AI chatbots

What are AI chatbots?

AI chatbots in WebGPT are dedicated AI experts built on your own documents and data. Unlike generic AI models, they know your business — your products, your policies, your data. When someone asks a question, the chatbot searches your knowledge base for the most relevant information, then generates a natural, accurate answer grounded in that data.

This makes AI chatbots ideal for:

• Website AI expert — Embed on your site to answer visitor questions 24/7, using your own data
• Internal team assistant — Let your team find answers across company documents instantly
• Document analysis — Analyze legal contracts, financial reports, medical records, or any specialized documents with an AI that knows the context
• Content creation — Use your chatbot as the AI source when creating articles or running bots — every piece of content informed by your proprietary knowledge
• Sales and support — Help visitors learn about products, get pricing info, or resolve issues

Knowledge base

The knowledge base is the foundation of your chatbot — it's where you upload the documents and information your chatbot will draw from when answering questions. The better your knowledge base, the more accurate and helpful your chatbot's responses will be.

Supported file formats

Uploading and processing documents

Navigate to the Knowledge base section to manage your content. To upload a new document:

1. Upload the file
   Drag and drop your file onto the upload area, or click to browse your computer. Supported formats are listed above.
2. Enter a title
   Give the document a descriptive title. This helps you identify and organize documents in the knowledge base list.
3. Configure chunking settings
   Your document is split into smaller pieces called "chunks" for efficient AI retrieval. Configure two important settings:
   • Chunk size (200–2000 tokens, recommended: 800) — How large each chunk is. Smaller chunks give more precise retrieval but may lose context. Larger chunks preserve more context but may include irrelevant information.
   • Overlap (0–500 tokens, recommended: 150) — How many tokens overlap between adjacent chunks. Overlap ensures that information that falls on a boundary between chunks is not lost.
4. Choose an embedding model
   Select the model used to convert your text into mathematical vectors (embeddings) for similarity search:

   Model	Dimensions	Cost	Best for
   text-embedding-3-large	3072	$0.13 per 1M tokens	Higher accuracy, important documents
   text-embedding-3-small	1536	$0.02 per 1M tokens	Lower cost, general purpose

5. Review cost estimate
   Before processing, WebGPT shows a real-time cost estimate based on your document size, chunk settings, and chosen embedding model. Review this to understand the processing cost.
6. Process the document
   Click Process to start. The system extracts text from your document, splits it into chunks, generates embeddings for each chunk, and stores everything for fast retrieval.

Organizing with tags

Assign tags to your documents to keep the knowledge base organized. Tags are especially useful when you have many documents and want specific chatbots to access only certain subsets. You can filter documents by tag when configuring a chatbot's knowledge source.

Document status

• Active / Inactive — Toggle whether a document is available for chatbot retrieval. Inactive documents remain in your knowledge base but are not searched during conversations.
• Indexed — Indicates whether the document has been processed and its embeddings are ready for use.

Deleting documents

When you delete a knowledge base document, WebGPT shows you exactly which chatbots use it and how many conversations would be affected — so you can make an informed decision before removing anything. If a deleted document was the only knowledge source for a chatbot, that chatbot is automatically removed to prevent broken experiences.

Creating a chatbot

Once your knowledge base has at least one document, you're ready to create a chatbot. Navigate to AI → Chatbots and click Add chatbot.

Basic settings

1. Name and status
   Give your chatbot a name (e.g., "Customer Support Bot" or "Product Knowledge Assistant") and set its status to active or inactive.
2. AI model selection
   Choose the AI model that powers the chatbot's responses. Available options include OpenAI, Google Gemini, Anthropic Claude, DeepSeek, Grok, and Mistral AI. Different models have different strengths — GPT-5.2 and Claude Opus are excellent for nuanced responses, while faster models like Gemini Flash are better for high-volume chatbots.

Instructions

After saving your chatbot settings, you'll be guided to the Instructions tab. This is where you define your chatbot's personality, role, and behavior rules. You have two options:

Guided setup (recommended)

The built-in wizard guides you through creating optimized instructions in 5 simple steps — no AI expertise required. It works for any use case: customer support, document analysis, research, education, content creation, and more.

