AI Search, live store search or scanned pages: how your chatbot finds products

A chatbot in an online store is only as good as the way it finds products. Ask a store assistant for "a lightweight rain jacket for running, under 100" and you expect two or three concrete suggestions with a price and a link. Whether the chatbot can deliver that depends far less on the AI model and far more on where its product knowledge comes from and how fresh it is.

There are three ways to give an e-commerce chatbot access to your products:

1. AI Search - the chatbot works on a product knowledge base built from your catalog and finds products by meaning. Supported native integrations can check store prices and stock when answering; XML setups use the latest refreshed feed data.
2. Live store search - the chatbot passes the customer's phrase to your e-commerce platform's search endpoint.
3. Training on scanned store pages - the chatbot is trained on a scanned copy of your store pages.

This article explains how each option works, where it shines, where it fails, which plan you need and how to choose. If you only want the summary, jump to the side-by-side comparison. For a short overview with examples, see the AI Search page.

Option 1: AI Search (indexed product catalog with live data)

AI Search is the recommended way to give a store chatbot its products. It combines the strengths of the two other options. ChatLab reads your catalog through the integration or a product feed and builds a dedicated product knowledge base: every product with its name, full description, attributes, categories, tags, codes (SKU, EAN), price and images. Indexing starts by itself when the shop is connected. After that the catalog is re-indexed on the schedule you set, as often as daily, and only changed products are re-indexed. Scheduled re-indexing is available on every E-commerce plan, including Free, and with every product pack. Product pages, documents and FAQs are kept separately in the regular training data and do not count against the product limit.

How it finds products: the customer's question is matched against product meaning, not words. "A warm but light jacket for hiking in autumn, under 300" is understood as a category, a use case and a budget. AI Search narrows by price and category and ranks the products that fit the description. Supported native integrations can then read product details from the store; XML setups use the latest successfully refreshed feed data. Codes also work as exact matches, so a question such as "is 5901234123457 in stock" can identify the product, with availability based on the connected source.

Advantages

• Contextual search. Synonyms, typos, other languages, descriptions of needs and comparisons ("something like X but cheaper") all lead to the right products.
• Prices and stock from the connected source. Supported native integrations can read store data when answering. XML feeds reflect the latest successful refresh and can lag behind changes in the store.
• Filters that customers actually use. Price range and category are applied as filters, not guessed as keywords.
• Recommendations, not just lookups. When the exact item is sold out or does not exist, the chatbot proposes alternatives that match the same need.
• Clean product knowledge. Only product data is indexed, without menus, banners or "you may also like" widgets. Nothing from the catalog uses up your training characters.
• Scales with the catalog. Thousands of products are searched as fast as ten, and a refresh only touches products that changed.
• Works without a native integration through a Google Merchant XML feed or an uploaded XLSX or CSV product file, so any store can use it. See Integration with XML feed.

Disadvantages

• Requires an integration, a product feed or a product file, and a product limit that covers your catalog size.
• New products appear after a successful catalog refresh, not the second they are published. Query-time price and stock checks for indexed products depend on the native integration.
• Feed-based setups use the latest successfully refreshed feed, which may itself contain older data. A feed does not provide order status or add to cart; those require a native integration that supports the specific action.

Best for: any store where customers describe what they need instead of typing an exact name, catalogs with hundreds or thousands of products, and stores that want the chatbot to sell, not only to answer.

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. See E-commerce plans and limits.

Option 2: Live store search (product search through the store API)

When your store runs on a platform with a ChatLab integration that supports product search (such as Shopify, WooCommerce, PrestaShop, Shoper, CS-Cart or Abicart), the chatbot can call the platform's API. This is live store search: the product search action sends the customer's phrase to your store's search endpoint and presents what comes back as product cards. BaseLinker and Wix currently support order status, not product search.

How it finds products: exactly the way your store's own search box does. The chatbot extracts a phrase from the conversation, the platform matches it against product names and sometimes descriptions, and returns the hits.

