Buy a chatbot platform if you need support answers fast and your questions are mostly about orders and policies. Build your own shopping agent if you want it to sell, with your catalog and rules behind it. Either way, it only answers as well as your product data and policy pages, and you're responsible for every answer it gives.
That's my view in one line. A store chatbot is only as good as your policy pages. The model matters less than most vendors admit. What it reads matters more.
This guide covers the two routes, what the bot must know, where people still matter, and how to test it. It's part of my wider guide on how to use AI in ecommerce.
Who is responsible for what a store chatbot says?
You are. In February 2024, a Canadian tribunal ruled that Air Canada had to compensate a passenger after its website chatbot wrongly told him he could claim a bereavement fare after travel.1 Air Canada argued the chatbot was responsible for its own actions. The tribunal called that a remarkable submission and said the chatbot was still just part of Air Canada's website.1
That case is from a small claims tribunal in British Columbia, so it isn't binding law elsewhere. The logic will travel, though. If your bot promises a 60-day return on a 30-day policy, expect to honour it.
So the first job isn't picking a vendor. It's making sure every fact the bot might quote is written down, current and in one place.
What does a store chatbot need to know?
Three kinds of data, and a way out to a person.
Where a store agent finds each answer
- Product data answers fit, materials, compatibility and stock.
- Policy pages answer delivery times, costs, returns and refunds.
- The order system answers "where is my order?" and "can I change it?"
- A person handles exceptions, complaints and anything involving money back outside the rules.
Most bad chatbot answers come from the second row. Policies written as vague prose ("we aim to deliver quickly") give the AI nothing to quote, so it fills the gap. The free AI policy checker scans your returns and shipping text and shows which facts an agent can't find.
Should you buy a chatbot platform or build your own agent?
It depends on the job you want done.
| Chatbot platform | Your own shopping agent | |
|---|---|---|
| Main job | Support: orders, returns, FAQs | Selling: advice, comparisons, carts |
| Time to launch | Days | Weeks |
| Reads your catalog | Usually through an app | Directly, with your own rules |
| Who sets the guardrails | The vendor, with your settings | You |
| Cost model | Monthly fee, often per conversation | Build cost plus model usage |
| Best for | Small and mid-size stores | Brands with a big or complex catalog |
Platforms are the fast route. Gartner's 2026 Magic Quadrant for conversational AI platforms covers vendors such as Google, Salesforce, Kore.ai and Yellow.ai, and says the market is shifting towards agentic AI and stronger governance.2 For a Shopify store, the app store has dozens of smaller options.
Your own agent is the route when the bot should sell. Anthropic's commerce blueprint gives developers working code for a shopping agent that searches the catalog, builds carts and hands them to your checkout.3 I compare the ways Claude works with a store in Claude for ecommerce.
My own keyword data points the same way. US searches for "conversational chatbot platform for ecommerce" dropped to nearly zero over the year, while searches for Claude and AI with ecommerce rose.6 Store owners are asking about agents, not chatbot widgets.
Where do people still matter?
At the edges, and the edges are where loyalty is won or lost.
Klarna is the well-known example. In February 2024 it said its AI assistant had taken over two-thirds of its customer service chats, 2.3 million conversations in its first month.4 By May 2025 it was hiring human agents again, and its CEO said he wanted customers to always be able to reach a person.5
The research backs a mixed model. In an NBER study of 5,179 support agents, an AI assistant raised issues solved per hour by 14%, and by 34% for new and less skilled staff.7 The gain came from AI helping people, not from removing them.
So design the hand-off before the bot. Every conversation should have a visible "talk to a person" option, and the bot should pass the full conversation along so the customer doesn't repeat themselves.
How do you test a store chatbot before launch?
Ask it the 20 questions that cost you money when answered wrong.
- Collect real questions from your support inbox, the awkward ones included.
- Write the right answer for each, with the source page.
- Ask the bot each one, then ask it again in different words.
- Try to make it promise something, like an extra discount or a late return.
- Fix the source page, not the bot's prompt, wherever an answer is wrong.
- Repeat monthly, and after every policy change.
Step 5 is the one people skip. If the bot gets returns wrong, the returns page is usually unclear. Fix it there, and every AI that reads your site gets the right answer too, ChatGPT and Google included. That's the idea behind conversational commerce done well.
Questions people ask
What is the best AI chatbot for ecommerce?
The one that reads your real catalog and policies and hands off cleanly to a person. For support, a platform built for your store system is the quickest start. For selling, build an agent on your own data.
Can I use ChatGPT as a chatbot on my store?
Pasting a ChatGPT link won't do it. You need a chatbot platform or an agent built on a model like GPT or Claude, connected to your product data, policies and orders.
Is a store responsible for its chatbot's answers?
Yes. In Moffatt v. Air Canada (2024), a Canadian tribunal held the airline responsible for wrong information its website chatbot gave a customer.
How much does an ecommerce chatbot cost?
Platforms usually charge a monthly fee, often based on conversations. Your own agent costs a build plus model usage. Compare both against the support hours it saves.
Will a chatbot replace my support team?
Rarely. AI handles routine questions and helps staff answer faster. Klarna hired human agents again in 2025 after leaning hard on AI.