Train an AI Agent on Shopify, Notion & WhatsApp

A lot of stores run on the same three tools: Shopify holds the products and orders, Notion holds the policies and how-tos, and WhatsApp is where customers ask. Training an AI agent on that stack means getting the first two into the agent’s knowledge and letting it answer on the third.

This walkthrough uses ZynfoAI and is honest about one detail: ZynfoAI connects to Shopify and WhatsApp directly, but Notion goes in as exported files or published pages. For the bigger picture, from costs to choosing a tool, see our complete ecommerce chatbot guide.

What “training” means here

You aren’t retraining an AI model. You’re giving the agent sources to answer from, and rules for how to behave. When a customer asks a question, the agent searches your sources, answers from what it finds, and shows the source. If nothing covers the question, it sends your fallback message and opens a help-desk ticket for your team. Our post on how AI chatbots train on business data goes deeper.

Step 1: Connect Shopify

Install the ZynfoAI Shopify app from the Integrations page and approve it in your Shopify admin. Two things happen:

• The agent reads your catalog live. It can answer product questions and help customers find products from your catalog.
• The agent tracks orders. It looks up an order’s status and replies with it. Refund, return and cancellation requests become help-desk tickets for your team.

Products are read live from Shopify, so catalog changes show up with nothing to resync.

Step 2: Bring in your Notion pages

Notion is where many teams keep returns rules, delivery timelines, size guides and care instructions. To get them into the agent:

• Export as PDF or HTML. Notion lets you export any page to PDF, HTML or Markdown. Upload the PDFs as sources.
• Or publish the pages to the web and add the URLs as website sources.
• Or keep the files in Google Drive and connect Drive, so the agent reads them from there.

Whichever you choose, re-upload or resync when a policy changes. Write the pages the way a customer would read them. “Exchanges: tags attached, unworn, within 30 days” is better than an internal memo.

Step 3: Write the rules

In the AI Agent instructions, set the tone and the limits:

• “Ask for the order number before checking an order.”
• “Never promise a refund or a cancellation. Open a ticket for the team.”
• “If someone reports a damaged item, open a ticket for the team.”
• “Keep replies to two or three sentences.”

Step 4: Connect WhatsApp

Connect your WhatsApp Business number through Meta’s Embedded Signup. The same agent then answers on WhatsApp from the same sources, and can reply with buttons and lists so customers tap instead of type.

When a customer needs a person, the agent turns the question into a help-desk ticket with the conversation attached. Live chat handoff to a teammate, where the AI stops replying while they handle it, is on the Pro plan.

Step 5: Test with real messages

Use messages from your actual WhatsApp history, typos and all:

Customer: hi where is order 1234

Agent: Hi! Order #1234 has shipped and is on its way. (order card: In transit)

Customer: can i exchange it? size is small

Agent: Yes, exchanges are possible within 30 days if the tags are still on. Shall I pass this to the team as a ticket so they can arrange it?

Source: Returns-and-exchanges.pdf (exported from Notion)

Check that each answer shows the right source, and that anything outside policy becomes a ticket rather than an improvised answer.

Common problems, and the fix

• The agent gives an old answer. A Notion page changed but the export didn’t. Re-upload it.
• Two sources disagree. An old FAQ page says 14 days, the Notion export says 30. Remove the old one.
• The agent says “I don’t know” too often. Check Unanswered Questions each week and add the missing page.

For WhatsApp specifics, read our WhatsApp AI chatbot guide. For real conversations to test against and everything else, go back to the ecommerce chatbot guide. ZynfoAI costs from $39 a month, with no per-resolution fees.