Build the chatbot's knowledge base

Before you start

You need the Grow or Unlimited (500 conversations/month) plan. On the Free plan the Knowledge Base page is locked. The chatbot must be switched on: see Turn on and customise the AI chatbot.

Where the chatbot gets its answers

The chatbot reads these on its own, without you doing anything:

• your store’s active products, with description and variants;
• collections;
• your store policies (returns, shipping, privacy, terms);
• published pages of your Online Store (for example “About us” or “Shipping”);
• published entries in your knowledge base.

It doesn’t read blog posts. It doesn’t take price and availability from this reading: it checks them on your store at the moment it answers.

When you edit a product, the app updates what the chatbot knows by itself. The status of the catalogue reading is in the Catalog indexing box on the AI Chatbot page.

Add a question and answer by hand (Grow and Unlimited)

Use manual entries for anything that isn’t written in your store: opening hours, real delivery times, gift wrapping, sizing, warranties.

1. Open Knowledge Base from the left menu.

2. In the Manual entries section press + Add entry.

   What you’ll see: the Add entry window opens with the Question and Answer fields.

3. In Question type the question the way a customer would ask it, for example “Do you offer gift wrapping?”. Up to 300 characters.
4. In Answer type the full answer. Up to 2,000 characters.

5. Press Save.

   What you’ll see: the window closes and “Entry saved.” appears. The entry is in the list with the Published label: from now on the chatbot uses it.

To leave the window without saving press Cancel.

Edit, unpublish or delete an entry

Next to each published entry there’s an Edit button and a More actions menu.

• Edit: reopens the window with question and answer. Change the text and press Save.
• More actions › Unpublish: the entry stays in the list with the Draft label, but the chatbot stops using it. To bring it back press Publish.
• More actions › Delete: removes the entry.

Warning: Delete doesn’t ask for confirmation and the entry can’t be recovered. If you only want to pause it, use Unpublish.

Automatic Q&A drafts (Unlimited only)

On the Unlimited (500 conversations/month) plan the app can read products and policies and prepare draft questions and answers by itself. You check them and publish only the right ones. On Grow this section is locked, with a button to change plan.

1. In Knowledge Base, in the Automatic Q&A drafts section, press Scan now.

   What you’ll see: when it’s done, the “Scan complete” box shows how many new entries were generated and how many products and policies were analysed. The entries appear under Automatically generated entries, with the auto and Draft labels.

2. Read each draft. If needed, open More actions › Edit, fix the text and press Save.
3. To publish one, press Publish next to the entry.

4. To publish many at once, tick the box next to each (or the box at the top, which selects them all) and press Publish selected.

   What you’ll see: for example “5 entries published” appears and the entries switch to Published.

Warning: a new scan replaces the not-yet-published drafts that come from the same sources. Entries you’ve already published stay. Publish or fix the good drafts before scanning again.

Automatic scans have a daily and a monthly safety limit. If you reach it, “Safety limit reached” appears: you can scan again the next day or the next month, as the message says.

Troubleshooting

The chatbot doesn’t use an entry I wrote. Check it has the Published label. Entries in Draft aren’t used.

“Scan complete” appears with the notice “sources not updated (previous drafts kept)”. Some products or policies weren’t read this time. Earlier drafts from those sources stay. Try again later with Scan now.

“Scan failed” appears. The app couldn’t read products and policies from the store. Try again in a few minutes.

After Save the window stays open with an error. The question or the answer is empty: fill in both. The error text under the field is currently shown in Italian (“La domanda non può essere vuota.” / “La risposta non può essere vuota.”).

Next step

Conversations and leads