Chatbot for your website: a guide for online store owners

A visitor who lands on your store at ten at night with a shipping or sizing question has three options: dig for the answer, send a message and wait until morning, or close the tab. A chatbot for your website gives them a fourth option: ask and get an answer right away. But a chatbot that gives vague answers, or invents a policy you never had, does more harm than good. This guide explains the types of website chatbot, what a store's chatbot needs to do, how to prepare the knowledge it answers from, how to install it, and how to check that it actually helps. It is part of the guide to AI for business.

What is a website chatbot

A website chatbot is a chat window, usually in the corner of the screen, that answers visitors without a person typing. Three very different tools go by that name.

Types of chatbot for websites

Who each type suits:

• A rules bot: three or four fixed questions.
• Live chat: a business with a team watching the screen during working hours.
• A grounded AI chatbot: open questions at any hour, as long as it answers only from your store's information and not from the model's general knowledge. That is the kind this guide focuses on.

Must-have features of an AI website chatbot

When you compare tools, these are the capabilities to check, roughly in order of importance:

• Answer only from your content. Shipping and return policies, FAQs and product details. Ask the vendor exactly where answers come from, and whether you can see and edit that source.
• Admit when it does not know. This is the most important capability and the least talked about. A chatbot that guesses a price or promises a delivery date you never offered commits you to it in front of the customer.
• Point to a person. When there is no answer, the customer needs to know where to go next: a rep in chat, WhatsApp, phone or email.
• Recommend a product with a link. A customer who is unsure wants one specific suggestion, not a list of the whole catalog.
• Reply in the visitor's language. Customers do not always write in the language your site is in.
• Look like part of your site. Color, position, name, avatar and welcome message.
• Show you every conversation. Without reading conversations, you cannot tell what knowledge is missing or where the bot goes wrong.

Pay attention to what a tool does not promise, too. A chatbot that is not connected to your stock and orders cannot tell a customer whether an item is available right now or where their parcel is. An honest tool says so plainly.

Preparing the chatbot's knowledge base

A chatbot is only as good as the content behind it. Most problems store owners blame on AI are really missing, outdated or contradictory information on the site. Before you install:

1. Update your policy pages. Shipping, returns, warranty and payment methods. If one page says "3 business days" and another says "up to a week", the chatbot will pick one.
2. Write a real FAQ page. Take the ten questions you get most by email and messaging, and write a short, clear answer to each.
3. Complete your product descriptions. Dimensions, materials, fit for common uses and what is in the box. What is not written down, the chatbot does not know.
4. Write rules. What it must never promise ("no guaranteed delivery dates"), what always goes to a person ("refunds"), and how to sound ("friendly, plain, no exclamation marks").
5. Fill in contact details. A WhatsApp number or phone the chatbot can point to when it has no answer.

Once the tool has scanned your site, review what it extracted, and switch off or delete whatever should not be there:

• Duplicate products picked up from two pages.
• Badly worded questions from the scan.
• Products you no longer sell.

How to install a chatbot on your website

Most website chatbots install with one line of code that the tool gives you. You paste it into the header or footer area of your site's theme:

• WooCommerce on WordPress: usually through the theme settings, or a plugin that adds code to every page.
• Shopify: add the code in the theme's code editor.
• A site someone else built: it is a small job for whoever built it.

After installing, check three things:

1. The window shows up on mobile without covering the add-to-cart button.
2. It does not appear where it gets in the way, such as checkout, if you do not want it there.
3. It answers your five most common questions correctly.

Welcome message and chat name

The welcome message is what most visitors see, even if they never type a word. "Hi, how can I help?" tells them nothing. A few rules:

• Say what the chat can actually help with, for example "Ask me about shipping, returns and sizing, or ask for a product recommendation". Visitors know what to ask, and their expectations match what the chatbot really knows.
• Use your store's name or a plain label, not a human name. A customer who thinks they are talking to a rep expects things a bot does not do, like checking an order or approving a refund.
• Never ask for sensitive details. No card numbers or ID numbers in chat.

Bubble: StoreChart's chatbot for websites

Bubble is StoreChart's chatbot for your website, and it is built on the principles above:

• Its knowledge comes from a scan of your site, up to about 250 pages, that extracts products, articles and FAQs. You complete it with a business profile, instructions and rules, and every item in the knowledge base can be edited or switched off.
• It replies in the language the visitor writes in, from the same knowledge base.
• It recommends products from the scanned catalog: it asks one question at a time, recommends a product and explains why, with a link to the product page.
• When the information is not there, it says so and points the visitor to your store's WhatsApp or phone, whichever you set in the profile.
• It looks like your site: color, position, name, avatar, welcome message and theme, set per store.
• You see every conversation, and can fill gaps in the knowledge based on what visitors ask.

What's not included in Bubble

Know in advance what it does not do:

• No live stock check, so switch off products that are sold out for a while.
• No handoff to a live rep, and no ticketing.
• No access to orders, and it does not create leads.

If most of your customers message you on WhatsApp, StoreChart also has a WhatsApp AI agent. It is a separate tool with its own knowledge and settings, and it can hand a conversation to your team.

How to check the chatbot helps

In the first weeks, read the conversations, not just the numbers. Each of these findings is a fix in the knowledge base:

• Questions the chatbot could not answer.
• Correct answers that are badly worded.
• Referrals to WhatsApp in cases where the answer is on your site.

After that, track a few simple measures:

• How many conversations end with a referral to a person. If that share does not drop over time, the knowledge is not being updated.
• How many repeat questions still reach your email and messaging on topics the chatbot should cover.
• The conversion rate of visitors who chatted compared with visitors who did not, if the tool or your analytics can separate them.

A website chatbot is one entry point to customer service. How it fits with WhatsApp and a human team is covered in the guide to AI customer service for online stores.