Training a chatbot on a product catalogue sounds like a big technical task. In practice, it is mostly about making sure your product pages say what customers need to know, then letting the assistant read them.
This guide walks through setting up a store assistant, from preparing your pages to the weekly habit that keeps answers accurate.
Step 1: check your product pages first
An assistant answers from your content. If product pages are thin, answers will be thin. Before training anything, check your best sellers for:
- Dimensions, materials and weight.
- Fit notes for clothing and footwear.
- Compatibility for tech and accessories.
- Care instructions.
- What is included in the box.
These are the details customers ask about. As we showed in what an unanswered product question costs a small store, missing details become unanswered questions, and unanswered questions become lost sales.
Step 2: prepare your policy pages
Product pages cover products. Policies cover everything else customers ask:
- Delivery times, costs and countries.
- Returns and exchanges.
- Payment options.
- Order tracking.
- Contact details and hours.
Write these in plain language. They answer a large share of presale questions.
Step 3: crawl your store
With SpideyChat, you point the crawler at your store's address and it reads your pages, up to 150 of them, building a searchable index. For larger catalogues, prioritise categories, best sellers and policy pages. The training guide covers each option.
Want a 24/7 AI Employee for Your Website?
See how SpideyChat can answer customer questions, capture leads and handle enquiries while your team is offline.
Step 4: add Q&A for common questions
Some questions apply across your whole range, like "do you offer gift wrapping?" or "are your products vegan?" Add these as question and answer pairs. The free FAQ generator can draft a starting set from your website.
Step 5: test with real questions
Do not test with "what do you sell?" Test with the questions customers actually ask, taken from your inbox, in their words. Score each answer as right, vague or wrong. Vague and wrong answers almost always mean a product page is missing information.
The same test works for software documentation, described in how to train a website assistant on product docs.
Step 6: install it on your store
Adding the assistant is one script tag. The installation guide covers Shopify, WooCommerce, Wix, Squarespace and other platforms. SpideyChat for ecommerce stores then answers presale and order questions at any hour, and captures questions it cannot answer with the shopper's email.
Step 7: review the gaps weekly
After launch, the assistant logs questions it could not answer. Each is a missing detail on a product page or policy. Fix the most common ones each week and re-crawl.
Over a few weeks, product pages get noticeably better. That helps every shopper, including the many who never ask, and it helps your search visibility. The value of those presale questions is covered in shoppers who ask before buying are your best customers.
Keeping the catalogue current
Products change. New ranges launch, items are discontinued, prices move. Re-crawl after significant changes, and remove pages for discontinued products so the assistant does not recommend things you no longer sell. For more ecommerce-specific setup tips, see how to train an ecommerce chatbot on your product catalogue.
What good looks like
A well-trained store assistant answers most product and policy questions instantly and accurately, and hands the rest to you with context. Your product pages are more complete than when you started, and customers who ask get answers in seconds rather than hours.