Editorial review
This article is reviewed by the Yanitly editorial team for product accuracy, privacy and security claims, and integration references. Technical claims are kept aligned with official documentation or observed Yanitly product behavior.
Updated: 2026-10-18
In social commerce, shoppers rarely type a product name or SKU. Fashion, footwear, accessories, cosmetics, and home textile brands receive screenshots, shared posts, and in-store photos every day. A shopper sends an image on WhatsApp, Instagram DM, or Facebook Messenger and asks "Do you have this in black?", "Is size M still available?", or "How much is this jacket?"
Text-only bots stall at this point: they either ask for a product code the shopper does not know or drop the conversation into the agent queue. Yanitly's e-commerce live support automation extracts product attributes from the shopper's image, compares them with your catalog, and, when it is confident, sends the correct color photo of the matching product into the chat. When it is not confident, it does not guess; it offers a few candidates or asks for the product name and color.
The most important rule of selling from photos is never to quote a price, stock level, or order detail from an image you are not sure about. Trust comes from accurate matching combined with a firm no-guessing rule.
Mobile shopping starts with visual discovery. A shopper who sees an item in a Reel, a creator's post, or the brand's own feed does not know its exact catalog title. A product listed as "Double-Breasted Linen Blend Blazer" is simply "that beige jacket" in the shopper's mind.
Visual questions on messaging channels usually arrive in four formats:
Teams that handle this by hand spend the day searching the catalog for photos, checking stock sheets, and picking images from a gallery. In the evening and during ad peaks, every delayed reply cools the shopper's intent to buy.
Instead of letting a general chat model interpret the image freely, Yanitly passes it through a step-by-step pipeline that can be audited. The output of each step is the input of the next, so everything the assistant says is backed by a real catalog record.
When an image arrives, the system first checks that the file is a readable picture. An AI model then extracts only commercial product attributes in a structured form: category, color, pattern, cut, neckline or sleeve type, and the look of the material. If a person appears in the photo, the system does not try to work out who they are; it only notes that a person is present and keeps the analysis on the garment, shoe, or accessory.
Data sent to the AI API used for this analysis is not used for model training by default. The photo itself stays in your conversation history like any other conversation media and follows your own retention and deletion rules. There is also a daily usage limit per business; when it is reached, the image is processed as a short description and the conversation continues.
The extracted attributes are compared with your active catalog through rule-based scoring. Category, color family, variant details, and product tags are scored, and the strongest few candidates are selected. Their catalog photos are then reviewed side by side with the shopper's photo to settle the ranking.
The result is stored on the message. The assistant can only reach it through the latest or the previous image in the conversation; it cannot make up an image address or a message identifier on its own.
The most expensive mistake in visual selling is looking at a blurry or multi-product photo and quoting the wrong item, or preparing an order summary for the wrong variant. Yanitly does not leave this risk to the model's good intentions; it limits it with server-side rules.
A photo is treated as matching a single product only when several conditions hold at the same time:
If any of these is missing, the match counts as unconfirmed. While the match is unconfirmed, or while the product the assistant wants to quote contradicts the image, the cart, quote, and order steps do not run. Instead, the assistant presents up to three close candidates or asks for the product name and color.
If the file is corrupted, the photo is too dark, or no product can be picked out, the assistant does not guess. The first time, it politely asks for the product name or color; if a second unreadable image arrives, it hands the conversation to a member of your team instead of tiring the shopper.
The second half of visual selling is sending the right photo back. When a shopper asks "Can I see this dress in burgundy?", resending the general cover photo erodes trust.
The image tools only work on messaging channels that support images: WhatsApp, Instagram DM, and Facebook Messenger. Product photos are not sent in public comments; the shopper is invited to a private message instead. On phone calls, the voice assistant does not use image tools.
When the channel's AI mode is draft, FAQ only, or off, automatic image sending is also off and your team stays fully in control. Before publishing a change, you can use the AI Simulator in the panel to preview which photo each color returns. The same channel separation underpins our WhatsApp chatbot solution.
Once the match is clear, the conversation can move straight into an order:
Visual product search is only as good as your catalog. Uploading a primary photo for every color, keeping product names and tags simple, and keeping stock current directly improve match quality. Matches always move forward with the shopper's confirmation, so an assistant asking "Did you mean this product?" is not a failure but part of the design. To run the flow across your WhatsApp, Instagram, and Messenger channels together, explore our WhatsApp automation platform.
AI live chat for ecommerce teams: answer order, shipping, return, product and campaign questions faster while reducing cart hesitation.
Guides on visual product recognition, color-specific catalog image delivery, cash-on-delivery consent workflows, Click-to-WhatsApp ad funnels, and smart order/return verification across WhatsApp, Instagram DM, and Messenger.
Yanitly extracts product attributes such as category, color, pattern, and cut from the image in a structured form. It then compares them with your catalog through rule-based scoring and ranks the strongest candidates by reviewing their catalog photos side by side with the shopper’s photo.
No. A single match is accepted only when the score is high, the gap to the runner-up is clear, and the visual comparison is positive. Otherwise the assistant prepares no price or order summary; it offers up to three candidates or asks the shopper to confirm the product name and color.
Each color in the catalog has its own primary photo, and only that color’s photo is sent when a shopper asks for it. If no photo exists for the requested color, the assistant does not present the cover photo as that color; it says the photo is unavailable and suggests the colors that exist.
The analysis focuses only on product attributes such as clothing, shoes, and accessories; it does not try to identify the person and only notes that a person is present. Data sent to the API used for the analysis is not used for model training by default, and the photo itself is kept in the conversation history under your retention and deletion rules.
