A shop chatbot is useful when it finds the right product, shows a current price and can tell a customer where their order is. This guide walks through the setup. The main decision is how the chatbot finds products, and there are three paths:
-
Path A: AI Search (recommended) - ChatLab keeps its own index of your catalogue and the chatbot searches it by meaning.
-
Path B: live store search - the chatbot passes the customer's phrase to your shop's own search through the store API. Nothing is indexed.
-
Path C: training on scanned store pages (possible, not recommended for products) - ChatLab scans your product pages into the training knowledge base. It works on any plan and with any website, but prices and stock go stale between scans.
A chatbot uses AI Search or live store search for products, never both. When AI Search is on, it replaces live store search. Scanned pages can sit next to either of them, and they are the right source for informational content such as delivery, returns and FAQ.
Before you start
Create and train your chatbot before sharing it with customers. For installation, you also need permission to edit and publish the website. See How to create your first chatbot.
What Is Important in a Chatbot for E-Commerce
Start with the customer experience you want to support. Product discovery, cart actions and order lookup are separate capabilities with different prerequisites.
Help a customer find a product
Find products by describing a need
AI Search needs a connected, indexed catalog and an enabled search action. Indexed data and live lookups have different freshness guarantees.
Visitor: A light dress for a summer wedding, under $100.
Play the example or read it below. No real customer data is used.The customer describes a need rather than an exact product name.
Playback pauses when this example leaves the screen. This is a simulation, not a live chat. Widget controls inside the playback are not interactive.Read the complete example
- The customer describes a need rather than an exact product name.
Visitor: A light dress for a summer wedding, under $100.
- The bot searches the catalog.
Searching the product catalog...
- Matching products appear as Offer Cards.
Emerald wrap dress, $79. Viscose, midi length. Cobalt satin midi, $69. Viscose, midi length.
- The customer can narrow down the selection.
Visitor: Does the emerald one come in size 38?
- Variant information must come from the available product data.
Assistant: The product lists size 38. Check the product page for current availability before purchasing.
See AI Search for how the catalogue is indexed and searched, and Offer Cards for visual results.
Help a customer add a selected product
Choose a variant before adding to cart
Illustrative sequence: actual cart buttons and variant pickers depend on the platform and storefront. This playback does not execute a cart operation.
Emerald wrap dress, $79. Viscose, midi length.
Play the example or read it below. No real customer data is used.Search results provide a product card with a cart control on supported stores.
Playback pauses when this example leaves the screen. This is a simulation, not a live chat. Widget controls inside the playback are not interactive.Read the complete example
- Search results provide a product card with a cart control on supported stores.
Emerald wrap dress, $79. Viscose, midi length.
- The customer chooses a product and size using the supported cart controls.
Visitor: Emerald wrap dress, size 38.
- Only confirm success after the storefront accepts the cart operation.
Assistant: Example result: the selected variant has been added to your store cart.
- Payment and checkout remain in the store.
Assistant: Review your cart in the store to confirm quantity, price and delivery, then proceed to checkout.
This requires a supported integration and storefront setup, not just a product card. Read Adding products to the cart and test the actual store cart before enabling the feature for customers.
Help a customer track an existing order
Check an order
Requires a connected store and its enabled order action. Available fields depend on the integration. Matching an email address is not proof of identity.
Visitor: Where is my order?
Play the example or read it below. No real customer data is used.The customer asks about an existing order.
Playback pauses when this example leaves the screen. This is a simulation, not a live chat. Widget controls inside the playback are not interactive.Read the complete example
- The customer asks about an existing order.
Visitor: Where is my order?
- The default lookup needs both identifiers.
Assistant: Please give me your order number and the email address used at checkout.
- Use fictional data in this example.
Visitor: 41982, anna@example.com
- The action queries the connected store.
Checking order 41982...
- Report only information returned by the store.
Assistant: Order 41982 is marked as shipped. The carrier is DPD. Please check the tracking page for delivery updates.
Read Checking order status for lookup requirements, privacy considerations and unsuccessful lookups. Order details do not by themselves enable refunds or changes to an order.
Three Ways the Chatbot Finds Products: AI Search, Live Store Search, Scanned Pages
1. AI Search (recommended)
- ChatLab keeps its own index of the shop catalogue, separate from the training knowledge base. It does not use training characters.
- Indexed per product: name, description, categories, keywords and tags, attributes (for example colour, size, brand), product codes (SKU, EAN, barcode), price, URL and image.
- The customer asks in plain language, for example "a red dress under 300". The chatbot searches by meaning, not by exact phrase. A price range is applied as a strict filter, and exact product codes are matched too.
- Price, stock and variants are read live from the shop at the moment of answering, so the customer sees current data even between index refreshes.
