Ecommerce chatbot: types, costs and when it actually sells

In short: an ecommerce chatbot is software that answers shoppers in a window on your store, day and night. A rule-based chatbot follows ready-made buttons. An AI chatbot understands messages written in the shopper's own words and, when it is connected to your catalog and orders, shows products, checks deliveries and adds items to the cart. Only the second kind leads shoppers to an order. CloudCart's Aura Chat is that kind, with 5 skills you switch on one by one.

What is an ecommerce chatbot and how does it work?

A chatbot is software that holds a conversation with a person by text or voice. The first well-known chatbot, ELIZA, was built at MIT between 1964 and 1967 by Joseph Weizenbaum and replied using keywords and fixed templates (Wikipedia: ELIZA). Most AI chatbots today run on a large language model (LLM), the same kind of technology behind ChatGPT, Claude and Gemini.

In an online store, a chatbot goes through 4 steps with every message:

1. It understands the question. A rule-based bot looks for a keyword or waits for a button tap. An AI bot understands "I need something quiet for the bedroom, the room is small" even when no word from the product name appears.
2. It finds the data. The answer comes from a stored text (rule-based bot) or from the store's own data: catalog, stock, published pages and orders.
3. It acts. It shows a product card, puts a product in the cart, checks an order status or files a return request.
4. It logs the conversation. You see what shoppers asked and which conversations ended in an order.

For sales, steps two and three make the difference. A chatbot that answers from general knowledge sounds clever, yet it does not know your prices, your stock or your shipping terms. A chatbot that reads only from your store gives narrower answers, and they are correct.

Types of ecommerce chatbots

Website chatbots come in 5 types: button menus, keyword bots, general AI chatbots, AI chatbots connected to the store's data, and live chat with people. The first two follow rules, the next two use AI, and in the last one a person replies.

Type	How it answers	Good for	Weak spot
Button menu (scripted flow)	The shopper picks from buttons: "Shipping", "Returns", "Opening hours"	3-10 repeated questions	Does not understand anything outside the menu
Keyword bot	Looks for a word in the message and returns a stored text	Short questions such as "shipping cost"	Gets lost with a typo, a synonym or two topics in one sentence
General AI chatbot	A language model trained on your website text or uploaded files	Questions about services and terms	Cannot see stock or orders; may invent an answer if it has no limits
AI chatbot connected to the store	A language model with access to the catalog, stock, pages and orders	Product choice, advice, order status, returns, discounts	Only as good as your product data and pages
Live chat	A member of your team replies	Complex cases, B2B, expensive products	Only during working hours and only as many people as you have

Aura Chat, the AI shopping assistant in the AURA family of CloudCart, is the fourth type. It answers from the store's data: the catalog, published pages, orders and physical locations. It does not answer from general knowledge and does not invent a product, a price or a condition. The one exception is the "Product research" agent, which is off by default: it reads the manufacturer's website and manuals for technical data, but never takes a price, stock level, shipping or return terms from there. Where the two disagree, your catalog wins. For live chat with people, CloudCart connects to the Live Chat app (for the LiveChat service) and Zopim (Zendesk Chat).

AI chatbot vs rule-based chatbot: which one fits your store?

A rule-based chatbot is enough for a store where shoppers ask the same 5-10 things. An AI chatbot is the better choice when shoppers choose between many products, write with typos or in another language, and want advice. The table compares the two on real store tasks.

Task	Rule-based chatbot	AI chatbot connected to the store
"How much is shipping to a parcel locker?"	Yes, if you wrote the answer in advance	Yes, quotes your Shipping page with a link
"A gift for my dad under €50, he loves fishing"	No	Asks follow-up questions with buttons and shows products
"Should I take this one or the other one I looked at?"	No	Compares and recommends one product with a reason
"wo ist meine bestellung" on an English store	Rarely recognised	Understands, answers in German and asks for the order number and email
Out-of-stock product	Recommends it if it is in the script	Hides it when you set a minimum stock level
New product in the catalog	You rewrite the script	Finds it once it is in the search index (usually within seconds)
Maintenance	By hand, after every change	You keep products and pages complete; the bot reads from them

Cross-border sellers feel this most. Aura Chat replies in whatever language the shopper writes in, whatever the store's language, so a shopper from Vienna and one from Warsaw both get an answer they can read. The best practice is a mix: the AI answers freely, inside limits you set. In Aura Chat those limits are the skills you switch on, the catalog rules (minimum stock, minimum price, categories, brands and tags it never offers) and up to 8,000 characters of your own instructions about tone and store rules.

Can a chatbot actually sell? 5 conditions

A chatbot sells when it takes the shopper from the question to the cart without sending them off to search on their own. A bot that only replies with text and a link serves customers, but rarely closes an order. To sell, a chatbot has to meet 5 conditions.

