ICE T Framework

Overview

The ICE T framework is a specialized prompting structure designed for creating reusable templates that can be applied across similar tasks. Unlike ROSIE which is optimized for one-off interactions, ICE T focuses on scalability and consistency for repetitive workflows. ICE T stands for Instructions, Context, Examples, and Task.

Framework Structure

ICE T is built around creating modular prompts where most elements remain consistent, and only the task-specific data changes:

I - Instructions

Define the core AI behavior and process that will remain consistent across all uses of this template. This is the reusable "engine" of your prompt.

Example: "You are analyzing blog posts and creating engaging summaries for social media sharing."

C - Context

Provide the background information that stays the same for all instances of this task type. Include your business details, industry context, and any persistent guidelines.

Example: "We are a digital marketing agency specializing in content strategy for B2B tech companies."

E - Examples

Show sample inputs and outputs to demonstrate the expected quality, style, and format. These examples train the AI on your standards.

Example: Provide a blog post input with its corresponding social media summary output.

T - Task

This is the variable component containing the specific data for each individual task instance. This changes each time you use the template.

Example: "Title: The Future of Remote Work, Key Points: Statistics on remote work growth, benefits for productivity, challenges with collaboration"

Example Implementation

Here's a complete ICE T framework for generating social media summaries from blog posts:

You are analyzing blog posts and creating engaging summaries for social media sharing.

We are a digital marketing agency specializing in content strategy for B2B tech companies. Our summaries should be concise (under 200 characters), engaging, and include a call-to-action. Focus on the key insights and practical value for our audience.

For example:
<ExampleInput>
Title: The Rise of AI in Customer Service
Key Points: 70% of customers prefer AI chatbots, reduces response time by 50%, challenges with complex queries, best practices for implementation
</ExampleInput>

<ExampleOutput>
🚀 AI is transforming customer service! 70% of customers now prefer AI chatbots, cutting response times by 50%. Learn best practices for implementation and overcome complex query challenges. Read our latest insights! #AI #CustomerService
</ExampleOutput>

Respond with just the social media summary.

Title: The Future of Remote Work
Key Points: Statistics on remote work growth, benefits for productivity, challenges with collaboration, tools for success

Another Example: Job Candidate Evaluation

You are evaluating job candidates for software development positions to provide structured feedback.

We are a growing SaaS company focused on collaboration tools. We prioritize technical skills, cultural fit, and growth potential. When evaluating candidates, consider experience level, technical proficiency, communication skills, and alignment with our company values.

For example:
<ExampleInput>
Position: Senior Frontend Developer
Candidate: Sarah Johnson
Experience: 5 years React, 3 years team leadership
Skills: JavaScript, TypeScript, CSS, Agile methodology
Interview Notes: Strong technical knowledge, good communication, expressed interest in mentoring junior developers
</ExampleInput>

<ExampleOutput>
**Candidate Evaluation: Sarah Johnson - Senior Frontend Developer**

**Overall Rating: Strong Hire (4.5/5)**

**Technical Skills: Excellent**
- 5+ years React experience with modern best practices
- Solid TypeScript and CSS proficiency
- Familiar with Agile development processes

**Cultural Fit: Very Good**
- Clear communication skills demonstrated in interview
- Interest in mentoring aligns with our growth culture
- Team player mentality evident

**Growth Potential: High**
- Leadership experience suggests readiness for senior role
- Eager to contribute beyond individual contributions

**Recommendation: Proceed to final round**
</ExampleOutput>

Position: Junior Backend Developer
Candidate: Michael Chen
Experience: 2 years Node.js, recent bootcamp graduate
Skills: JavaScript, SQL, basic API development
Interview Notes: Eager to learn, good problem-solving, limited professional experience

Why It Works for AI

The ICE T framework is particularly effective for AI because it leverages the technology's strengths while mitigating its limitations:

Template Efficiency

  • Reusable Instructions create a stable foundation that the AI can reliably execute
  • Consistent Context ensures the AI maintains proper domain knowledge across uses
  • Variable Tasks allow for flexibility while maintaining quality standards

Quality Control Through Examples

The examples serve as training data that shapes the AI's understanding of your expectations. Each use of the template reinforces the desired output patterns, leading to increasingly consistent results.

Scalability

ICE T enables organizations to create "AI workers" for repetitive tasks. Once the template is perfected, it can be used by multiple team members with minimal training, ensuring consistent outputs across the organization.

Reduced Error Rates

By keeping the core prompt stable and only varying the task data, ICE T minimizes the risk of prompt drift or inconsistent interpretations that can occur with ad-hoc prompting.

When to Use ICE T

ICE T is ideal for:

  • Repetitive tasks that follow similar patterns
  • Team workflows where consistency across users is important
  • High-volume processes like quote analysis, content generation, or data processing
  • Quality-critical outputs where standards must be maintained
  • Scalable automation where the same process will be used repeatedly

Use ICE T when you need to create standardized, reusable AI capabilities that can be deployed across your team or integrated into automated workflows. It's perfect for building "AI assistants" for specific business functions.