- Works with Shopify, WooCommerce, PrestaShop, Shoper, CS-Cart, Ecwid, BigCommerce, an XML product feed (Google Merchant format) and an uploaded XLSX or CSV product file.
Plan you need: included in every E-commerce plan (Free 100 products, Start 2,000, Grow 5,000, Scale 20,000). On a Universal plan, add the E-commerce integrations add-on ($30 a month) to connect the store, plus an AI Search product pack ($29 a month, 5,000 products) assigned to the chatbot. The pack is what turns AI Search on for that chatbot.
2. Live store search
- The chatbot passes the customer's phrase to the shop's own search through the store API and answers with what the shop returns. Nothing is indexed, nothing has to be kept in sync, and the data is always live.
- It is a keyword search. It finds what the shop's own search would find for that phrase. It does not understand descriptions of a need, synonyms or typos the way AI Search does.
Plan you need: a Universal plan with the E-commerce integrations add-on ($30 a month). No product pack is needed. On E-commerce plans AI Search is the default and replaces live store search.
3. Training on scanned store pages (possible, not recommended for products)
- ChatLab scans your website, product pages included, and stores the text of the pages in the training knowledge base. No integration is needed.
- Good for informational content: delivery, returns, payment methods, size guides, about us and FAQ.
- Weak for products: prices and stock go stale between scans, and product pages consume training characters.
Plan you need: any plan, including Free. It is limited by the plan's training characters and number of pages. Recommended for informational pages, not for products.
A chatbot uses AI Search or live store search for products, never both. Scanned pages are the fallback for products when a store has no integration, and they sit next to either of the other two for everything that is not a product. If you are not sure which one to pick, start with AI Search. The full comparison is in AI Search, live store search or scanned pages.
Available E-Commerce Integrations
Navigate to your bot and select the Connect tab to view all available integrations.
ChatLab supports the following e-commerce platforms:
- Shopify - AI Search or live store search, order status
- WooCommerce - AI Search or live store search, order status
- PrestaShop - AI Search or live store search, order status
- Shoper - AI Search or live store search, order status
- CS-Cart - AI Search or live store search, order status
- Ecwid - AI Search, order status
- BigCommerce - AI Search or live store search, order status
- Wix - order status only
- BaseLinker - order status only
- Abicart - live store search by phrase, order status
- Product feed (XML) - AI Search over a Google Merchant feed, without order status
Path A: Set Up AI Search
Use this path on an E-commerce plan, or on a Universal plan with the E-commerce integrations add-on and an AI Search product pack assigned to the chatbot.
- Connect the shop. Open the Connect tab and connect your platform. Follow the article for your platform in the list above. If your platform has no native integration, connect an XML product feed or upload an XLSX or CSV product file under Training > E-commerce catalog.
- Let the catalogue index. Indexing starts by itself when the shop is connected. You do not have to scan product pages.
- Check the catalogue. Open Training > E-commerce catalog and look through the Products and Categories tabs. This is what the chatbot can find. Shop > Products shows the Product catalog panel with the AI Search packs on this chatbot.
- Schedule automatic re-indexing. In Training > E-commerce catalog, use Schedule auto-retrain. Re-indexing can run as often as daily and is available on every E-commerce plan and with every product pack. Refreshes are incremental: unchanged products are skipped and removed products are deleted from the index. Train now or Retrain runs a refresh on demand.
- Review the search action. In Settings > Actions, open the AI product search action of your platform, for example Shopify - AI product search. Its options are Include product card, Products in stock only and Number of products to return.
- Enable Offer Cards. Turn on Enable Offer Cards in Settings > Model & Advanced so results appear as product cards, up to 5 per answer. See Offer Cards.
Then test like a customer: describe a need instead of typing a product name, add a budget, and ask for a product by its SKU or EAN.
Path B: Set Up Live Store Search
Use this path when you want product search without an index. It needs a Universal plan with the E-commerce integrations add-on. No product pack is needed.
- Connect the shop. Open the Connect tab and connect Shopify, WooCommerce, PrestaShop, Shoper, CS-Cart or Abicart. Follow the article for your platform.
- Review the actions. In Settings > Actions, filter by your platform and check that the product search action is enabled. There is no catalogue to index and no schedule to set.
- Enable Offer Cards in Settings > Model & Advanced so results appear as product cards.
- Help the chatbot build good phrases. Live store search is a keyword search, so the phrase matters. The custom instructions below help.
You can move to AI Search later by assigning an AI Search product pack to the chatbot. AI Search then replaces live store search.
Custom instructions for live store search
Configure custom behavior under Settings > Role & Behavior. Click Enable custom role and behavior instructions and add the instructions for your platform.