Advantages

• Store data at query time. Prices, stock, variants and product links come from the store's API response; available fields and freshness depend on the platform.
• No training and no character limits. The catalog is not copied anywhere, so its size does not matter.
• Great for exact lookups. Product names, SKUs and codes are found instantly.
• Other actions, such as order status, add to cart and in-stock filtering, depend on the integration. Product search alone does not enable them.

Disadvantages

• Keyword matching, one to one. Most platform search endpoints look for the literal words. "Sneakers for wide feet" finds products only if a name or description contains those words. Synonyms ("trainers", "running shoes"), typos, other languages and descriptive needs return nothing or the wrong items.
• The chatbot has to guess the phrase. From "something warm for my daughter, she is 8" the AI must invent a search term your store engine will accept. Every guess that misses costs a round trip and a message credit, and the customer sees a "no results" answer.
• No understanding of budget or purpose. "Under 200" or "for a beginner" are not search terms. Filtering usually depends on what the platform endpoint supports, and often it supports very little.
• Quality depends on the platform. Some store search engines are excellent, some match only titles. The chatbot inherits every limitation.

Best for: stores where customers know what they want and search by name or code. Order status and cart operations are separate capabilities available only where the integration supports them.

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.

Option 3: Training on scanned store pages

ChatLab crawls your store, extracts the visible text of every page and stores it in the knowledge base. Product pages, category pages, delivery terms, size guides and FAQs all become training data. A detailed setup guide is in Training a bot on an e-commerce store without an integration.

How it finds products: the customer's question is compared with fragments of scanned pages. The chatbot gets the fragments that look most similar and writes an answer based on them.

Advantages

• Works with any store on any platform, no integration or feed required.
• Captures everything a page says: long descriptions, materials, care instructions, comparison tables, buying guides.
• The only option for content that is not a product at all: policies, FAQs, brand story, blog articles.

Disadvantages

• Outdated prices and stock. The chatbot remembers the page as it looked at training time. A price change, a promotion or a sold-out product is invisible until the next retrain. A customer who hears one price in the chat and sees another in the cart loses trust in both.
• Noise instead of product knowledge. A product page carries a menu, footer, cookie banner, "customers also bought" widgets, reviews and newsletter forms. Without careful element filtering all of it lands in the training data and competes with the product itself for the chatbot's attention.
• Too much text, too little structure. The chatbot sees prose, not fields. It does not know that "299" is the price and "S, M, L" are variants, so it cannot filter by budget or size, only guess from text.
• Character limits. Large catalogs quickly exhaust the training character limit of your plan, and the whole catalog must be rescanned to pick up changes.
• Weak recall on large catalogs. With thousands of similar products the most similar text fragments are often not the best products, and the chatbot tends to repeat the same few items.

Best for: small catalogs that rarely change, stores where the chatbot mostly answers questions about service and policies, and as the informational layer next to a live product source.

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.

Side-by-side comparison

AI Search

• Source of product knowledge: indexed catalog plus supported store lookup, or the latest feed refresh
• Matching: semantic match on product meaning, so "a warm jacket under 300" is read as a category, a use case and a price filter
• Synonyms, typos, other languages: yes
• Prices, stock and variants: store lookup where supported; latest feed refresh for XML
• Price and category filters: yes
• Exact search by SKU or EAN: yes
• Alternatives when sold out: yes
• Training characters: not used, products are counted separately
• Large catalogs: fine, refresh is incremental
• New product visible in chat: after the next scheduled refresh
• Order status and add to cart: separate native integration actions where supported; not provided by XML
• Setup: connect the integration or a feed; indexing starts by itself
• Plan: every E-commerce plan, or a Universal plan with the E-commerce integrations add-on plus an AI Search product pack

