Stick to the Facts
Learn how to write prompts that keep Omnifact focused on real information and prevent it from creating details that aren’t accurate.
The Core Principle
Sometimes AI creates information that sounds believable but isn’t actually true or based on your materials. This includes made-up statistics, fictional details, or confident answers about things it doesn’t actually know. The solution is simple: Always tell Omnifact to stick to the information you provide and say when it doesn’t know something. To provide this information, you can copy and paste text or attach files directly to your message. If cloud storage integrations are enabled, you can import documents from Google Drive, OneDrive, or SharePoint using the attachment menu. In Spaces, the AI can also retrieve from Uploaded Files and Connected Sources. When you use these techniques, you:- Get more reliable answers for important business decisions
- Avoid spreading incorrect information to colleagues or clients
- Build appropriate trust in AI-generated content
“Using only the information in this report, summarize the key findings. If any important details are missing, say so clearly. REPORT: [your document]“
Choosing the Right Response Mode
Omnifact offers three Response Modes that affect how strictly the AI sticks to facts. Use this framework to pick the right mode before you send your message:Auto (Default)
Best for: Questions that may need web search, attached files, or document retrieval alongside your instructions.The AI decides which tools to use. Pair with grounding phrases when accuracy matters.
Quick Response
Best for: Brainstorming, drafting, or general knowledge when speed matters more than external sources.The AI answers from its training data only. Do not use this mode when you need facts from your documents or the web.
Simple Lookup
Best for: Answers that must come only from your Space’s documents.Available in Spaces with Uploaded Files. The most reliable option for preventing made-up details.
| If you need… | Choose… |
|---|---|
| Up-to-date information or multiple sources | Auto |
| A fast draft without tool lookups | Quick Response |
| Answers strictly from your team’s documents | Simple Lookup |
Using Simple Lookup Mode in Spaces
If you are working within a Space that has Uploaded Files, Omnifact offers a built-in feature to guarantee responses are completely grounded in your documents: Simple Lookup mode. When you switch your Response Mode to Simple Lookup (using the selector at the bottom left of the chat input), the AI assistant will:- Search only through the Space’s Uploaded Files.
- Ignore its general training data.
- Base its entire response strictly on the retrieved document content.
Helpful Phrases to Keep Responses Accurate
Use these phrases to keep Omnifact focused on facts:Set Clear Boundaries
- “Using only the information provided…”
- “Based solely on this document…”
- “If the answer isn’t in the content, say so”
- “Don’t guess beyond what’s written”
Encourage Honesty
- “If you’re not certain, say so”
- “Only answer what you can verify”
- “Tell me when information is missing”
- “If unclear, ask for clarification”
Real Examples You Can Use
- Market Research
- Contract Review
- Financial Data
- Customer Feedback
“What are the latest trends in our industry and how do they affect our strategy?”
“Based on the market research I’m providing, identify industry trends mentioned and their potential impact on our strategy. If you need additional recent information not in these documents, let me know.
RESEARCH DOCUMENTS: [your research]”
“Based on the market research I’m providing, identify industry trends mentioned and their potential impact on our strategy. If you need additional recent information not in these documents, let me know.
RESEARCH DOCUMENTS: [your research]”
Common Mistakes That Lead to Made-Up Information
Asking for Information You Haven't Provided
Asking for Information You Haven't Provided
Problem: Requesting details that aren’t in your materials
Instead of: “What’s our customer satisfaction score?” (when you haven’t shared survey data)
Try: “What does this customer survey data tell us about satisfaction levels?”
Instead of: “What’s our customer satisfaction score?” (when you haven’t shared survey data)
Try: “What does this customer survey data tell us about satisfaction levels?”
Asking for Predictions Without Historical Data
Asking for Predictions Without Historical Data
Problem: Asking for forecasts without past information
Instead of: “What will our sales be next quarter?”
Try: “Based on these past 4 quarters of sales data, what trends do you see?”
Instead of: “What will our sales be next quarter?”
Try: “Based on these past 4 quarters of sales data, what trends do you see?”
Asking for General Best Practices
Asking for General Best Practices
Problem: Asking for advice without your specific context
Instead of: “What are the best marketing strategies for SaaS companies?”
Try: “Based on our company profile and these market research documents, which marketing approaches might work best for us?”
Instead of: “What are the best marketing strategies for SaaS companies?”
Try: “Based on our company profile and these market research documents, which marketing approaches might work best for us?”
When It’s Still Not Working
Omnifact still makes assumptions
Omnifact still makes assumptions
Be more explicit about boundaries:
- Add “Don’t assume standard practices not mentioned”
- Use “If this information isn’t provided, state that clearly”
- Try “Limit your analysis to only what’s documented”
Responses are too vague
Responses are too vague
Ask Omnifact to cite sources:
- “For each point, tell me which section of the document it comes from”
- “Quote relevant sections to support your analysis”
- “Point to specific data from the materials provided”
Missing important warnings
Missing important warnings
Explicitly ask for limitations:
- “Clearly state any assumptions you’re making”
- “Tell me what additional information would be needed”
- “Point out any sections that are unclear or could be interpreted differently”
When to Be Extra Careful
Watch out for these situations where made-up information is more likely:- Requests for specific statistics without providing data sources
- Questions about recent events or current market conditions
- Requests for industry benchmarks without providing reference materials
- Legal or compliance questions without specific documentation
- Technical specifications not provided in your materials
Other Intermediate Guides
- Review Separating Data and Instructions to structure your prompts clearly
- Learn Thinking Step by Step to guide AI reasoning processes
- Practice Using Examples to show exactly what you want
- Return to the Beginner Level to review the basics.