These examples are for the live store search actions. If AI Search is on, it replaces them, so adapt the instructions to the actions actually enabled in Settings > Actions instead of forcing an unavailable tool. Use your store's product language, not automatically English. Bot Settings save automatically; wait for the saved status before testing.
Read more about role customization: Role & Behavior Settings
For WooCommerce
WooCommerce integration has searchProducts, getProductCategories, and searchProductBySku tools available:
First, classify the user question into one of:
- PRODUCT-BY-SKU (user provides a SKU, reference number, or product code)
- PRODUCT-SPECIFIC (variant/price/availability/recommendation)
- ASSORTMENT-OVERVIEW (brands we carry, categories, "do you sell X at all", broad offer)
If PRODUCT-BY-SKU:
- Call searchProductBySku with the provided SKU or reference code.
- Return the product details to the user.
If PRODUCT-SPECIFIC:
- Call searchProducts(text) first using a concise query in the shop's product language.
- If results are empty/weak:
1) Call getProductCategories(normalized noun) to choose best categoryId.
2) Call searchProductsInCategory with the chosen category ID and search phrase to show a few items as examples.
- Answer using tool results as the source of truth for name, price and availability.
If ASSORTMENT-OVERVIEW:
1) Call getProductCategories(normalized noun) to find relevant categoryIds.
2) For up to 3-5 top categories, call searchProductsInCategory with the chosen category ID and search phrase.
3) Aggregate from tool results; always label the list as non-exhaustive.
For Shopify
Shopify integration has a searchProducts tool with an option to show only in-stock products. Enable the Products in stock only toggle in the Shopify product search action to filter out unavailable items automatically:
First, classify the user question into one of:
- PRODUCT-SPECIFIC (variant/price/availability/recommendation)
- ASSORTMENT-OVERVIEW (brands we carry, categories, "do you sell X at all", broad offer)
If PRODUCT-SPECIFIC:
- Call searchProducts first using a concise query in the shop's product language.
- If results are empty/weak, try up to 3 reformulations (simplify, synonym, category).
- Answer using tool results as the source of truth for name, price and availability.
If ASSORTMENT-OVERVIEW:
- Call searchProducts to provide a few examples and label them "examples".
- Always be explicit when a list is non-exhaustive (e.g., "Examples include ...").
For PrestaShop
PrestaShop integration supports searchProducts and searchProductByReference tools. The reference search allows customers to find products by SKU, EAN-13 barcode, ISBN, or MPN code:
First, classify the user question into one of:
- PRODUCT-BY-REFERENCE (user provides a SKU, EAN barcode, ISBN, MPN, or reference code)
- PRODUCT-SPECIFIC (variant/price/availability/recommendation)
- ASSORTMENT-OVERVIEW (brands we carry, categories, "do you sell X at all", broad offer)
If PRODUCT-BY-REFERENCE:
- Call searchProductByReference with the provided reference code.
- Return the product details to the user.
If PRODUCT-SPECIFIC:
- Call searchProducts first using a concise query in the shop's product language.
- If results are empty/weak, try up to 3 reformulations (simplify, synonym, category).
- Answer using tool results as the source of truth for name, price and availability.
If ASSORTMENT-OVERVIEW:
- Call searchProducts to provide a few examples and label them "examples".
- Always be explicit when a list is non-exhaustive (e.g., "Examples include ...").
For Abicart, CS-Cart, and Shoper
These integrations support live store search, but action names vary. Use the enabled product search action in place of searchProducts in this example. Wix and BaseLinker provide order status only, not product search:
First, classify the user question into one of:
- PRODUCT-SPECIFIC (variant/price/availability/recommendation)
- ASSORTMENT-OVERVIEW (brands we carry, categories, "do you sell X at all", broad offer)
If PRODUCT-SPECIFIC:
- Call searchProducts first using a concise query in the shop's product language.
- If results are empty/weak, try up to 3 reformulations (simplify, synonym, category).
- Answer using tool results as the source of truth for name, price and availability.
If ASSORTMENT-OVERVIEW:
- Call searchProducts to provide a few examples and label them "examples".
- Always be explicit when a list is non-exhaustive (e.g., "Examples include ...").
Path C: Train on Scanned Store Pages
You can skip the integration and let ChatLab learn your products from the pages of your store. This path is available on any plan, including Free, within the plan's training characters and number of pages, and it works with any website.
It is possible, but we do not recommend it for products:
- Prices and stock go stale between scans. The chatbot answers with what the page said at the last scan.
- Product pages consume training characters, so a large catalogue fills the limit quickly.
- There is no live stock check, no add to cart and no order status.
