Most guides on building an AI chatbot either assume you have a developer on staff or stop short of the configuration details that make the bot useful. This guide does neither. You will pick a real platform, configure a real system prompt, embed real code on your site, and have a working chatbot answering customer questions by the time you finish reading. No coding background required, though the guide includes code snippets you copy and paste.
What You Need Before Starting
- An account on one of the platforms below (free tier works to start)
- A list of your 10-15 most common customer questions and their correct answers
- Access to your website backend, WordPress admin, or CMS
- About 60 minutes of focused time
The most important item is the second one. An AI chatbot trained on vague placeholder content produces vague answers. A chatbot trained on your specific products, pricing, policies, and FAQs answers real questions correctly. Before you open any platform, write down the 10 questions your team answers most often and the right answer to each.
Choosing Your Platform
| Platform | Best For | Free Tier | Setup Time | Coding Required |
|---|---|---|---|---|
| Chatbase | FAQ bots trained on docs or URLs | 30 msg/month | 15-20 min | No |
| Botpress | Multi-step conversation flows | Yes, limited | 45-60 min | Minimal |
| Flowise | Custom LLM chains, self-hosted | Free (open source) | 60-90 min | Light |
| Voiceflow | Voice and text multi-channel | Trial only | 30-45 min | No |
| OpenAI Assistants API | Fully custom bots in code | Pay per token | 2-4 hours | Yes |
This guide walks through Chatbase for the no-code path and the OpenAI Assistants API for anyone who wants full control. Both produce a working chatbot at the end.
Step 1: Sign Up and Create Your Chatbot (10 minutes)
Go to chatbase.co and create a free account. Click "New Chatbot." You will see three training options: upload files, enter a website URL to crawl, or paste text directly.
Choose "Website" and enter your domain. Chatbase crawls your pages and builds a knowledge base from the content it finds. This works best when your site has a clear FAQ page, an accurate services or products page, and pricing information. If your site is sparse, use the "Text" option and paste your FAQ document directly.
Give your chatbot a name. This appears in the chat widget and sets the tone. "Support Bot" is forgettable. "Ask Riley" or "Maya from [Company Name]" creates a friendlier first impression and tells the visitor there is something useful here worth trying.
Step 2: Write a Strong System Prompt (15 minutes)
The system prompt is the most important configuration in your chatbot. It tells the AI what role to play, what it knows, what tone to use, and what to do when it does not know something. A weak system prompt produces generic, unhelpful responses. A strong one makes the bot sound like your best support agent.
Copy this example for a plumbing company, replace the details with your own, and paste it into your chatbot's Instructions field in Chatbase:
You are a customer support assistant for Meridian Plumbing, a family-owned plumbing service in Austin, TX, serving Travis County since 2008. YOUR ROLE: Answer questions about services and pricing, help customers schedule appointments, and offer emergency call-backs for urgent issues. SERVICES: - Emergency plumbing repairs (24/7 line available) - Drain cleaning and unclogging - Water heater installation and repair (tank and tankless) - Slab leak detection and repair - Toilet, faucet, and fixture repair or replacement - Repiping for older homes PRICING (estimates -- exact quote requires a site visit): - Service call fee: $89, waived when we perform the repair - Drain cleaning: $149-$299 depending on type - Water heater replacement: $800-$2,400 installed - Slab leak detection: $295 assessment fee HOURS: Monday-Saturday 7am-7pm. Emergency line 24/7. PHONE: (512) 555-0198 SERVICE AREA: Austin, Round Rock, Cedar Park, Pflugerville TONE: Friendly, honest, direct. Never say "I cannot help with that." Offer to have a team member call them back. IF YOU DO NOT KNOW: Say "I want to make sure you get the right answer -- can I have one of our team call you back? What is the best number to reach you?" ESCALATE IMMEDIATELY (offer human call-back) if: - Customer mentions flooding or a burst pipe - Customer is clearly distressed - Question involves insurance or liability
This prompt gives the bot the business context, real pricing ranges, service area details, tone rules, and a clear escalation path. It will not invent services that do not exist because the prompt defines exactly what the company offers.
Step 3: Test Before You Publish (10 minutes)
Chatbase includes a test chat window in the dashboard. Run through these scenarios before embedding on your site:
- Ask each of your 10 most common questions and verify the answers are accurate
- Ask about something outside the bot's scope (a service you do not offer) and check it redirects gracefully
- Type a question with a typo or unusual phrasing
- Ask for a specific price to confirm the bot gives ranges and offers a quote
- Simulate an urgent situation ("There is water pouring through my ceiling") and confirm the bot escalates
If any answer is wrong or unhelpful, update the system prompt with the correct information before going live. Do not rush this step.
