How to Add an AI agent to Your Website

Why add an AI agent to your website?

Something I notice about almost every small business I talk to: the questions aren't hard. They're just repetitive, and they all arrive at once.

Someone wants to know if you're open on Sunday. Someone else wants to know whether you deliver to their postcode. A third wants to know what your cancellation policy is. You've answered all three of these before, probably this week, probably more than once each.

The cost isn't the time it takes to answer one question. It's that those questions arrive at 9pm on a Tuesday, or at 10am on a Saturday when you're already behind, and they still need an answer. So you either answer them late, or you don't answer them at all, and the customer goes to a competitor who did.

Before you buy anything, though, do a two-week count. Open your inbox and your phone, and tally how many enquiries are the same question over and over. Most owners are surprised by the number, and most of them are surprised in the same direction: somewhere between a third and half of all contact is a handful of questions you could write down once.

That's the part worth automating. Not the complicated stuff. The part where you're typing the same four sentences a week.

Types of website AI agents

There are three broad approaches, and they're worth understanding before you pick one, because most disappointing experiences come from expecting one type to behave like another.

Rule-based bots follow a decision tree. If someone types "hours", show the hours. If they type "book", show the booking link. They're fast, predictable and cheap, and they break the moment someone phrases a question in a way you didn't anticipate. Asking "are you guys open Sunday?" is a different string from "sunday hours" and a naive rule set treats it as gibberish.

Live chat widgets put a human on the other end. Excellent when you've got someone available. Entirely unavailable at 9pm, which is exactly when a fair number of people try to find out whether you exist.

AI agents trained on your own content answer in natural language by reading the material you give them. They handle the phrasing problem that breaks rule-based systems, they work at three in the morning, and the limit of what they can say is the limit of what you wrote down.

For most small businesses the third one is the right default. AI isn't cleverer than a person. It's the right default because it removes the phrasing problem entirely, and because it answers instantly at the exact moment a customer is deciding whether you're worth their time.

The caveat: an AI agent is only as good as the content it was trained on. Garbage in, confident garbage out. The rest of this guide is mostly about fixing that.

If you want the fuller comparison, AI agent vs live chat covers which one to pick for your situation, and The small business owner's guide to AI agents covers the basics if you have never used one before.

What you need before you start

Not much:

• Some content to teach it. A FAQ page, a price list, a policy document, or just the answers you give most often, typed into a document. PDF, Word, Markdown and plain text all work. If you have a website, you can point the bot at your own pages instead.
• A website. Any platform. If you can paste a snippet of HTML somewhere, you can install an AI agent.
• About ten minutes.

That's it. You don't need to write any code, and you don't need to know what an API is.

Step-by-step: adding ChatterEase to your website

Step 1: Create your account

Start for free. No credit card, so you can find out whether it's useful before you pay for anything.

Step 2: Create your first AI agent

Give it a name and pick how it should look. The name and colour show up in the widget, so "Ask about your booking" is more useful to a visitor than "Assistant v2". You can change both later without reinstalling anything.

Step 3: Train it on your content

This is the step that decides whether the bot is useful or a liability, so give it real attention. You have three ways to add content:

Upload documents. Drag in your FAQ, your price list, your terms, your cancellation policy. PDF, Word, Markdown or text, up to 10MB each. If a customer question can be answered from something you've already written down, that document belongs here.

Point it at your website. Give it your URL and it will read the pages it can find, skipping the usual noise like blog and news sections. This is usually the fastest route to a working bot, because your website already contains the answers. The problem is that visitors can't find them without knowing what to search for.

Type the common questions in directly. This sounds obvious and gets skipped most. The twelve questions you answer every week belong in there as explicit question-and-answer pairs, in your own words.

A note on what to leave out: anything internal. Draft pricing, staff notes, unpublished plans. A bot trained on a document it shouldn't have seen will happily tell a customer something you never meant to publish. The training best practices in our docs cover this in more detail.

Step 4: Customise the welcome and fallback messages

The welcome message is the first thing a visitor reads, so make it useful rather than cute. "Ask me about opening hours, booking, or our menu" tells a visitor what the bot is for. "Hi! How can I help you today?" doesn't.

The fallback message matters more than people expect. This is what appears when the bot doesn't know, which will happen, and that's fine. Make it honest and give it a way forward: offer to pass the question to your team, or point somewhere useful. A bot that admits it doesn't know and offers a human reads as competent. One that guesses reads as broken.

Step 5: Copy your embed code

Once the bot exists you'll get a snippet like this:

<script async src="https://chatterease.com/embed/YOUR_AGENT_ID/widget.js"></script>

Replace YOUR_AGENT_ID with the ID from your bot's embed tab. That's the only part that changes.

Note the async attribute. It means the script loads in the background and doesn't hold up the rest of your page. Keep it. Without it, your AI agent can slow down the page it's meant to help.

Step 6: Add it to your website

WordPress. Open your dashboard, go to Appearance, then Widgets or Blocks depending on your theme, and add a Custom HTML block. Paste the snippet in and save. There's a fuller walkthrough in the WordPress integration guide.

Shopify. Online Store, then Themes, then Customize, then Edit code. Open the theme.liquid layout and paste the snippet just before the closing </body> tag. Details in the Shopify integration guide.

Wix. Edit your site, then Add a section, then Custom Embed, and paste it in. The Wix guide has screenshots worth following if the editor looks unfamiliar.

Anything else. Paste it before the closing </body> tag. For a single-page app, put it in your main index.html instead.

Word count on the platform guides matters less than the placement rule: put it near the end of the body, not in the <head>. The HTML integration guide covers this.

Testing your AI agent

Test it the way a customer would, not the way you built it. Open your own site in a private window (incognito, or logged out) and try to get help with something.

Look for three things:

Does it know the things you really get asked? Ask your real top ten questions. Anything it misses is a content gap, not a bug.

Does it stay in its lane? Ask something it shouldn't answer, like a medical question or a refund authorisation. It should decline and offer to pass you to a person.

Does the fallback work? Ask something vague and unhelpful, like "hi". You want the honest "I don't know that yet" path, not an invented answer.

If it does badly, the fix is nearly always more content rather than different software. Add the answers it's missing and it improves immediately.

Tips for getting the most out of your AI agent

Write Q&A overrides for the answers it keeps getting wrong. Overrides are your own authored answers that always win, whatever the bot would have said. The moment you notice a question it handles badly, write the correct answer and pin it. Most bots get noticeably better in the first fortnight from this alone.

Turn on lead capture. Most AI agent conversations are a customer telling you something you want: a booking request, a quote enquiry, a callback. If the bot isn't collecting that, it's answering questions and throwing away the reason people asked them.

Read the analytics weekly, not never. The conversation log shows you what people are asking, which is a free list of what to fix on your site. Plenty of businesses find their AI agent answering the same question hundreds of times, which is a strong hint that the answer belongs on the page itself.

Keep it current. Prices, hours and policies change. If your bot was trained in March and you changed your hours in June, it will confidently tell people the wrong thing until you update the source.

Conclusion

Adding an AI agent to your website is a ten-minute job. The part that takes longer is deciding what it should say, and that part is worth the time. It's the same work as writing a good FAQ page, except it now answers the question at 9pm instead of waiting for you to wake up.

Start with your ten most common questions. Train the bot on those and your own website. Give it an honest fallback. Watch what it gets wrong for a fortnight and fix the top three.

If you want the reasoning rather than the mechanics, start with AI agent vs live chat to work out whether this is even the right tool for you, or the small business guide if AI agents are new to you.

Start for free: no credit card.