1. Purpose
   Choose what your chatbot will specialize in. Browse six categories — Customer-Facing, Industry Expert, Internal Tool, Document Expert, Educator, and Content Creator — or define a custom purpose. The wizard automatically adjusts all settings to match your choice.
2. Identity & tone
   Enter the name, a short description, and choose a tone of voice. You can select a Writing Style for consistent brand voice, pick up to 3 tone tags, or write custom tone instructions. The fields adapt to your chosen purpose — a customer support chatbot asks for your organization name, while a research chatbot asks for the subject or topic.
3. Core knowledge
   Select which data source topics the chatbot should know about and optionally add key facts. These are woven into the instructions so the chatbot can suggest relevant topics when it cannot find a specific answer.
4. Behavior rules
   Fine-tune how the chatbot handles different situations: language preferences, response length, formatting style, what to do with off-topic questions, and more. Each option comes with sensible defaults based on your chosen purpose, so you only need to change what matters to you.
5. URLs & preview
   Add important links (website, documentation, contact page, etc.) and preview the fully assembled instructions.

After completing the wizard, you'll see a summary of your settings and the generated instructions. All wizard settings are saved and can be edited anytime by clicking Edit Settings.

Custom instructions

Below the wizard-generated instructions, a Custom Instructions field lets you add specific rules and guidelines. These are appended to the generated instructions and won't be affected when you edit wizard settings — perfect for adding business-specific rules like "Always mention the free trial" or "Never discuss competitor pricing."

Write manually

Prefer full control? Choose Write Manually to write your own instructions from scratch. You can switch between guided and manual modes at any time.

Knowledge base configuration

This is where you connect your chatbot to its knowledge base and fine-tune how it searches for information:

• Select knowledge base documents — Choose which documents from your knowledge base the chatbot can access. You can select individual documents or filter by tags to include an entire group.
• Retrieval K (1–20, default: 5) — The number of document chunks retrieved for each question. Higher values give the AI more context to work with but increase processing time and cost. For most chatbots, 3–7 is a good range.
• Similarity threshold (0–1, default: 0.7) — The minimum similarity score a document chunk must have to be included in the context. Higher thresholds (e.g., 0.8) return only very relevant chunks. Lower thresholds (e.g., 0.5) return more results but may include less relevant information.
• Context window (0–3, default: 1) — When the chatbot finds a relevant chunk, it can also include surrounding chunks for additional context. A value of 1 means it includes 1 chunk before and 1 chunk after each match, helping the AI understand the full picture. Set to 0 to use only the exact matching chunks.
• Include citations — Toggle whether the chatbot's responses include references to the source documents. When enabled, users can see which documents the answer came from, which builds trust and allows verification.

Advanced AI settings

• Temperature (0–2, default: 0.7) — Controls the creativity and randomness of responses. Lower values make the chatbot more focused and factual (good for support bots). Higher values make it more creative and conversational.
• Max tokens (0–16000, default: 6000) — The maximum length of a single response. For customer support, 2000–4000 tokens is usually sufficient. For detailed technical explanations, you might need 6000+.

Notes

Add internal notes to your chatbot for your own reference. These are only visible to you in the dashboard and are never shown to chatbot users.

Widget design and customization

Make the chatbot feel like a natural part of your website. WebGPT gives you full control over the widget's appearance and behavior, so visitors get a seamless experience that matches your brand.

Position and layout

• Text direction — Choose LTR (left-to-right) or RTL (right-to-left) to match your website's language
• Corner position — Place the chat widget button in any corner of the screen
• Spacing — Adjust the distance of the widget from the screen edges

Branding

• Color picker — Choose a primary color for the chatbot widget that matches your brand
• Auto-detect brand colors — Enter any website URL and WebGPT will analyze it to extract the primary brand colors. This is a quick way to match the chatbot to your existing site design.

Text customization

Customize every piece of text displayed in the widget:

• Toggle button text — The text shown on the floating chat button (e.g., "Chat with us")
• Chat title — The header text when the chat window is open (e.g., "Customer Support")
• Leave chat text — The text shown when a user ends the conversation
• Input placeholder — Placeholder text in the message input field (e.g., "Type your question...")
• Welcome message — The first message displayed when a user opens the chat (e.g., "Hi! How can I help you today?")

