The common output shape should not be mistaken for a common ranking method.

The public documentation does not disclose an image-similarity score, matching model or threshold. It would therefore be inaccurate to assign a percentage match or claim that the first visual result is closest unless a separate source actually provides that measure. The safe description is that the endpoint returned a set of image-search results for the selected marketplace.

This limit is useful for readers. It separates what the API contract confirms—accepted inputs and returned fields—from assumptions about the internal retrieval system that the documentation does not support.

The image URL must be publicly accessible and use a recognized image format. Beyond that technical requirement, the picture itself can vary: it may show a product against a clean background, a cropped detail, a model wearing an item or a screenshot containing text and surrounding objects.

Those differences can affect the kind of result a visual system returns, even though the RizzitGO document does not publish scoring details. A close crop may emphasize a logo or texture while a full scene may emphasize shape and color. That is why visual search is better described as another discovery route, not a replacement for reading the source title, SKU and current listing.

Near-duplicate marketplace images create another ambiguity. Multiple sellers can reuse or edit similar promotional photography. Image resemblance can connect records worth reviewing, but it cannot by itself establish a shared seller, shared inventory or identical manufacturing source.

RizzitGO is not alone in exposing more than text search. Alibaba said in May 2026 that its Qwen shopping assistant was connected to Taobao's catalog and included multimodal features such as virtual try-on. That announcement describes Alibaba's own systems—not RizzitGO—but it shows why commerce discovery is moving beyond a single search box.

For spreadsheet-style sites, the opportunity is not to claim that one mode is superior. It is to preserve enough context for readers to understand why a result appeared and what still needs independent verification on the current listing.