Use Path C for products only when your store has no integration and you cannot provide an XML product feed or a product file. In that case follow Training a bot on an e-commerce store without an integration: it explains how to split the scan into an informational training and a product training, and gives ready element selectors for the popular shop platforms.
Whichever path you choose for products, the website scan remains the right source for informational content: delivery, returns, payment methods, size guides, about us and FAQ. See "How to Scan the Informational Pages of Your Store" below and Adding New Website Sources.
Order Status and Add to Cart
Order status and add to cart are separate actions of the store integration. They work the same way on both paths.
- Order status - available for every native store integration, including Wix and BaseLinker. See Checking order status.
- Add to cart - available where the integration and the storefront support it. See Adding products to the cart.
An XML product feed and an uploaded product file give the chatbot products for AI Search, but not order status.
Plans and Limits
Every E-commerce plan includes AI Search for a set number of products, a monthly number of chatbot responses and one shop per chatbot. Bigger catalogues add AI Search product packs. Universal plans can buy the E-commerce integrations add-on and AI Search product packs too. The full breakdown, including what counts as a product and what happens above the limit, is in E-commerce plans and limits.
Choose and Test Your AI Model
Use a model that handles your catalog and customer questions reliably. Test ambiguous product descriptions, exact product codes, variants, and order lookups. Compare quality and message credit usage before changing models.
Change the model in Settings > Model & Advanced.
How to Scan the Informational Pages of Your Store
Plan scanning your website in portions rather than all at once for better control. Scanning has two filtering levels: URL filtering (which pages enter training) and element filtering (which parts of each page become content). The checklist below covers the essentials.
-
Define proper scanning scope - use URL filtering:
- Include only URLs of informational pages (e.g., delivery, returns, FAQ, contact)
- Exclude product and category paths (e.g.,
/products/and/collections/) when products come from AI Search or live store search - Exclude low-value paths like
/tags/and archive pages - For multilingual stores, restrict scan to a primary language
-
Exclude repeating elements - use HTML Element Exclusion filter for header, footer, sidebar
-
Include only relevant content - use CSS Selector Inclusion to focus on the main content of each page
-
Use Sitemap if available - helps identify the pages more efficiently than crawling
For the full method - both filtering levels explained in depth, the recommended split into separate informational and product trainings, and ready-to-use element selectors for WooCommerce, Shopify, PrestaShop, Magento, Shoper, CS-Cart and more - see Training a bot on an e-commerce store without an integration. It is meant for stores that cannot use an integration, an XML feed or a product file.
Read more about optimizing your website scan: How to Reduce Training Characters When Scanning a Website
Additional Features Worth Enabling
Offer Cards
Enable this setting under Settings > Model & Advanced to allow the chatbot to return product cards with names, images, and prices. The cards are built from AI Search or live store search results, up to 5 per answer with AI Search.
Add this to your custom instructions to avoid duplicate links:
If knowledge base provides URLs of the source information, provide them as comma separated list in the response as a reference - only after you have answered user's question in the conversation and only if the links are not included in the product list
Read more: Offer Cards
Suggested Questions
Pre-populate quick-reply chips (shipping, returns, opening hours) under Settings > Chat Conversation. Enable dynamic suggested follow-ups for AI-generated suggestions.
Read more: Dynamic Suggested Follow-ups
Human Hand-off
Activate the Contact Human form and provide your support email for escalations.
Read more: Human Support Contact Form
Conversation & Client Summaries
When enabled, ChatLab automatically generates a summary for each conversation and maintains an up-to-date profile summary for returning customers.
Read more: Chatbot Summaries and Memory
Chat Frontend API
You can pass visitor context such as a user ID or email through the frontend API. Browser-supplied context is not proof of identity and does not automatically authorize order-history access. Never expose store API keys or private security tokens in widget configuration.
Read more: Chat API
Language Setup
ChatLab supports 90+ languages. Read more: Chatbot Language Setup
When using live store search with Shopify, WooCommerce, or other e-commerce platforms, consider what language your shop returns products in, because the phrase is matched by the shop's own search. Add this instruction to your Role & Behavior settings:
When searching for products use [shop language] product and category names in singular form.
Monitor & Iterate Daily
To ensure ChatLab remains accurate and effective:
- Review conversation logs regularly. Fine-tune answers or add content where the bot hesitates. Read more: Chatlogs
- Enable AI Conversation Insights to identify patterns and issues. Read more: AI Conversation Insights
- Leverage the automated Daily Activity Report. Read more: Daily Summary Email
Optimizing Chatbot Responses
Even well-configured chatbots benefit from regular fine-tuning. Analyze conversation logs to identify weak spots, misinterpreted queries, or repetitive fallback answers. Refine prompts, update training material, and experiment with system instructions.
Read more: How to Improve Chatbot Responses