What an AI Chatbot Should Answer on a Shopify Store (and What It Shouldn't)
Shipping, returns, sizing and stock questions are perfect for a support bot. Order lookups, refunds and exceptions belong with you. Here's how to split it.
Published Updated 6 min read
On a Shopify store, an AI chatbot should answer the questions shoppers ask before they buy: shipping times and costs, returns, sizing, materials, compatibility and "do you ship to my country?". It shouldn't try to handle anything about a specific order, refunds, or exceptions to your policies. Those need a person, or Shopify's own order tools.
I build ChatBento, and most of what makes a store chatbot useful or annoying comes down to drawing that line in the right place. Here's how I'd draw it.
Why pre-purchase questions are the sweet spot
A shopper on a product page with a question has two options: find the answer, or leave. Most won't email you and wait a day to find out whether a jacket runs small.
Pre-purchase questions have three things in common that make them ideal for a bot:
- The answer is the same for everyone. Your returns window doesn't change from shopper to shopper.
- The answer is already written down somewhere, or should be: a product page, a policy page, a size chart.
- Speed decides the sale. An answer in five seconds keeps the cart alive. An answer tomorrow usually doesn't.
That's exactly what an AI chatbot trained on your store's pages does well: it finds the relevant text and answers in the shopper's own words.
What it should answer
Here are the question types I'd make sure your bot handles well, with where the answer should come from.
| Question type | Example | Source of the answer |
|---|---|---|
| Shipping | "How long to Germany?" "Is shipping free over $50?" | Shipping policy page, plus a Q&A for cut-off times |
| Returns and exchanges | "Can I return sale items?" "Who pays return shipping?" | Refund policy, with the exceptions written as Q&As |
| Sizing and fit | "Does the medium fit a 38-inch chest?" | Size charts as text, fit notes on product pages |
| Materials and care | "Is it machine washable?" "Is this real leather?" | Product descriptions |
| Compatibility | "Does this case fit the Pixel 9?" | Product descriptions, a compatibility Q&A |
| Stock and restocks | "Will the blue one come back?" | A Q&A you keep current, or a handoff |
| Payment and discounts | "Do you take PayPal?" "Is there a student discount?" | FAQ or Q&A, including the "no" answers |
| The business itself | "Are you a real shop?" "Where do you ship from?" | About page, contact page |
Two notes on that table.
Size charts often live in images. If your size chart is a picture, a bot that reads your pages can't read the numbers in it. Add the chart as text on the page, or as a Q&A.
Stock changes faster than content. A bot answering from your pages knows what the page said when it last read it, not what's in the warehouse right now. For "is it in stock?", the product page itself is the source of truth. Keep the bot to general answers ("sizes that are sold out show as unavailable on the product page") and let it hand off restock questions you can't answer in advance.
What it shouldn't answer
ChatBento can't look up orders or change anything in your store: it answers from your content, and that's all. That's a deliberate limit, and I'd keep the same line even with a tool that could do more.
"Where is my order?" This is the most common post-purchase question in ecommerce, and it's specific to one customer. The answer lives in Shopify: the order confirmation email, the shipping confirmation with tracking, and the order status page linked from both. The bot's job is to point there ("Your shipping confirmation email has a tracking link. If you can't find it, leave your email and order number and we'll check"), not to guess.
Refunds and cancellations for a specific order. The bot can explain your policy. It shouldn't promise that a particular order qualifies. Anything that commits money goes to a person.
Exceptions. "It's two days past the return window, can you make an exception?" Maybe you will. That's your call, not the bot's. I wrote about why this matters in Your Chatbot's Answers Are Your Promises.
Anything needing personal data. Addresses, payment problems, account changes. Collect a contact email and reply from your inbox.
Set up the handoff for post-purchase questions
Since the bot will see "where's my order?" questions whatever you do, make the handoff good:
- Answer the general part first. Point to the tracking link in the shipping email and the order status page.
- Then offer the form. In ChatBento, when the bot can't answer, it offers a short contact form (name, email, message) and emails it to you with the conversation.
- Ask for the order number in the form message. Add a Q&A like "For questions about a specific order, include your order number" so the bot tells shoppers what to send.
- Reply fast. A handoff that sits for two days is worse than no chat at all.
Teach it your store, not just your homepage
Most bad store chatbots are trained on too little. Before launch:
- Let it read your whole storefront. Product pages, collections, and your policy pages: shipping, refunds, privacy, terms. ChatBento reads your published pages by following your links and sitemap.
- Write Q&As for what isn't on any page. Dispatch cut-off times, holiday delays, which countries you don't ship to, how exchanges actually work, whether gift wrap exists. Our free FAQ generator can draft ten questions from any page, which you then check and correct.
- Write the "no" answers. "We don't ship to PO boxes." "Sale items can't be returned." A bot that has never seen the no is the one that says yes.
- Remove old promotions. A Black Friday page from last year still quoting 30% off will be quoted back to shoppers.
Test it like a shopper
Before you turn it on, ask it these, worded the way your customers write:
- "how long does shipping take to [your second-biggest country]"
- "can i return something i bought on sale"
- "what size should i get if im between sizes"
- "where is my order" (it should point to tracking and offer the form, not invent a status)
- "can you give me a discount" (it should only mention discounts you actually offer)
- "whats the weather like" (it should politely decline)
Every answer that's more generous or more specific than your real policy is a content problem. Fix the page or the Q&A, then ask again.
Where to put it
On a Shopify store, the chatbot should be on every storefront page, because questions come up on product pages, not on the contact page. ChatBento installs with one line pasted into your theme's theme.liquid file, above </body>. Shopify's checkout pages are separate from your theme, so the bot shows while people browse, not while they pay, which is usually what you want.
Check your product pages on a phone after installing. The chat bubble should never cover the "Add to cart" button. If something else already sits in that corner, move the bubble to the other side.
The short checklist
- Shipping, returns and payment questions answered from your policy pages
- Size charts available as text, not only images
- Q&As for cut-off times, exceptions and the "no" answers
- "Where's my order?" points to tracking and the contact form
- Refunds, cancellations and exceptions always go to a person
- Old promotions removed from what the bot reads
- Tested with real shopper wording on a phone
The step-by-step install for your theme is in the Shopify guide. It takes about two minutes, and ChatBento has a free plan, so you can test it on your own store before paying anything.