1. It sees the catalog and stock. Recommending a sold-out product loses the shopper. In Aura Chat you set a minimum stock level below which a product is not shown; a product without tracked stock counts as available.
2. It shows the product with a cart button. The product card in Aura Chat has a photo, brand, price (the old one struck through during a sale) and "Add to cart", which puts the product in the shopper's real cart.
3. It asks questions when the shopper is unsure. The "Finding products" skill asks with cards of 3-6 buttons instead of demanding an exact product name.
4. It answers the last doubt. The "Advising on products" skill gives a recommendation with a reason, the return window from your pages and the product's rating, then offers the purchase.
5. It speaks first at the right moment. Proactive messages fire on rules: seconds on the page, pages viewed, whether the shopper came back to the same product. At most once per visit, and by default never in the cart, checkout or account.

A sixth condition helps but is optional: a discount. Aura Chat gives a single-use personal code only for the reasons you described, up to the cap you set (percentage, amount or free shipping), valid for 1 to 90 days. A reason you did not describe gets a "no", however politely the shopper asks. How codes are set up in the store is in the guide to promo codes and discounts in CloudCart.

Customer service chatbot: which questions can it take over?

A customer service chatbot takes the repeated after-purchase questions: where is my order, how do I return an item, how much is shipping and where is your shop. Aura Chat has separate skills for them, and your shipping and return terms come from your published pages even when the skills are off.

Shopper's question	What Aura Chat does	What you need
"Where is my order?"	Asks for the order number (or code) and the email on the order, then gives the status, payment, contents and how far the parcel has got with the courier	"Orders" skill switched on
"I want to return a product"	Says whether a return is possible and how long is left, sends a 6-digit code to the email on the order and files the request	"Returns" skill and the Aftercare app
"How much is delivery to my country?"	Answers from your shipping page and never quotes a page it has not opened	A separate Shipping page
"Do you have a shop in town?"	Shows a card for the location with the address, a call button and directions in Google Maps	Physical locations entered in the admin
"Why doesn't my code work?"	Explains what the code is valid for and why it was rejected	"Promo codes" skill

Returns are a good example of why the data matters. Under EU consumer law, a shopper buying online has 14 days from delivery to cancel the purchase without giving a reason (Your Europe: returns). The chat states your return window from your own Returns page, so the page has to say it clearly. Your team then answers only the cases that need a person. How Aura Chat works together with Aura Search and Aura Studio is covered in the article on the AURA AI assistant in CloudCart.

How much does an ecommerce chatbot cost? 4 pricing models

The cost of a chatbot depends on how you pay: a monthly subscription to a chat vendor, an agency project, your own build on a language model API, or credit for conversations inside your store platform. Compare the total cost for a year, including the time spent maintaining it.

Model	How you pay	Hidden cost	Sees catalog and orders?
Chat service subscription	Monthly, by the vendor's plan	Connecting the catalog, a separate login and separate reports	Depends on the integration with your platform
Agency build	One-off project fee plus maintenance	Every change to the script or catalog is a new job	Only if the project includes it
Own bot on an API (OpenAI, Anthropic, Google)	Per processed token (OpenAI API pricing) plus development	Developer time, security, GDPR compliance	Only what the developer connects
Aura Chat in CloudCart	Prepaid credit for conversations	No separate integration: the chat is part of the store	Yes: catalog, stock, pages, orders, returns

Aura Chat is charged by conversation. A new install gets a small starting credit so you can try the chat with real customers. Credit you buy is added to the balance and does not expire. A visitor who never opens the chat costs nothing, and a proactive message only starts spending when the shopper replies. The "Usage" tab shows the average cost of one conversation and roughly how many conversations the remaining credit covers, based on the average conversation over the last 30 days. When the credit runs out, the shopper only sees that the chat is temporarily unavailable, with no mention of payments.

Important: CloudCart takes no commission on sales, including orders that follow a chat. You pay your plan and the chat credit; 0% transaction fee on sales applies at any revenue.

How to add an AI chatbot to your store in CloudCart: 7 steps

You add an AI chatbot to a CloudCart store from the admin, with no code: install Aura Chat, switch on the skills, set the limits and style the window. The chat appears on the storefront right after installation.

