Strategy
Chatbot or AI agent: what changes for an online store
Chatbot or AI agent for an online store: three questions to tell them apart and know when each is enough. A ShopGuru guide. Book a 30-minute call.
Some online stores have already paid for a chatbot that replied "Sorry, I didn't understand your question". The question we hear now is a different one: is a chatbot with AI enough, or does the store need an AI agent for online store customer service? The difference comes down to three questions, and the answer isn't always the agent.
Article summary
- A chatbot talks to customers from a script, a menu or a list of frequently asked questions.
- An AI chatbot writes more natural replies, but it's still limited to the text it was given and, in most cases, doesn't act in the store.
- An AI agent answers with the store's knowledge, checks systems such as the order or the carrier, and takes actions within limits set by the store owner.
- Three questions separate a chatbot from an agent: where it gets the answer, whether it can do anything, and what it does when it doesn't know.
- A chatbot is enough when the repeated questions are general; an agent makes sense when they depend on each order and on the store's exceptions.
- The store is answerable for what its chatbot or agent tells customers, so what it does when it doesn't know matters more than how fluent it sounds.
- A 30-minute call with a ShopGuru engineer helps you see whether moving to an agent makes sense or whether a chatbot will do.
Chatbot, AI chatbot and AI agent: three different things
The word chatbot covers very different things today. A rule-based chatbot works with menus and keywords: "1 for orders, 2 for returns". It's the format of many WhatsApp chatbots. It answers well what someone foresaw and programmed, and fails when the customer phrases things differently.
An AI chatbot uses a language model to understand the question and write the reply. It reads the text it was given, such as the FAQ and the store's policies, and answers more naturally. As a rule, that's where it stops: it talks, but it doesn't look up an order or change it.
An AI agent is a different category. In Anthropic's definition in the guide Building effective agents, an agent is a system in which the model directs its own steps and the tools it uses to reach the result, instead of following a fixed path written in code. In a store, those tools are the order, the carrier, invoicing and the catalogue.
"A chatbot answers what was foreseen. An agent deals with what the customer needs, within the store's rules." ShopGuru
Three questions to tell a chatbot from an AI agent
Labels don't help much. These three questions work with any tool and can be answered in a demo.
Where does it get the answer?
A chatbot answers from the text it was given. If the returns policy says 30 days, it repeats 30 days. It doesn't know whether that order has already left the warehouse, whether the customer has already exchanged twice or whether the carrier is running late.
An agent adds the facts of the case to the store's knowledge: the real status of the order in the store and with the carrier. The answer becomes about that order.
Can it actually do anything?
Some chatbots end the conversation with a link or with "contact us by email". Others show an order's status, if someone programmed that flow, but they can't step outside it.
An agent takes actions: it corrects an address while the order is still in the warehouse, opens a return within the return period, sends the label. At ShopGuru, the store owner chooses one of three levels for each type of action: does it alone, asks for approval or never does. On day one, every action with an effect asks for approval.
What does it do when it doesn't know?
The rule-based chatbot replies "Sorry, I didn't understand". The AI chatbot carries the opposite risk: it writes a plausible, well-phrased reply that the store never approved.
The ShopGuru agent says "I don't know. I'll ask.", tells the customer when the answer will come and hands the conversation to your team with a summary and a suggested action. The question is logged as a gap: your team answers it once and, from then on, the agent answers it. After the Implementation, ShopGuru engineers keep testing the answers and, if they find a wrong one, propose a correction. We describe this handover in after-hours customer service.
"The difference between a chatbot and an agent shows in the question nobody foresaw." ShopGuru
The same message with a chatbot and with an agent
An example, with a fictional kitchenware store. At 10:40pm, a customer writes on WhatsApp: "Has order 20841 shipped yet? I'd like to add a 28 cm frying pan."
A rule-based chatbot replies with the menu. An AI chatbot politely explains that changes are requested by email. Neither knows whether the order is still in the warehouse.
The agent checks the order: it hasn't left and only ships tomorrow at 11am. It confirms the 28 cm frying pan is in stock. Adding an item to a paid order is an action this store set to "asks for approval", so it tells the customer the team will confirm by 10am and leaves the action ready to approve, with the amount to charge. In the morning, the decision takes seconds and the order ships at 11am with the frying pan.
| Question | Rule-based chatbot | AI chatbot | AI agent |
|---|---|---|---|
| Where it gets the answer | Script and menus | FAQ and policy text | Store knowledge and order data |
| Unforeseen question | "I didn't understand" | Plausible reply, not always right | "I don't know. I'll ask." and handover to your team |
| Actions in the store | Only the ones programmed | As a rule, none | The ones the store owner allowed, by autonomy level |
The store is answerable for what its chatbot says
In February 2024, British Columbia's Civil Resolution Tribunal, in Canada, ordered Air Canada to compensate a passenger. The website chatbot had told the passenger they could apply for a bereavement fare after buying the ticket, which the airline didn't allow.
Air Canada argued it wasn't liable for information given by the chatbot; in the decision Moffatt v. Air Canada, 2024 BCCRT 149, the tribunal wrote that the airline is responsible for all the information on its website, whether it comes from a page or a chatbot.
The case isn't about a store or about Portuguese law, but it shows the customer's yardstick: what the store writes, the store honours. That's why fluency matters less than behaviour when there's no answer.
"Better an 'I don't know, I'll ask' than a made-up answer with good grammar." ShopGuru
When a chatbot is enough for an online store
Not every store needs an agent. A chatbot, rule-based or AI, can be enough when the repeated questions are general (opening hours, shipping, payment methods) and your team can deal with whatever it doesn't solve the next day, at no cost.
An agent starts to make sense when "where is my order" arrives every day, exchanges and address changes take up your team's time, and the store has exceptions no script covers, such as different shipping for the islands. If the store already uses its e-commerce platform's AI, the right comparison is in platform AI or custom agent.
Frequently asked questions
Isn't an AI chatbot already an AI agent?
Not always. An AI chatbot writes replies from the text it received. It becomes an agent when it checks the store's systems, takes actions within set limits and hands over to your team whatever falls outside the rules.
Is a WhatsApp chatbot enough for a small store?
It can be, if the questions are general and your team answers the rest the next day. When most messages are about specific orders, an options menu hands the work back to your team. What changes between the WhatsApp app and the API is covered in WhatsApp Business for your online store.
Can I keep the chatbot I already have?
It depends on what it does today and which channels it's on. In the Diagnostic, we map the channels and tools the store already uses, and the plan says which ones we connect, which we don't and which are still to be assessed.
Where to start
The first step is to read the store's last 50 conversations and count how many depend on a specific order or an exception. With that count done, a first 30-minute call with a ShopGuru engineer helps you decide between a chatbot and an agent. Book a call.
About this article
Written by ShopGuru's engineers based on the design of the service. Published and reviewed on 28 September 2026. The store, the customer and the messages in the example are fictional. Sources: the guide Building effective agents, by Anthropic (December 2024), and the decision Moffatt v. Air Canada, 2024 BCCRT 149, by British Columbia's Civil Resolution Tribunal (14 February 2024). This article is for information only and is not legal advice.