AI Support for Shopify: What It Should and Should Not Answer

Between those two groups is a middle zone: refunds, cancellations, discount codes and address changes. These are fine to automate when you've given permission and written a rule for the case. This guide gives you a simple test and a four-zone map, so you can decide what to hand over on day one and what stays with your team. For the full list of ticket types, see 7 tickets to automate and 5 to keep for your team, listed at the end of this article.

Want an agent that answers only what you allow and passes everything else to your team? Start the 14-day trial.

The test: is the answer already written down?

Before you let AI handle a type of ticket, ask three questions:

• Is the fact in Shopify or your knowledge base? Order status, tracking number, fulfillment date and product materials are all stored somewhere. If a fact only exists in your head, the agent can't know it.
• Is the rule written? "Returns accepted within the window on unworn items" is a rule. "We look at it case by case" is not.
• Is a wrong answer small and easy to fix? A wrong tracking link costs one follow-up email. A wrong answer about an allergen can hurt someone.

If all three answers are yes, the agent can handle the ticket. If any answer is no, the ticket goes to your team.

The test also works in reverse. When a ticket you expected the agent to handle ends up escalated, look for the missing piece. Usually a fact was never written down, or a rule was left vague.

The four zones of Shopify support tickets

• Zone 1, answer alone: facts from Shopify or your policies. Typical tickets: "Where is my order?", "Do you ship to Canada?", "Is this jacket machine washable?" The agent finds the data and replies.
• Zone 2, act within permissions: actions you allowed, backed by a written rule. Typical tickets: cancel an unfulfilled order, change an address before shipping, refund a missing item. The agent checks the rule, acts and confirms.
• Zone 3, facts clear, decision human: the policy is known but the call isn't automatic. Typical tickets: return outside the window, damaged item claim, bulk order request. The agent gives the policy, then escalates.
• Zone 4, hand off immediately: judgment, risk or emotion. Typical tickets: allergic reaction, legal threat, chargeback, "I want to talk to a person." The ticket is escalated to your team.

Most stores already know what belongs in zone 1. Zones 2 to 4 are where you have to make decisions.

Zone 1: what AI should answer on its own

These tickets come in all day. The answer is in your data, and the customer mostly wants it quickly.

• Order status and tracking. The agent finds the order from the email address or order number and replies with the current status and tracking link.
• Shipping questions. Delivery times, countries served and shipping costs, all taken from your written shipping policy.
• Return and exchange eligibility. The agent checks the order date and the item against your return rules.
• Product facts. Sizes, materials and care instructions, as long as they're on the product page or in the knowledge base.

Example: a customer writes "Ordered 4 days ago, still nothing." The agent finds the order, sees it shipped yesterday, and replies with the carrier and tracking link. Your team never has to open the ticket.

There's one condition. The fact has to be written down and up to date. If a supplier changed the fabric and nobody updated the product page, the agent will repeat the old description. When that happens, fix the source, not the reply.

Zone 2: actions taken only with your permission

Relay can refund, send a discount code, cancel an order or change an address, but only with the permissions you give it and only when a written rule in your policy covers the case. Each rule says when the action is allowed and what to do when it isn't. Here are examples of rules you could write:

• Refund. "Refund an item the customer reports as missing if the order shows it was never fulfilled. Otherwise, escalate."
• Cancellation. "Cancel any order that hasn't been fulfilled yet. If it has shipped, explain the return policy instead."
• Address change. "Update the shipping address if the order hasn't been fulfilled. If it has, escalate."
• Discount code. "Send a one-time apology code when an order ships later than the date shown at checkout."

If a case doesn't match the rule, the agent doesn't act. It passes the ticket to your team. To set this up step by step, read our guardrails guide for Shopify AI support, listed at the end of this article.

Write your cancellation, address and refund rules once, then see how they apply to your real tickets. Start the 14-day trial.

Zone 3: give the facts, let a person decide

In these tickets the facts are clear but the decision needs a person. The agent shouldn't refuse and it shouldn't decide. It explains the policy that applies and passes the ticket to your team.

What you write in your rules decides how this zone works. Some rules you can add:

• Return outside the window. "State our return policy, then escalate. A team member decides on exceptions." For return cases in detail, see our guide to AI for Shopify returns and refunds, listed at the end of this article.
• Damaged or wrong item. "Apologize, ask for the order number and a description of the problem, then escalate."
• Warranty claims. "Ask the customer to describe the fault, then escalate."
• Bulk, wholesale or custom requests. "Tell the customer the team will reply, then escalate."

The more precise your rule, the less your team has to ask when they open the ticket. A vague rule like "handle damaged items carefully" gives the agent nothing to apply. A rule that says what to ask and when to hand off does.

Zone 4: what AI should never answer

Some tickets should go to a person right away. To keep them out of automation, list them in your rules as "always escalate":

• Health and safety. Allergic reactions, skin irritation, injuries, a product that broke in a dangerous way. A wrong word here can harm a customer and expose your business.
• Legal and payment disputes. Mentions of a lawyer, threats to report the store, chargebacks. These need a person who knows your history with the customer and your obligations.
• Account security. Requests to change the email on an account or anything that looks like someone else's order.
• A clear request for a person. If the customer asks for a human, they get one.

Add one more rule as a safety net: anything your rules don't cover goes to your team. The agent shouldn't fill a gap with a guess.

For the full inventory of ticket types, including the ones to keep for your team, see 7 tickets to automate and 5 to keep for your team, listed at the end of this article.

Where your zones come from in Relay

The zones aren't a setting you switch on. They come from three things you control. Your knowledge base fills zone 1. Your permissions and written rules define zone 2. Everything else goes to your team in the ticket center. If you're still deciding which tickets to hand over first, read when to automate customer support in a Shopify store, listed at the end of this article.

Sort your tickets into four zones and let Relay handle only what you allow. Start the 14-day trial.

FAQ

What questions should AI support answer on Shopify? Questions where the answer is a fact in your store data or a written policy. That covers order status, tracking, shipping times, return eligibility, product details and store basics.

Can an AI support agent issue refunds on Shopify? Yes, if you allow it. In Relay, refunds, cancellations, discount codes and address changes only happen when you've given Relay that permission and your store policy has a written rule for the case. Everything else goes to your team.

When should AI hand a ticket to a human? When the ticket involves health or safety, legal threats, chargebacks, account security, or a customer who asks for a person. Write these as "always escalate" rules. It should also hand off when the answer isn't in your knowledge base or your rules.

What happens if the AI doesn't know the answer? It shouldn't guess. Add a rule that sends anything your rules don't cover to your team. Each ticket escalated this way shows you a fact or a rule you still need to write.