Training Overview
Your chatbot is only as good as its training data. Learn how to create a comprehensive knowledge base that enables accurate, helpful responses.
How AI Training Works
Unlike traditional chatbots that match keywords, Chatmefy uses advanced AI to truly understand your content and generate contextual responses.
1. Content Ingestion
Upload documents, crawl websites, or add Q&A pairs
2. AI Processing
Content is chunked, embedded, and indexed for semantic search
3. Response Generation
AI finds relevant content and generates natural responses
Privacy note: replies are generated with Google Gemini on Vertex AI, in a Google Cloud project used only for Chatmefy.
Training Sources
Document Upload
Upload PDFs, Word docs, text files, and more
DOCX
TXT
CSV
Website Crawling
Automatically import content from your website
HTML
Blog posts
Help articles
Custom Q&A
Manually add question and answer pairs
Direct input
CSV import
Bulk upload
Conversation Learning
Learn from real customer interactions
Chat history
Corrections
Feedback
Training Quality Indicators
Monitor your knowledge base health in the dashboard:
| Indicator | What It Means | Target |
|---|---|---|
| Knowledge Coverage | % of questions your bot can answer confidently | >85% |
| Response Accuracy | Based on user feedback (thumbs up/down) | >90% |
| Content Freshness | How recently content was updated | <30 days |
| Fallback Rate | % of questions that trigger fallback | <15% |
Getting Started with Training
- 1 Start with your FAQ Add your most common questions and answers first. This covers the majority of visitor inquiries.
- 2 Add product/service documentation Upload PDFs, help docs, or crawl your documentation site to cover detailed product information.
- 3 Review unanswered questions Check your dashboard weekly for questions the bot couldn't answer, then add the missing information.
- 4 Iterate and improve Training is ongoing. Regularly add new content and refine existing answers based on feedback.
Training Tips
Quality over quantity — well-written content beats large volumes of poor content
Use natural language — write answers how you'd actually speak to customers
Include variations — add different ways customers might ask the same question
Keep it current — outdated information can harm trust more than no information
Be specific — vague answers lead to follow-up questions and frustration
Test regularly — ask your bot questions and refine based on responses
Start Training Your Bot
Choose the training method that works best for your content: