Answer product and catalogue questions with your bot

A product bot is for online shops and anyone with a catalogue: products, courses, properties, plans or parts. Shoppers ask what a product page doesn't answer quickly: which one is right for me, what's the difference between these two, will it fit, is it in stock, where's my order. The bot answers from your product content, recommends options that match what the shopper describes, and shows them as cards they can click through.

For facts that change by the minute, such as stock levels, delivery dates and order status, the bot calls your own systems with the API request power-up instead of relying on content that could be hours old.

What you'll set up

1. Knowledge sources. Pick the one that matches how your catalogue is published:
   • Product pages on your site: a website data source with URL patterns that include your product pages, and a content selector for the product details.
   • A spreadsheet export: upload a CSV. Each row becomes its own entry, so include the name, price, key specs, product URL and image URL in every row.
   • A short range or unpublished details: manual content, one entry per product.
   
   Add your delivery, returns and sizing pages too. Turn on automatic syncing (Standard and above) so prices stay current.
2. Retrieval. Product rows are short, so they often miss the default match bar. Lower Document relevance to around 0.5 and raise Max context documents to 8–10 so the bot can compare several products. See Retrieval settings.
3. Instructions. Write the Prompt using the example below. See Write your bot's instructions.
4. Power-ups.
   • Show cards with Display cards for product recommendations, and Display table for side-by-side comparisons. Displays show in the website widget and API; other channels get text.
   • Call your API for live stock, prices or order status, for example "Look up an order by order number and email address".
5. Channel. Install the chat widget across your shop. On product pages, the JavaScript SDK can tell the bot which product the shopper is looking at (needs Advanced SDK features).
6. Handoff. Add Talk to a human for complaints, trade orders and anything involving payment.
7. Analytics. Review Conversations for products people ask about that you don't stock. On Pro and Enterprise, Analytics shows the most common questions.

Example instructions

You are the shopping assistant for Acme Outdoors, which sells hiking and camping gear.

Help shoppers choose the right product:
- Ask what they'll use it for (trip length, season, budget) if they haven't said.
- Recommend up to three products from the information provided, and explain in one line why each fits.
- Use Display cards to show recommendations, with the product name, price, image and link.
- Use Display table when someone asks to compare products.

Facts:
- Only mention products, prices and specs that appear in the information provided. Never invent a product.
- For stock levels and order status, use the Check stock and Order lookup power-ups. Don't guess.
- If we don't sell what they need, say so honestly.

Keep replies short and friendly.

More starting points: product questions prompts in the prompt library.

Example conversations

Shopper: I need a tent for two people for summer festivals, under £150.

Bot: Here are three that fit:

(cards: Ridge 2 at £129, quick to pitch; Meadow 2 Lite at £99, lightest to carry; Dome 2 Plus at £145, most headroom)

The Ridge 2 is the most popular for festivals because it pitches in about five minutes.

Shopper: Is the Ridge 2 in stock in green?

Bot: Yes, there are 14 in green, and orders placed before 2pm ship today.

Pitfalls

• Stale prices and stock. Knowledge is a snapshot from the last sync. Don't let the bot quote stock from content; use an API lookup, or tell it to link to the product page for current availability.
• Rows without context. A CSV row that says only "Ridge 2, 129" gives the bot nothing to match "festival tent" against. Include a plain-language description and the use case in each row.
• Retrieval too strict. If the bot says it can't find products you know exist, lower Document relevance. If it recommends unrelated items, raise it again slightly.
• Promising what the API doesn't return. If your API has no delivery estimate, tell the bot not to give one.
• Handling orders without identity checks. Ask for the order number and the email on the order before looking anything up, and have your API check they match.