1. Prepare your pages. The chat answers about shipping, returns, warranty and payment only from your published pages and finds them by title. Create separate pages with clear titles: "Shipping", "Returns", "Warranty". A condition that exists only in an image, a PDF or a banner is invisible to the chat.
2. Install the app. In the admin: Apps → Aura Chat (/admin/apps/aura_chat) → Install. The chat is live on the storefront at once. On install, all skills are off and the chat answers only from your pages.
3. Switch on the skills. In "Knowledge & Skills" (/admin/apps/aura_chat/knowledge) turn on "Finding products" and "Advising on products" for sales, "Orders" for order status, "Returns" for returns (needs the Aftercare app) and "Promo codes" for personal codes.
4. Set the catalog limits. In "Finding products": minimum stock (1 hides sold-out products), minimum price and up to 100 categories, brands and tags the chat never offers.
5. Write the instructions. In "Instructions for the assistant" (up to 8,000 characters) describe your tone and store rules, for example: "We sell tools. Ask what the customer is working on before recommending anything. Don't promise next-day delivery." Your instructions are added to the assistant's core rules and cannot switch them off: it still will not invent a product, a price or a condition.
6. Style the window. In "Settings": store name (the assistant speaks for the store, as "we"), brand colour, position left, centre or right, up to 6 opening questions, and the line under the message box that says an AI is answering.
7. Set up proactive messages. In "Proactive Sales" a new store has one rule: after 10 seconds on a page, the chat asks "Can I help you with something at {store}?". You can have up to 5 rules, each with a message of up to 300 characters, for example "45 seconds on a product page". They are checked top to bottom and the first match speaks. The starter sentence comes in English or Bulgarian; if your store runs in another language, rewrite it in that language. A product name inside the message and the "page type" condition need the Datalayer app.

If your store is new, an AI assistant such as Claude Code, ChatGPT Codex, Gemini CLI or Cursor can do the groundwork from step 1 for you. With a PAT token from the admin, it reads your products and pages, lists the products that lack a description or attributes and fills the gaps from one sentence. The more complete your products are, the more precisely the chat answers. Bulk changes like this are covered in AI agent for an online store: how many hours and how much money it saves.

How do you know if your chatbot pays off?

A chatbot pays off when the profit from orders after a conversation is higher than the cost of the conversations. In Aura Chat both numbers sit in one app: the "Overview" tab shows revenue after a chat, and the "Usage" tab shows the average cost of one conversation.

Metric in Aura Chat	How it is calculated	What it tells you
Conversations	Conversations started in the period	How many shoppers use the chat
Orders assisted	Orders from the same browser or customer after a conversation	How many buyers talked to the chat before buying
Conversion after a chat	Orders assisted ÷ conversations	The conversion rate of shoppers who wrote
Revenue per conversation	Assisted revenue ÷ conversations	Compare it with the cost of one conversation
Encouraged orders	Orders containing a product the chat showed on a card	Closer to the chat's direct effect
Complaints per 100 conversations	Thumbs-down ratings per 100 conversations	Where answers don't help

A worked example with sample numbers: a store has 400 conversations in a month, 24 assisted orders and an average order of €60. Conversion after a chat is 6%, assisted revenue is €1,440 and revenue per conversation is €3.60. Compare €3.60 with the cost of one conversation in "Usage": the gap also has to cover the cost of the goods sold, because revenue is not profit.

Read the two order counts differently. An "assisted" order means the shopper chatted and then bought, and the two may be unrelated. An "encouraged" order contains a product the chat showed, so it is closer to the direct effect. Orders are matched to conversations every hour. Compare conversion after a chat with your store's overall conversion rate: the formula and typical figures are in online store conversion rate: formula and benchmarks, and the AI agent can calculate your other store metrics from one sentence, as shown in ecommerce KPIs: 10 metrics.

The "Conversations" tab shows every conversation with the full transcript, topic, rating and cost. Transcripts are kept for 6 months. The questions shoppers ask most often are ready-made topics for new pages, FAQs and product descriptions.

Chatbot plus AI agent: how chat questions make your store better

The chatbot talks to shoppers, and the AI agent works for you. Aura Chat answers on the storefront, while an AI assistant such as Claude Code, ChatGPT Codex, Gemini CLI or Cursor, connected to the store with a PAT token, changes products and prices from one sentence. Together they close a loop: the chat shows what shoppers ask, the agent fills in what is missing, and the chat starts answering more precisely.

The starting point is conversation topics. Aura Chat puts every finished conversation into one of up to 12 topics you define, and a new store starts with standard topics such as "Product recommendation", "Product information", "Availability", "Order status", "Shipping" and "Returns and complaints". The "Overview" shows how many conversations each topic has. The topic that grows is next week's job for the AI agent.