Leave details form

Optionally collect visitor contact information:

• Enable/disable — Toggle the leave details form on or off
• Form title — The heading text for the contact form
• Thank you message — Message displayed after the user submits their details
• Name and phone labels — Customize the field labels to match your needs

Conversation starters

Add clickable prompt suggestions that appear when a user first opens the chat. These help guide visitors toward common questions and demonstrate the chatbot's capabilities. For example:

• "What are your pricing plans?"
• "How do I get started?"
• "What are your support hours?"

Users can click any starter to send it as their first message, or type their own question instead.

Additional settings

• File uploads — Allow users to upload files (.txt, .doc, .docx, .rtf, .md, .pdf, .xlsx) during the conversation for the AI to analyze
• Voice recording — Enable voice input so users can record audio messages (transcribed to text by the AI)
• Email notifications — Receive email alerts when users start new conversations or leave their contact details
• Message limits — Set a maximum number of messages per conversation to control costs
• Allowed domains — Restrict which websites can embed and use the chatbot widget. This prevents unauthorized use of your chatbot on other sites.

Embedding the chatbot on your website

Once your chatbot is configured and active, you can embed it on any website:

1. Copy the embed code
   Open the chatbot's Embed code tab (via the actions menu on any chatbot card, or from the third tab when editing a chatbot). Click the copy button to copy the JavaScript snippet to your clipboard.
2. Paste into your website
   Add the copied code snippet just before the closing </body> tag on any page where you want the chatbot to appear. Most website builders and CMS platforms have an option to add custom HTML/JS.
3. Save and test
   Save the changes to your website and refresh the page. You should see the chat widget button appear in the corner you configured. Click it to open the chat and test it with a few questions.

Chat interface features

The chatbot's chat interface provides a rich, interactive experience for users:

• Real-time streaming — Responses appear word by word as the AI generates them, giving users immediate feedback.
• Markdown rendering — The AI's responses are rendered with proper formatting including headings, bold text, lists, code blocks, and links.
• Citations — When enabled, source documents are referenced in responses so users can verify the information.
• Message rating — Users can rate individual responses on a 1–5 star scale. This feedback helps you evaluate and improve your chatbot's performance.
• Token usage and cost tracking — Each conversation tracks the number of tokens used and the associated cost, visible in your dashboard.
• Stop generation — Click the stop button during a response to halt generation. Any content already generated is kept.
• Save and download — Use the Save dropdown below any response to save it as a draft article for publishing, or download it as a formatted DOCX file.
• File upload — If enabled, users can upload documents for the AI to read and discuss.
• Enter-to-send toggle — Users can choose whether pressing Enter sends the message or creates a new line.
• Conversation starters — If configured, clickable prompt suggestions appear when opening a new conversation to help get started quickly.

Conversation history

Every chatbot conversation is automatically saved and accessible from the Conversations tab. Each conversation card shows useful context at a glance: the chatbot name, visitor's domain, country, device type, message count, and total cost — so you can quickly understand who is using your chatbot and how.

For each conversation, you can:

• View full transcript — Read the entire conversation including the chatbot's responses, source citations, and per-message costs
• Edit topic — Click the topic to rename it inline for easy reference
• View contact details — If the visitor submitted their information via the leave details form, their name and phone number are linked directly from the conversation card
• Export as JSON — Download the complete conversation data in JSON format for analysis or integration with other tools. Also available from the chat interface menu.
• Export as DOCX — Download the conversation as a formatted Word document for sharing or archival. Also available from the chat interface menu.
• Delete — Remove a conversation from your history

Tags management for knowledge base

Use tags to organize your knowledge base documents into logical groups. Navigate to the tags management section to create, rename, or delete tags. Common tagging strategies include:

• By topic — "pricing", "technical", "onboarding", "policies"
• By product — "product-a", "product-b", "general"
• By department — "support", "sales", "engineering"
• By language — "english", "spanish", "hebrew"

When creating a chatbot, you can filter the knowledge base by tags to give each chatbot access to only the documents it needs. For example, a sales chatbot might only use documents tagged with "pricing" and "product-a", while a support chatbot uses "technical" and "onboarding".