What this automation does
This chatbot sits on your website or in your app and answers customer questions by pulling from your existing documentation. Unlike a basic FAQ page, it understands natural language questions and finds the relevant answer even when the customer phrases things differently than your docs.
It uses Retrieval Augmented Generation (RAG) — your documents are indexed, and when a question comes in, the system finds the most relevant sections and feeds them to the AI along with the question. This means the AI only answers based on your actual docs, not made-up information.
Tools you need
- Your documentation: Help docs, FAQ pages, knowledge base articles, PDF guides
- OpenAI API or Claude API: For understanding questions and generating answers
- Pinecone or Supabase: Vector database to store and search your document embeddings
- Chatbot UI: Chatbase, Voiceflow, or a custom widget
How to set it up
Step 1: Gather all your documentation — help articles, FAQ pages, policy documents, product guides. Export them as text or markdown files.
Step 2: Use a platform like Chatbase or CustomGPT to upload your docs. These platforms handle the RAG pipeline for you — they chunk your documents, create embeddings, and set up the retrieval system. No coding needed.
Step 3: Customize the chatbot's system prompt. Set your brand name, tone, and rules — for example, never make promises the docs don't support, always offer to connect with a human for complex issues, and include your support email.
Step 4: Embed the chatbot widget on your website. Test it with 20 real customer questions from your recent support history. Check that it finds the right docs and gives accurate answers.
Step 5: Add a fallback — when the chatbot can't find a relevant answer (low confidence), it should say so and offer to create a support ticket or connect with a human.
Cost breakdown
| Item | Cost | Notes |
|---|---|---|
| Chatbase or similar | $40-$100/mo | Includes hosting and RAG infrastructure |
| OpenAI API (if self-hosted) | $20-$50/mo | Based on conversation volume |
| Setup time | 2-4 hours | One-time, mostly uploading docs |
| Total monthly | $40-$100/mo | Handles 70%+ of repetitive questions |
Frequently asked questions
The chatbot will tell users it doesn't have enough information and offer to connect them with a human. Use the questions it can't answer to identify gaps in your documentation — this is actually a great way to improve your docs.
RAG significantly reduces hallucination because the AI only answers based on retrieved documents, not its general knowledge. Add instructions in your system prompt like 'Only answer based on the provided context. If you're not sure, say so.' Test with adversarial questions to verify.