Most people have had the experience: you open a website chat, ask a simple question, and get a menu of options that do not include your question. After three loops you close the tab.
Modern AI models have made website chat far more capable. They have not made bad chatbot design go away. This guide covers how to add an AI assistant to your site that customers actually find useful, and that helps your business rather than hiding it.
Decide what the chatbot is for
Before choosing a tool, decide the job. A chatbot that tries to do everything usually does nothing well. The most valuable jobs for a business website are:
- Answering pre-sales questions: pricing ranges, timelines, service areas, what is and is not included.
- Qualifying and capturing leads: understanding what the visitor needs and getting their contact details or a booking.
- Handling routine support: order status, rescheduling, opening hours, returns policy.
- Routing to a human when the question is complex, sensitive or high value.
Pick one or two of these to start. Write down what a successful conversation looks like for each, for example "visitor gets a price range and books a call".
Ground it in your own information
A general AI model knows a lot about the world and nothing about your business. Left alone, it will guess, and a confident wrong answer about your pricing or refund policy is worse than no answer at all.
The fix is grounding. The chatbot should answer from a knowledge base you control:
- Service and product descriptions
- Pricing information you are happy to share publicly
- Policies (delivery, refunds, cancellations)
- Your FAQ and past answers your team gives often
The technique behind this is usually called retrieval-augmented generation, or RAG: the system finds the relevant passages from your content and gives them to the model along with the question. The model then answers from that material rather than from memory. Just as important, you instruct it to say "I don't know, let me connect you with the team" when the answer is not in the material.
Design the conversation, not just the model
The AI model is one component. The experience around it matters as much.
Open with a clear promise
"Ask me about pricing, availability or how we work" sets expectations far better than "Hi! How can I help you today?". Visitors immediately know what they can ask.
Keep answers short
Chat windows are small. Two or three sentences, with a link to the full page if needed, beats a wall of text. Instruct the model to be brief and to ask one question at a time.
Make the human handoff obvious
There should always be a visible way to reach a person. When the bot hands off, it should pass along the conversation so far. Making a customer repeat themselves is the fastest way to undo everything the chatbot achieved.
Capture leads naturally
Do not demand an email address before answering anything. Answer first, build trust, then offer: "Want me to send this quote breakdown to your email?" or "I can book you a 20-minute call, does Thursday work?".
Common mistakes
- Auto-opening the chat on page load. It interrupts visitors who were reading. A small, visible launcher is enough.
- Hiding contact details behind the bot. Your phone number and email should still be on the page.
- No guardrails. The bot should not offer discounts, make legal or medical claims, or discuss competitors unless you have decided it should.
- Never reviewing transcripts. The conversation logs are a goldmine. They show what customers actually ask, where the bot fails, and which pages on your site are missing information.
- Treating launch as the finish line. The first version will have gaps. Plan to review and improve it weekly for the first month.
Privacy and data
Customers type surprising things into chat boxes, including phone numbers, addresses and sometimes sensitive personal details. Decide up front what you store, for how long, and who can see it. Tell visitors what happens to their data, use AI providers whose terms do not allow training on your conversations, and follow the data protection rules in the markets you serve.
Off-the-shelf or custom?
For a simple FAQ assistant, an off-the-shelf chat widget connected to your help content can be enough. A custom build makes sense when the chatbot needs to:
- Read live data, such as order status, availability or account details
- Write into your CRM, calendar or booking system
- Follow specific business rules for qualification and routing
- Match your brand voice closely and appear in more than one channel, like the website and WhatsApp
How to measure success
Track the outcomes the chatbot was meant to produce: questions resolved without a human, leads captured, bookings made and handoffs completed. Also track the honest signal: how often visitors ask for a human straight away. If that number is high, the bot is not earning its place on the page yet.
We build AI assistants that answer from your own data and plug into your CRM and booking flow. Tell us what you want your website to handle and we will tell you whether a chatbot is the right answer.