What the chat shows	What you tell the AI agent	What the store gains
"Product information" is growing	"Show me products in Jackets without material or a size chart and fill in the attributes from the descriptions."	The chat compares and advises instead of saying it has no data
"Availability" is growing	"Which products are below 5 units? Give me a table I can reorder from."	Fewer "out of stock" answers right when the shopper is ready to buy
The same shipping and returns questions keep coming	"Draft a Shipping page with prices and delivery times per country and courier."	The chat quotes the page with a link; the shopper doesn't email support
Shoppers ask for a discount on the same product	"Take 15% off everything in Winter until Sunday."	The promotion is in the catalog and the chat shows it with the old price struck through
Conversion after a chat is above the store average	"Calculate orders, revenue and average order value for the last 30 days and compare them with the previous month."	You see whether the chat lifts sales for the whole store without building reports

The same chat questions are the best topics for the rest of your marketing. A question shoppers type into the chat is usually typed into Google too: check it in the Search Console queries report (Google Search Console for an online store) and turn it into a page or an FAQ. "How do I wash this?" or "Which size should I take?" is a ready email for customers who have already bought (email marketing for an online store: 7 automated emails). Running the store through an AI agent saves time: instead of opening ten admin screens, you write one sentence and get a finished change or a table.

Chatbot or live chat with people?

A chatbot is the better choice for repeated questions, answers outside working hours and choosing between many products. Live chat is the better choice for complex cases: a large business order, a complaint, a made-to-order product. For a small store the chatbot is a convenient start, because it answers instantly at any hour with nobody on shift, which also matters when your shoppers sit in several time zones.

	AI chatbot (Aura Chat)	Live chat (Live Chat, Zopim)
When it answers	24/7	When an agent is available
Knows catalog and orders	Yes, from the store's data	The agent checks them by hand in the admin
Setup in CloudCart	Apps → Aura Chat, skills and limits	A LiveChat license key or the Zendesk Chat embed code
Where it shows	Proactive messages on rules; by default not in cart, checkout or account	Live Chat loads on every page, with no exceptions
Handover to a person	None; when needed it gives the store's contact details	The whole conversation is with a person

Aura Chat has no handover to an agent: conversations in the admin are read-only. So put the phone number and email where shoppers can reach you for complex cases into the assistant's instructions. If you want an agent answering live, use the Live Chat or Zopim apps from the CloudCart App Store (connecting them is covered in the guide to live chat in CloudCart); agents, opening hours and styling are set in each service's own admin.

What does EU law say about chatbots?

From 2 August 2026, the EU AI Act requires that people are told they are interacting with an AI system, unless this is obvious to a reasonably attentive person (Regulation (EU) 2024/1689, Article 50). The information has to be clear and given at the latest at the first interaction. The transparency rules apply across the EU from August 2026 (European Commission: AI Act). For a store, that means a clear line in the chat window.

Aura Chat has that line by default: "{name} can make mistakes. This is an assistant powered by artificial intelligence." It can be up to 200 characters and you can change it. Don't delete it: an empty field removes the line completely.

Conversations contain personal data: an email, an order number, sometimes an address. Under the General Data Protection Regulation (GDPR), your store's privacy policy should say that the chat records conversations and how long it keeps them. In Aura Chat, transcripts are deleted 6 months after each message. To show an order, the chat asks for both the order number and the email on the order, so a stranger cannot see someone else's order, and a return request is confirmed with a 6-digit code sent by email.

7 mistakes that stop a chatbot from selling

A chatbot sells poorly most often because of incomplete products and pages or bad settings. These are the common mistakes and how to fix them:

1. Terms live only in the Terms and Conditions. The chat finds pages by title. A return clause buried in a long Terms page is hard to find; a "Returns" page is found on the first try.
2. Products have no descriptions or attributes. The bot cannot compare two products when it only knows the name and price. Fill in size, material and compatibility.
3. Sold-out products are not hidden. Set minimum stock to 1 if you don't want the chat to recommend a product you cannot ship.
4. The chat speaks up on every page after 10 seconds. The starter rule is for testing. Replace it with specific moments: hesitation on a product page, a second look at the same product, many pages viewed with an empty cart.
5. The "Never speak here" list was deleted. It is the only thing that keeps the cart, checkout and account clear of interruptions. By default it holds /checkout, /cart* and /account*; leave them there.
6. Discounts without a reason. A code given to everyone who asks is a discount for everyone. Describe specific reasons (hesitation over price, a delayed delivery) and a cap, and for abandoned carts use the dedicated tools from the article on abandoned carts.
7. The window covers the Buy button on mobile. Open your store on a phone and check where the chat sits against the cart button; move it to the other side if needed. What else to check is in the article on your store's mobile version.

A full overview of the AI tools in the store, from search with Aura Search to texts and photos, is in AI for ecommerce: 12 tasks AI can do in your online store.

Frequently asked questions

Create a store for €1, install Aura Chat and switch on the first two skills this week. Questions? Write to support@cloudcart.com.