Training Your AI Twin

Your AI Twin’s intelligence comes from the content you provide. This guide covers what content works best, how to customize personality, and how to keep your AI updated.

Content Types

We can process and train your AI on almost any type of digital content.

Audio Content

Podcasts: Public or private podcast episodes
Voice Memos: Recorded thoughts and teachings
Seminar Recordings: Workshop and event audio
Interviews: Conversations and Q&A sessions
Audio Books: Narrated books and courses

Video Content

YouTube Videos: Public channel content with auto-sync
Online Courses: Pre-recorded course modules
Webinars: Live or recorded webinar sessions
Event Recordings: Conference talks and presentations
Vlogs: Personal or business video content

Text Content

Books: Published or unpublished manuscripts
PDFs: Guides, workbooks, manuals
Articles: Blog posts and long-form articles
Whitepapers: Research and thought leadership documents
Emails: Email newsletters and courses
Website Content: Website copy and knowledge bases
Transcripts: Transcribed conversations and teachings

Content Best Practices

Quality Over Quantity

More content isn’t always better. Focus on:

High-quality, authoritative content that represents your best thinking
Well-organized materials with clear structure
Consistent messaging across all content
Updated information that reflects your current perspective

Content Organization

Before sharing content with us, organize it by:

Core Teachings: Fundamental concepts and frameworks
Advanced Material: Specialized or advanced topics
Practical Application: Examples, case studies, exercises
FAQs: Common questions and answers

This helps us prioritize what the AI learns first and ensures the most important content is emphasized.

Personality & Tone Customization

During your onboarding call, we’ll discuss:

Voice & Style

Tone: Professional, casual, inspirational, direct, nurturing, etc.
Vocabulary: Technical jargon vs. simple language
Formality Level: Formal business tone vs. conversational style
Humor: Whether the AI should use humor and what kind

Behavioral Guidelines

Response Length: Short and punchy vs. detailed and thorough
Question Handling: Socratic method vs. direct answers
Proactivity: Reactive (only answers) vs. proactive (guides conversation)
Empathy Level: How much emotional intelligence to display

Knowledge Base Control

You decide the AI’s boundaries and how it handles different scenarios.

Constraint Options

Strict Constraint (Recommended):

AI only answers based on your provided knowledge base
Ensures every answer has a verifiable source
Minimizes hallucination risk
Best for protecting IP and brand accuracy

Adaptive Mode:

AI can tap into wider knowledge when appropriate
Useful for general questions outside your specific domain
Still prioritizes your content first
Requires careful guardrail configuration

Handling Off-Topic Questions

Configure how the AI responds when users ask questions outside the knowledge base:

Option 1: Polite Decline

“That’s outside my area of expertise. I’m specifically trained on [your domain]. Is there something related to [your domain] I can help you with?”

Option 2: Escalation

“That’s a great question, but it’s outside my training. Let me connect you with our team for help with that.”

Option 3: General Knowledge (if Adaptive Mode)

AI provides a general answer while noting it’s not based on your specific content

Guardrails & Boundaries

During onboarding, we set strict guardrails to prevent the AI from:

Discussing off-limit topics (competitors, pricing debates, etc.)
Making promises or commitments on your behalf
Providing advice in regulated areas (legal, medical, financial) unless appropriate
Sharing information not in the knowledge base (if in Strict mode)
Engaging with inappropriate or abusive users

Custom Guardrails

You can specify additional boundaries:

Topics to Avoid: Specific subjects the AI should not discuss
Required Disclaimers: Specific warnings or disclaimers to include
Escalation Triggers: Keywords or scenarios that should escalate to human support
Brand Guidelines: Specific terminology to use or avoid

Keeping Your AI Updated

Automatic Content Syncing

If your content is from a public source we’re connected to, our system will automatically sync with new material:

YouTube Channels: New videos automatically added
Podcast Feeds: New episodes automatically ingested
RSS Feeds: Blog posts and articles automatically processed

Manual Content Updates

For private content, you have two options:

Via Dashboard:

Log in to dashboard.steno.ai
Navigate to Content Management
Upload new files
Content is processed within 24-48 hours

Via Shared Folder:

Add new content to your shared Dropbox/Google Drive folder
Notify our team
We process and integrate the new content

Update Frequency

Recommended Schedule:

Quarterly: For evergreen content businesses
Monthly: For actively creating new content
Weekly: For rapidly evolving topics or frequent content creation

Hallucination Prevention

We use several methods to ensure the AI provides accurate answers:

Source-Based Responses

The primary method is constraining the AI to answer questions only based on your provided knowledge base. This ensures every answer has a verifiable source from your content.

Confidence Scoring

The AI is trained to recognize when it’s uncertain and either:

Ask clarifying questions
Admit uncertainty
Offer to escalate to human support

Regular Quality Checks

Our team periodically reviews conversation logs to identify and correct:

Responses that drift from source material
Inconsistent messaging
Areas where guardrails need adjustment

Testing & Refinement

After receiving your demo AI Twin:

Test Extensively: Try various questions and scenarios
Flag Issues: Use thumbs up/down feedback in the interface
Provide Context: Explain why certain responses should be different
Iterate: We refine based on your feedback

See the Dashboard Access & Testing guide for more details on testing your AI.

Questions about training your AI Twin? Contact support@steno.ai.