An AI chatbot for your Shopify store: what it should answer, and what it must hand to a person

Most Shopify stores get the same twenty questions all day: is this in stock, does it fit, when will it arrive, where is my order, can I return it. An AI chatbot that answers those well pays for itself in the first week. One that answers them badly — confidently wrong about a return policy, inventing a delivery date — costs you the customer and the chargeback.

This is the spec we built doTask's AI Agent to, and it is the checklist we'd suggest for judging any Shopify chatbot app.

What the bot should own

Product questions, from the live catalogue. "Do you have the linen shirt in medium?" should be answered from your store's actual products and inventory, not from a document you uploaded last month. The bot should show the product with its picture, price and a link, and say when something is out of stock rather than guessing. On the Shopify App Store version of doTask, the catalogue is read live through Shopify; on other platforms it can be pushed through the API or a Google Sheet.

Policies, quoted from the source. Shipping times, return windows, warranty terms. The bot should answer from your policy pages and say which page it took the answer from. A citation is not decoration: it is how the customer trusts the answer and how you catch a wrong one.

Order status, once identity is confirmed. "Where is my order?" is the most common support question in e-commerce. The bot should look it up — but only after the customer gives something that proves it's theirs: the order number plus the checkout email, or a signed-in storefront session. Never surface another customer's order because someone typed a plausible number.

Store facts. Opening hours, locations, payment methods, whether you ship to Sabah and Sarawak, whether you do gift wrapping. Boring, high-volume, perfect for a bot.

The first message at 2 am. Even when the answer is "a person will reply in the morning", an instant acknowledgement that captures the question and the customer's contact keeps the enquiry. Silence loses it.

What it must hand to a person

Anything involving money moving. Refunds, cancellations after fulfilment, partial credits, chargebacks. The bot can explain the policy and collect the details; a person makes the decision. In doTask this is a confirmation step: the assistant proposes the action and a team member approves it from the inbox.

Complaints and anger. A customer who writes "this is the third time" does not want a bot. The bot's job is to recognise the tone, apologise once, and hand over with the whole context attached, so the person doesn't ask them to repeat it.

Questions it can't source. If there is no page, document or product that answers the question, the right response is "I don't have that — I'll get someone to answer" and a notification to the team. A bot that improvises is a liability. Ask any vendor what their bot does when it doesn't know; the honest answer is the only acceptable one.

Anything medical, legal or safety-related about the product. Hand over.

How the handover should work

The bot and the team should share one thread. The customer keeps talking in the same WhatsApp, Instagram or web chat; the only thing that changes is who is typing. Things to check:

• The bot stops when a person starts. If a team member replies — from the inbox or from the Instagram app — the bot should stay quiet on that thread until it is handed back.
• The person sees what the bot said. Every bot message in the thread, with the source it used.
• Notifications reach someone. Email, a browser notification, a Slack message — whichever your team actually looks at.
• Business hours are respected. Outside hours the bot can say when a person will reply and take the question, rather than pretending someone is there.
• On WhatsApp, Instagram and Messenger, Meta's 24-hour window applies. The bot should not try to send after the window closes. On Instagram and Messenger a person can reply for up to seven days only through Meta's Human Agent feature, which the app must be approved for. How the window works.

Judging a chatbot app before installing it

Questions worth asking on a demo or a trial:

1. Where do answers come from? The right answer is "your pages, your files, your products, your policies — and it tells you which." The wrong answer is "it's trained on general knowledge."
2. What happens when it doesn't know?
3. Can I see and correct what it said? You should be able to read every conversation, mark answers as wrong, and add the correct answer as a Q&A that the bot uses next time.
4. Does it know my orders — safely? How does it verify the customer before showing order details?
5. Which channels? Storefront chat is table stakes. Your customers also write on WhatsApp and Instagram; does the same bot, with the same knowledge, answer there, in one inbox?
6. Can my team take over in the same thread?
7. What does it cost as I grow? Per conversation, per seat, or per message — and what counts as a conversation.
8. How does it get onto the store? A theme app embed you switch on is the clean way; pasting code into theme.liquid is the fragile way.

What this looks like in practice

A customer opens the chat on your product page at 11 pm: "Does the oversized tee shrink?" The bot answers from your care-instructions page, quotes it, and links the page. "Ok how long to Penang?" — from the shipping policy: 2–4 working days, tracked. "I ordered last week, #1043, where is it?" — the bot asks for the checkout email, matches it, and reads the fulfilment status from Shopify: shipped yesterday, here's the tracking link. "Actually I want to return it" — the bot explains the 14-day window, collects the reason, and hands the thread to your team with a note; in the morning someone approves the return from the inbox.

Five messages, one human decision, and a customer who was answered at 11 pm.

doTask is on the Shopify App Store; the free plan includes the storefront chat and the AI Agent. Installing it on Shopify takes about five minutes, the pricing page lists every plan, and how doTask compares with Gorgias, Tidio and Shopify Inbox sets the alternatives side by side.