Live store search

• Source of product knowledge: a call to the platform search endpoint
• Matching: keyword match, usually one to one; needs literal keywords
• Synonyms, typos, other languages: rarely
• Prices, stock and variants: fields returned by the store API
• Price and category filters: depends on the platform
• Exact search by SKU or EAN: yes
• Alternatives when sold out: no
• Training characters: not used
• Large catalogs: fine
• New product visible in chat: immediately
• Order status and add to cart: only if the integration supports the specific action
• Setup: connect the integration
• Plan: a Universal plan with the E-commerce integrations add-on, no product pack

Training on scanned store pages

• Source of product knowledge: static copy of scanned pages
• Matching: text similarity on page fragments; guesses budget and purpose from text
• Synonyms, typos, other languages: sometimes
• Prices, stock and variants: as of the last training
• Price and category filters: no
• Exact search by SKU or EAN: only if the code is on the page
• Alternatives when sold out: no
• Noise from menus, banners and widgets: high without element filtering
• Training characters: yes, the whole catalog
• Large catalogs: hits character limits, weaker recall
• New product visible in chat: after a retrain
• Order status and add to cart: no
• Setup: scan and filter pages
• Plan: any plan, including Free

Which option should you choose?

• You sell a few dozen products and the chatbot is mainly there for delivery, returns and FAQs. Website scraping is enough. Follow the element filtering guide so product pages come in clean.
• Your platform has an integration that supports product search and customers search by product name or code. Live store search through the integration is enough and needs no product pack. Move to AI Search when you notice "no results" answers for descriptive questions.
• You have hundreds or thousands of products, customers ask in their own words, and you want the chatbot to recommend and upsell. AI Search is the right choice. It is the default in every ChatLab E-commerce plan, and a Universal plan gets it with the E-commerce integrations add-on plus an AI Search product pack.
• Your platform has no native integration. Export a Google Merchant XML feed and use AI Search over it. You get contextual search and product cards; order status and cart operations stay unavailable.

In practice the best setup is layered: informational pages through scraping, the catalog through AI Search, and order status or cart actions through a native integration that supports them. Check the integration's article for its available actions and storefront requirements.

How to enable AI Search in ChatLab

1. Open your chatbot and go to the Connect tab. Connect your store with Shopify, WooCommerce, PrestaShop, Shoper, CS-Cart or Ecwid, or add an XML feed or an XLSX or CSV product file if no suitable catalog integration is available.
2. Make sure the plan covers AI Search and your catalog size. See E-commerce plans and limits.
   • On an E-commerce plan: AI Search is included for a number of products (Free 100, Start 2,000, Grow 5,000, Scale 20,000). Bigger catalogs add AI Search product packs (5,000 products for $29 a month each).
   • On a Universal plan: add the E-commerce integrations add-on ($30 a month) to connect the store, then assign an AI Search product pack to the chatbot in Shop > Products. The pack is what turns AI Search on for that chatbot.
3. Indexing starts by itself when the shop is connected. Wait for the first indexing to finish. The indexed catalog is listed under Training > E-commerce catalog on the Products and Categories tabs, where you can check what the chatbot can find.
4. In Settings > Actions, confirm that the AI product search action of your integration is enabled. When AI Search is active it replaces the plain phrase search, so the chatbot always uses the smarter path.
5. Turn on Enable Offer Cards in Settings > Model & Advanced so results appear as product cards with an image, price and link. See Offer Cards.
6. Set the re-indexing schedule with Schedule auto-retrain in Training > E-commerce catalog, so the catalog is refreshed automatically, as often as daily. Train now or Retrain runs a refresh on demand.

Then test it the way a customer would: describe a need instead of typing a name, add a budget, ask for something similar but cheaper, and ask about an item you know is sold out. Compare the answers with the store, check the latest refresh for feed-based catalogs, and confirm final price and availability on the product page or at checkout.

Related articles

• E-commerce plans and limits
• How to set up ChatLab for your e-commerce store
• Training a bot on an e-commerce store without an integration
• Integration with XML feed
• Integration with Shopify
• Integration with WooCommerce
• Offer Cards