> ## Content Index > Fetch the complete content index at: https://kandid.ai/blog/llms.txt > Use this file to discover other available public pages before exploring further. # Shopify AI Chatbot Product Discovery: How-To - URL: https://kandid.ai/blog/shopify-ai-chatbot-product-discovery-how-to/ - Published: 2026-10-29T16:47:53.000Z - Updated: 2026-10-29T16:47:53.000Z - Author: Pulkit Garg - Tags: Chatbot Setup & Operations, Product Discovery & Recommendations, Shopify AI Chatbots, Product Discovery, Shopify, Personalization, Chatbot Setup, Product Recommendations > **Quick Summary:** Shopify chatbots can only recommend products they can read, so clean product details, fill and map metafields such as size and material, and ground answers in your Shopify Catalog. Turn responses into fast product cards with in-chat add-to-cart, clear upsell rules, and human handoff, while testing real shopper questions before launch. Measure the discovery funnel weekly and separate attributed revenue from incremental lift through A/B testing, with Kandid supporting catalog-grounded discovery and cart actions. A shopper asks your Shopify AI chatbot product discovery prompt for something waterproof in size 10, and it suggests jackets you don't stock. The bot answered well but never read your size metafields. This guide fixes Shopify AI chatbot product discovery end to end: map the Shopify Catalog channel, clean product data, and measure what chat actually sells. Kandid builds for this stack, so the steps reflect real store setups. ## Before You Start: Make Your Catalog AI-Readable A chatbot can only recommend what it can read. Your product data needs three things: clear titles and descriptions, structured metafields, and correct channel mapping. Without them, recommendations miss. ### Audit titles, descriptions and metafields A Shopify AI chatbot needs clean titles, benefit-led descriptions, and metafields for attributes like size, material and fit. Run this check first: - Are titles specific ("Merino running socks, ankle") rather than generic? - Do descriptions answer sizing and care questions? - Are variants complete: color, size, stock? ### Map custom fields for AI channels Metafields store your custom data, but AI channels only see fields that are mapped. In Shopify, open Catalog, then check channel mapping and product metafields so attributes like fabric or use case reach the chatbot. Fill gaps in metafield definitions before testing any chatbot prompt. Any AI assistant, including [Kandid's Shopify AI sales agent](https://kandid.ai/shopify-ai-sales-agent?ref=kandid.ai), recommends better when attributes are structured this way. ## Step 1: Configure the Chatbot to Ground Answers in Your Catalog ### Give the assistant your product and store content An ungrounded bot guesses, so it recommends jackets when a shopper wants waterproof size-10 shoes. Feed it your product titles, descriptions, variants, prices and stock status, plus policies and sizing guides. Clean [Shopify Catalog](https://shopifyranked.com/glossary/shopify-catalog/?ref=kandid.ai) data helps AI channels read products accurately, and filled metafields give the bot attributes like material or fit to filter by. ![Merchant adding product attributes on laptop](https://assets.snowseo.com/organization-813abc9e-b233-44b2-ae76-4bd670b7e4d1/brand-R0iMcAIFDKu8Brq0IQGR6SjFzQ1H1NWI/library/ai-images/ai-image-1793292425230-e9li18.webp) Merchant adding product attributes on laptop ### Write discovery questions and intents Map the questions shoppers actually ask: budget, use case, size, gifting, delivery. For each one, define an intent and the product attributes it should filter on, for example "gift under ₹2,000" filters price and category. Tools like Kandid can train directly on your catalog data, which cuts manual question writing. > Test five real shopper questions before launch. Vague answers usually point to thin product content, not a broken bot. ## Step 2: Turn Answers Into Recommendations and Cart Actions ### Design in-chat product presentation Show each recommendation as a product card, not plain text. Include the image, title, price and one or two attributes the shopper asked about, like waterproofing or size. A quick view button lets people check details without leaving the chat. Keep cards to three or four per reply so the choice stays easy. Cards should match your theme and load fast on mobile, where most chat traffic happens. ![Shopper viewing product cards in a mobile chat](https://assets.snowseo.com/organization-813abc9e-b233-44b2-ae76-4bd670b7e4d1/brand-R0iMcAIFDKu8Brq0IQGR6SjFzQ1H1NWI/library/ai-images/ai-image-1793292445327-udt0bh.webp) Shopper viewing product cards in a mobile chat ### Set cart actions, upsell rules and handoff Every card needs a next step. Add-to-cart should work inside the chat so shoppers never lose their place. For upsells, set simple rules: recommend a size-matching accessory or a bundle only when the cart value fits your margin. Define handoff triggers too, so a human takes over when a shopper hesitates or asks about shipping delays. Tools such as Kandid support this add-to-cart flow, and the trade-offs of AI versus rules are covered in [Shopify AI Chatbot vs Rule-Based Bot: Which Should You Choose?](https://kandid.ai/blog/shopify-ai-chatbot-vs-rule-based-bot-which-should-you-choose/). ## Step 3: Measure Discovery Without Overclaiming To measure sales attributed to a Shopify AI chatbot, tag chat sessions first, then count orders where a chat session appears in the customer journey as assisted conversion. Treat that as attributed revenue, not proof the chatbot caused the sale, because many shoppers already intended to buy. ### Track the discovery funnel Count four stages weekly: sessions with product interactions, add-to-cart rate from chat, checkout starts and completed orders. Compare chat sessions with non-chat sessions on the same traffic source. If chat visitors convert no better, discovery is not working. For worked examples, see our guide to tracking chatbot revenue and AOV impact. ### Separate attributed revenue from incremental lift Run A/B tests where half of eligible shoppers see the assistant. The lift versus the control group is your incremental revenue. Attributed numbers usually overstate it, so report both figures and the test period together before you scale spend. ![Kandid homepage](https://assets.snowseo.com/organization-813abc9e-b233-44b2-ae76-4bd670b7e4d1/brand-R0iMcAIFDKu8Brq0IQGR6SjFzQ1H1NWI/screenshots/screenshot-homepage.png?v=1782889980333) Kandid homepage Your bot should suggest products your shoppers actually want, not guess. Kandid's [AI sales agent](https://kandid.ai/shopify-ai-sales-agent?ref=kandid.ai) uses your store and product data to guide discovery. See how it works. ## Frequently Asked Questions ### Q1: How do I drive product discovery with a Shopify AI chatbot? Clean your product data first, fill size and material metafields, map the Shopify Catalog channel, ground the bot in your catalog, then test real shopper questions and track chat add-to-cart rate weekly. ### Q2: Which metafields matter most for chatbot recommendations? Attributes shoppers filter by: size, fit, material, color, price and use case. Map them in Shopify Catalog so AI channels can actually read them. ### Q3: Why does my chatbot recommend products I don't sell? Usually thin or unmapped product data, not a broken bot. The assistant guesses when metafields are empty or channel mapping misses your catalog. ## Conclusion Clean product data, mapped Shopify Catalog metafields, catalog-grounded answers, product cards with add-to-cart, and honest measurement: these five steps turn a support bot into a discovery engine that recommends what you actually sell.