Home AI agent Training the AI agent

Training the AI agent

Last updated on Aug 14, 2025

The AI agent needs relevant information about your store and support processes to respond accurately to customers. Two methods accomplish this:

Knowledge is your information library from Shopify (products, site pages) and helpdesk (help articles, training responses, canned responses) that the AI agent accesses. Knowledge is dynamically provided based on how well information matches customer requests.

Guidance consists of custom instructions you write in natural language—like explaining processes to a new teammate. Use guidance to define AI agent behavior and responses for specific shopper scenarios. For example, guidance tells the AI agent what to say and do when someone requests a return.

Knowledge sources available to the AI agent

The AI agent accesses multiple information types to handle customer queries effectively. View all available data by navigating to AI agent → Knowledge sources.

Available knowledge sources:

  • Help articles: Help portal content from your help center

  • Canned responses: Response templates created by human agents

  • AI training responses: Saved responses from successful human-agent interactions. Manually add question-and-answer pairs through the Knowledge portal

  • Shopify products: Product catalog data from your connected Shopify store

  • Shopify website pages: Store website content and page information

AI training responses

Responses you can save from good human-agent interactions. These provide the AI agent with examples of how to handle various customer questions, improving its ability to respond effectively.

How to add an AI training response

You can find an "Add AI training response" button by hovering over any message sent by a human agent. It opens a popup where you can edit the response (automatically polished by AI) and save it to your knowledge library.

In the playground, you'll see a "Revise answer" button when you hover over any message sent by the AI agent. You can give the AI the expected correct response and save it to the training responses.

Guidance

When the AI agent picks up a ticket, it considers guidance before any other knowledge source or action. While shoppers don't see guidance directly, it significantly shapes their experience with your store. Your guidance enables the AI agent to handle shopper inquiries consistently with your brand.

How guidance works

Guidance provides specific, pre-defined responses and actions when the AI agent recognizes customer requests or scenarios.

Response examples

If a customer asks about refunds, the AI agent automatically provides your pre-written refund instructions.

Action examples

Define specific actions like assigning conversations to particular agents, adding labels, or automatically resolving issues.

Efficiency benefits

Regular guidance updates help the AI agent handle common requests efficiently, saving team time on repetitive inquiries.

Writing effective guidance

Intent field

Use specific, customer-focused scenarios that match how customers describe their issues:

  • Good: "Customer wants to return item after 30 days"

  • Good: "Customer received wrong size shoes"

  • Good: "Customer asking about shipping delay"

  • Poor: "Returns" or "Shipping" or "Order issues"

Description field

Add context and variations to help the AI agent recognize the scenario. Include alternative phrases customers might use:

Example 1: "Customer received the wrong size, color, or style and wants to exchange or return. May say 'this is too small,' 'wrong color,' 'not what I ordered,' or 'doesn't match the photo.'"

Example 2: "Customer's order is taking longer than expected to arrive. May ask 'where is my order,' 'when will it ship,' 'tracking shows no updates,' or 'it's been X days.'"

Example 3: "Customer wants to cancel their order that hasn't shipped yet. May say 'cancel my order,' 'I don't want this anymore,' 'changed my mind,' or 'ordered by mistake.'"

How to respond field

Provide step-by-step instructions with specific actions:

Example for wrong size return:

IF order was placed within 30 days, THEN:
1. Apologize for the inconvenience
2. Confirm we accept returns within 30 days
3. Send link: [Your return portal URL]
4. Explain: "You'll receive a prepaid return label within 24 hours"
5. Add label: "return-requested"

IF order was placed over 30 days ago, THEN:
1. Apologize that the return window has expired
2. Explain our 30-day policy
3. Add a private note with summary of the issue
4. Handover to a support agent

Testing and troubleshooting guidance

Test guidance using the AI agent playground where you can impersonate a customer and their order history. Ensure guidance is enabled in the Guidance settings.

Common issues:

Missing description variations

Instead of: "Customer wants refund"

Use: "Customer wants refund for defective, damaged, or unsatisfactory product. May say 'this is broken,' 'poor quality,' 'not as described,' or 'want my money back.'"

Unclear response instructions

Instead of: "Process return"

Use: "1. Verify purchase date, 2. Send return form link, 3. Add 'return-processing' label, 4. Snooze for 3 days"

Unsupported actions

Only instruct the AI agent to perform available actions.

Here's the full list of available actions

Order management:

  • Get latest order information for customers

  • Add tags to Shopify orders

  • Remove items from orders

  • Change shipping addresses

  • Cancel orders

  • Hold order fulfillment

Conversation management:

  • Mark conversations as resolved

  • Mute conversation notifications

  • Change conversation status (open, resolved, snoozed)

  • Set or clear conversation priority

  • Assign conversations to teams or agents

  • Escalate to human agents

  • Add private notes to conversations

  • Add labels to conversations