❄️ Winter Sale: 40% OFF AIWU
WINTER_SECRET
Valid until Mar 1st
Train Your Chatbot with Embeddings: Connect Your Knowledge Base - AIWU – AI Plugin for WordPress
Table of Contents
< All Topics

Train Your Chatbot with Embeddings: Connect Your Knowledge Base

You’ve set up embeddings and created a dataset. Now you’ll connect it to your chatbot so it actually uses that knowledge when answering visitor questions. This guide covers the connection step, configuration options, and how to verify the chatbot is pulling from your training data.


Before You Start

You’ll need:

If you haven’t created your dataset yet, start with the Embeddings guide first — this guide picks up where that one ends.


Step 1: Open Your Chatbot Settings

Go to WordPress Admin → AI Copilot → AI ChatBots. Click your chatbot to open its settings.


Step 2: Go to the Knowledge Tab

Click the Knowledge tab (sometimes labeled Training or AI Training).

You’ll see a Datasets section listing all available embeddings datasets from your AI Training module.


Step 3: Enable Your Dataset

Find your dataset in the list and toggle it on. If you have multiple datasets (e.g., one for FAQs, one for products), you can enable all of them simultaneously — the chatbot will search across all enabled datasets when answering questions.

Configure retrieval settings:

Setting Recommended What it does
Results per query 3–5 How many matching document chunks the AI retrieves per question. More = more context, higher token cost.
Similarity threshold 0.7 Minimum relevance score for a result to be included. Lower = more results but potentially less relevant. Higher = fewer but more precise.
Search mode Hybrid Combines semantic (embeddings) and keyword search for best accuracy. Use Semantic-only for natural language questions, Keyword-only for exact term lookups.
💡 Tip: Start with defaults (3 results, 0.7 threshold, Hybrid mode). Adjust only if you notice the chatbot missing relevant answers or pulling irrelevant context.

Step 4: Update Your Chatbot Instructions

Go to the Context tab and add a line to your chatbot instructions that tells the AI how to use the knowledge base:

You are a helpful assistant for [Your Business Name]. When answering questions, use the information from the knowledge base as your primary source. If the answer is in the knowledge base, answer from it directly and accurately. If the answer is not in the knowledge base, say: "I don't have that information — please contact us at [email] for assistance." Never make up information that isn't in the knowledge base.

⚠️ This instruction matters. Without it, the AI may use its general knowledge instead of your training data — which means generic answers and potentially inaccurate information about your business.

Step 5: Save and Test

Click Save Changes. Then test immediately:

  1. Open your site in an incognito window
  2. Ask the chatbot a question that is specifically answered in your training data — for example, a specific product price, a policy detail, or a FAQ answer
  3. The chatbot should respond with the exact information from your dataset

Good test questions:

  • “What is your return policy?” (if you trained on your return policy)
  • “How much does [Product X] cost?” (if you trained on your product catalog)
  • “Do you offer free shipping?” (if you included shipping info in training data)
✅ Getting specific, accurate answers from your business data? Your chatbot is now trained. It knows your products, policies, and FAQs — and will answer visitor questions from that knowledge instead of making things up.

Updating Your Training Data

When your business information changes (new prices, updated policies, new products), update the embeddings:

  1. Go to AI Training → your dataset
  2. Edit or replace the affected documents
  3. Click Re-generate Embeddings for the changed documents
  4. No chatbot reconfiguration needed — it automatically uses the updated knowledge

Common Issues

Problem: “Chatbot still gives generic answers even with dataset enabled.”
Fix: Check that the dataset toggle is on AND saved. Re-read the chatbot instructions — remove any lines that say “use your general knowledge” or override the knowledge base instruction. Try asking a very specific question that’s directly in your training data.

Problem: “Chatbot answers some questions from training but misses others.”
Fix: Lower the similarity threshold to 0.6 to retrieve more potential matches. Also check whether the missed question’s answer is actually in your dataset — search your training data directly.

Problem: “Chatbot mixes training data with hallucinated information.”
Fix: Strengthen the instruction in Step 4 with: “Only use information from the knowledge base. Do not supplement with information from your general training.”


What’s Next


Last verified: AIWU v.4.9.2 · Updated: 2026-02-25

Scroll to Top