Using examples means providing Omnifact with samples of the kind of output you want before asking it to perform the same task. Instead of just describing what you want, you show it.Think of it like training a new employee—you show them examples of well-done work, then ask them to do something similar.This gives you consistent formatting, appropriate tone, and better accuracy.
The Task: Classify customer support emails by priority and type.With Examples:
Copy
Here are examples of how to classify customer support emails:EXAMPLE 1: Email: "My account has been charged twice this month. Please refund the duplicate charge immediately."Classification: HIGH PRIORITY - Billing IssueEXAMPLE 2:Email: "I'd like to know more about your premium features. Do you have a comparison chart?"Classification: LOW PRIORITY - Sales InquiryNow classify this email:"Our entire team can't log in to the system. This is blocking all our work today."
When you need to handle different scenarios, provide examples for each:
Copy
Here are examples of how to respond to different types of customer feedback:POSITIVE FEEDBACK EXAMPLE:Customer: "Love the new feature! Makes my job so much easier."Response: "Thank you for the positive feedback! We're thrilled the new feature is helping your workflow. Please let us know if you have any other suggestions."NEGATIVE FEEDBACK EXAMPLE:Customer: "The interface is confusing and slow. Very disappointed."Response: "Thank you for bringing this to our attention. We understand your frustration and are actively working on interface improvements. I'll connect you with our product team to discuss specific issues."FEATURE REQUEST EXAMPLE:Customer: "Would be great if you could add dark mode to the app."Response: "Great suggestion! Dark mode is actually on our roadmap for the next quarter. I'll add your vote to the request and notify you when it's available."Now respond to this customer feedback:[Customer message]
Here are examples of project status updates with increasing detail:SIMPLE PROJECT:Status: On Track | Timeline: 2 weeks remaining | Blockers: NoneMODERATE PROJECT:Status: Minor Delays | Timeline: 3 weeks remaining (1 week behind) | Blockers: Waiting for client approval on designs | Next Steps: Follow up with client by FridayCOMPLEX PROJECT:Status: At Risk | Timeline: 6 weeks remaining (2 weeks behind) | Blockers: (1) Technical integration issues with third-party API, (2) Key team member out sick, (3) Scope creep from stakeholder requests | Next Steps: (1) Schedule technical review meeting, (2) Identify backup resources, (3) Stakeholder alignment meeting to confirm scopeNow create a status update for this project:[Project details]
Problem: Only providing one example, which may not capture the variation you need Solution: Provide 2-3 examples that show different scenarios or complexity levels Instead of: One example of email classification Try: Examples of high-priority, medium-priority, and low-priority classifications
Inconsistent Examples
Problem: Examples that use different formats or styles Solution: Ensure all examples follow the exact same structure and format Check for: Same headings, same level of detail, same tone across all examples
Examples Too Complex
Problem: Examples that are too detailed or contain irrelevant information Solution: Keep examples focused on the specific task at hand Instead of: Full email with headers, signatures, and conversation history Try: Just the essential content needed for the classification task
Make examples as close to your actual work as possible. If you classify support tickets, use realistic support ticket examples.
Show Edge Cases
Include examples of tricky or borderline cases that Omnifact might encounter in your real work.
Be Consistent
All examples should follow exactly the same format, style, and level of detail you want in the final output.
Test and Refine
Try your few-shot prompt with different inputs and adjust your examples if the results aren’t consistent.
The time you spend creating good examples pays off immediately. Well-crafted examples can eliminate hours of back-and-forth refinement on complex tasks.