How to Build a Custom AI Chatbot for Client Websites Using Free Tools in 2026

This article mentions a tool I use; the link at the end is an affiliate link.

Small business owners are desperate for AI chatbots that actually understand their business, but most can't afford enterprise solutions. If you can build custom chatbots trained on a client's specific data, you can charge $500-2000 per implementation. Here's exactly how to do it using mostly free tools.

Why This Works in 2026

Generic ChatGPT plugins don't cut it anymore. Clients want chatbots that know their product catalog, answer questions about their specific services, and match their brand voice. The technical barrier has dropped enough that you can build these without being a senior developer, but it's still high enough that most business owners won't DIY it.

What You'll Build

A custom chatbot widget that:

• Trains on the client's website content, PDFs, and FAQs
• Embeds directly on their site
• Answers questions using their specific information
• Costs them nearly nothing to run (under $10/month)

Step 1: Gather and Prepare Client Data

Start by collecting everything the chatbot needs to know:

• Scrape their website using a tool like BeautifulSoup (Python) or just manually copy key pages
• Request their FAQ documents, product specs, and pricing sheets
• Get 10-20 common customer questions they receive via email

Convert everything to plain text files. One trick: create a "master document" that combines all this information with clear section headers. This makes the next step much easier.

Step 2: Set Up Your Vector Database

You need somewhere to store this information in a way AI can search it. Use Pinecone's free tier (gives you one index, plenty for starting out).

1. Sign up at Pinecone and create a new index
2. Use OpenAI's embedding API to convert your text chunks into vectors
3. Upload these vectors to Pinecone with metadata tags

Here's the key: chunk your text into 500-1000 character segments with 100 character overlap. This ensures the AI can find relevant context without getting overwhelmed.

# Simple chunking example
def chunk_text(text, chunk_size=800, overlap=100):
    chunks = []
    start = 0
    while start < len(text):
        end = start + chunk_size
        chunks.append(text[start:end])
        start = end - overlap
    return chunks

Step 3: Build the Retrieval Logic

When a user asks a question, you need to:

1. Convert their question to a vector using the same embedding model
2. Query Pinecone for the 3-5 most relevant chunks
3. Send those chunks plus the question to GPT-4 or Claude
4. Return the AI's response

Use LangChain to handle this flow. Their RetrievalQA chain does exactly this in about 10 lines of code:

from langchain.chains import RetrievalQA
from langchain.vectorstores import Pinecone

qa_chain = RetrievalQA.from_chain_type(
    llm=your_llm,
    retriever=vectorstore.as_retriever(search_kwargs={"k": 4})
)

Step 4: Create the Chat Interface

Don't overthink this. Use Streamlit or a simple HTML/JavaScript widget.

For a professional embeddable widget:

• Create a simple Flask or FastAPI backend with one POST endpoint
• Build a minimal chat UI with vanilla JavaScript or React
• Use an iframe to embed it on the client's site

The entire frontend can be under 200 lines. Focus on making it mobile-responsive and matching the client's brand colors.

Step 5: Deploy and Monitor

Deploy your backend to Railway or Render (both have generous free tiers). For clients who need guaranteed uptime, move to their paid tiers at $5-10/month.

Set up basic logging to track:

• Questions the bot couldn't answer well
• Response times
• Daily usage

This data helps you improve the bot and shows the client you're actively managing their solution.

Scaling This Into a Real Side Hustle

Once you've built one chatbot, the process gets much faster. Your second client takes half the time. By your fifth, you've got reusable templates.

Where to find clients:

• Local business Facebook groups
• Cold email to businesses with outdated websites
• Upwork and Fiverr (yes, really—filter for higher-budget projects)

Price it as a one-time setup fee ($500-1500) plus optional monthly maintenance ($50-200). The maintenance is where recurring income happens.

Optional: Using Pre-Built Frameworks

If you want to move faster, some platforms bundle these steps together. Around the time I was scaling my own chatbot service, I tested Perpetual Income 365 which provided templates for client onboarding and email sequences that helped me systematize my outreach. It's not necessary—you can build everything from scratch—but it saved me time on the business side while I focused on the technical implementation.

Common Pitfalls to Avoid

Pitfall 1: Making the bot too conversational. Business owners want accurate answers, not personality.

Pitfall 2: Not setting clear expectations. Tell clients the bot will answer 70-80% of questions perfectly and escalate the rest to humans.

Pitfall 3: Forgetting to update the knowledge base. Schedule quarterly reviews with clients to refresh their data.

Real Numbers From My Experience

I built my first chatbot in January 2026 for a local HVAC company. Took me 12 hours total. Charged $800. They got 340 conversations in the first month and the owner said it saved his receptionist 10+ hours.

Second client was a law firm. Charged $1,200 because they needed more complex intake forms integrated. That one took 8 hours.

By client five, I had it down to 4-6 hours per implementation. That's $100-300/hour if you price it right.

Next Steps

Pick one business you know needs this. Maybe a friend's company or a local shop you frequent. Offer to build it at cost just to get the first one under your belt. Document everything you do. That documentation becomes your process.

The technical skills you'll build—working with embeddings, vector databases, and API integration—are valuable beyond this specific side hustle. You're learning the foundation of modern AI applications.

Start with one chatbot. Get it working. Get paid. Then scale.

---

The tool mentioned above is an affiliate link (disclosed at top): Perpetual Income 365