Step 4: Embed the Chat Widget on Your Website (10 minutes)
In Chatbase, go to Connect, then Embed. You will see a JavaScript snippet. Copy it and paste it into your website before the closing body tag. Here is what it looks like:
<script>
window.chatbaseConfig = {
chatbotId: "your-chatbot-id-here",
initialMessage: "Hi! Ask me anything about our services.",
primaryColor: "#1a6b3c",
position: "bottom-right"
}
</script>
<script
src="https://www.chatbase.co/embed.min.js"
chatbot-id="your-chatbot-id-here"
defer>
</script>
Where to paste it on popular platforms:
- WordPress: Install the "Insert Headers and Footers" plugin and paste the code in the Footer field.
- Shopify: Online Store, then Themes, then Edit Code, then paste into
theme.liquidbefore the closing body tag. - Squarespace: Settings, then Advanced, then Code Injection, paste in Footer Code.
- Webflow: Project Settings, then Custom Code, paste in Footer Code.
Save your changes, visit your live website, and run through the test scenarios one more time. Seeing it work on the real site is important because sometimes configuration behaves differently in the live embed versus the test window.
Step 5: Custom API Build with OpenAI Assistants (Optional, for Developers)
If you want full control with no third-party platform between you and the AI, the OpenAI Assistants API takes about two hours to set up. The core logic in Python is straightforward:
from openai import OpenAI
client = OpenAI(api_key="your-api-key")
# Create the assistant once and save the returned assistant.id
assistant = client.beta.assistants.create(
name="Meridian Plumbing Support",
instructions="Your full system prompt text here",
model="gpt-4o-mini",
tools=[{"type": "file_search"}]
)
# Start a conversation thread per user session
thread = client.beta.threads.create()
# Add the customer message
client.beta.threads.messages.create(
thread_id=thread.id,
role="user",
content="What does drain cleaning cost?"
)
# Get the response
run = client.beta.threads.runs.create_and_poll(
thread_id=thread.id,
assistant_id=assistant.id
)
messages = client.beta.threads.messages.list(thread_id=thread.id)
reply = messages.data[0].content[0].text.value
print(reply)
Wrap this in a Flask or FastAPI endpoint and connect it to a frontend chat widget. GPT-4o mini at $0.00015 per 1,000 input tokens makes this extremely cost-effective even at high volume. A 500-token customer message costs about $0.000075.
After Launch: What to Monitor
Check your Chatbase conversation logs weekly for the first two months. You will find questions the bot answers poorly, topics it deflects unnecessarily, and phrasing it consistently misunderstands. Each one is a specific update to make to the system prompt or training data. The improvement curve is steep in the first four weeks, then levels off once the main gaps are filled.
Track these metrics after the first month: percentage of conversations that end with the user getting a useful answer, percentage that escalate to human follow-up, and average conversation length. If escalation is above 25%, your system prompt likely needs more specific information about your services or policies.
Frequently Asked Questions
How much does it cost to run an AI chatbot on my website?
Chatbase's free tier covers 30 messages per month. Paid plans start at $19/month for 2,000 messages. If you build directly on the OpenAI Assistants API, costs depend on volume: GPT-4o mini at $0.00015 per 1,000 input tokens means a 500-token message costs about $0.000075. For most small businesses handling 500-1,000 conversations per month, direct API costs run well under $5/month.
Will the chatbot give customers incorrect information?
A well-configured bot with a specific system prompt that includes your actual services, pricing, and policies gives accurate answers for the scenarios it was trained on. The risk of wrong information is highest when the system prompt is vague or when customers ask about things not covered in training data. Always include a clear fallback: "If you are unsure, offer to have a team member follow up."
Can the chatbot book appointments or take orders?
Yes, but this requires integration work beyond the basic setup. Chatbase and Botpress both support Calendly integration so the bot can link directly to a booking page. For payment processing or order taking, you need a custom API integration. Start with information and lead capture, then add transactional features once the basic bot is performing well.
What happens when the chatbot cannot answer a question?
With a good system prompt, the bot acknowledges it does not have that information and offers a next step: a callback, a contact form link, or a direct phone number. A bot that says "I cannot help with that" and stops is poorly configured. Define your escalation path explicitly in the system prompt so every dead end becomes a lead capture opportunity.
How is this different from a live chat tool?
Live chat requires a human to be available at the keyboard. An AI chatbot handles questions 24 hours a day, seven days a week, instantly. It covers the 60-70% of questions with known answers, captures contact details from visitors who would otherwise leave, and routes the complex or urgent cases to your human team. The two work best together rather than as alternatives.
Common Mistakes to Avoid
The most common mistake is going live with a vague system prompt and hoping the AI figures it out. It will not. Write out your actual services, actual pricing ranges, actual service area, and actual escalation rules before you publish anything.
The second most common mistake is not updating the bot after launch. Check the conversation logs. Real customer questions reveal gaps that no amount of planning catches in advance. A chatbot that is not being maintained stops being useful within a month as your products and policies change.