ICIO Framework
Four components built for data work. ICIO separates what to do from what to do it with — keeping your Instruction, Context, Input Data, and Output Indicator distinct so the AI processes your information exactly the way you need.
Origin: ICIO is a community framework from 2023 that was designed specifically for data processing tasks. Unlike persona-based frameworks that begin with a role assignment, ICIO leads with the Instruction — the task itself. The acronym stands for Instruction, Context, Input Data, and Output Indicator. Its key innovation is the explicit separation of Input Data from the instruction and context, making it particularly effective for tasks involving data transformation, analysis, and structured output generation.
Modern LLM Status: ICIO remains especially relevant for data-centric AI workflows. As organizations increasingly use LLMs to process, transform, and analyze data, the need to clearly separate “what data to work with” from “what to do with it” becomes critical. Modern LLMs handle data better when the input is explicitly demarcated from the instructions — reducing the risk of the model treating your data as part of the prompt instructions rather than as material to process. ICIO’s structure naturally enforces this separation.
Separate the Data from the Directive
When you ask an AI to process data, two things can go wrong. First, the model might confuse your data with your instructions — treating a column header as a command or interpreting data values as context. Second, the model might produce output in an unexpected format because you never specified what the result should look like.
ICIO solves both problems by design. The framework creates four distinct zones in your prompt: the Instruction (what operation to perform), the Context (why and under what conditions), the Input Data (the actual material to process), and the Output Indicator (the exact format and structure of the result). This separation prevents cross-contamination between instructions and data.
Think of it like a factory workflow: the work order (Instruction) is separate from the manufacturing context (Context), the raw materials (Input Data), and the product specification (Output Indicator). Each component serves a distinct purpose, and keeping them separate prevents errors.
LLMs process prompts as a single stream of text. Without clear boundaries, the model cannot always distinguish between “this is an instruction” and “this is data to process.” ICIO’s explicit Input Data component creates a clear signal: everything in this section is material to be processed, not instructions to follow. This is especially critical when working with user-submitted content, CSV data, log files, or any input that might contain text the model could misinterpret as directives.
The ICIO Process
Four components that structure data-centric AI tasks
Instruction
Define the operation to perform. Unlike role-first frameworks, ICIO leads with the task itself. Use a clear, specific verb: “classify,” “extract,” “transform,” “summarize,” “translate,” “reformat.” The instruction tells the AI what kind of processing to apply to the input data.
“Extract all email addresses and phone numbers from the following customer support tickets and categorize them by issue type.”
Context
Provide background information that shapes how the instruction should be executed. Context includes business rules, domain-specific conventions, handling exceptions, and any assumptions the AI should make. This is the “why” and “under what conditions” behind the instruction.
“These tickets are from our European customer base. Phone numbers may use international format with country codes. Some tickets contain multiple issues. Prioritize the primary issue for categorization. Our issue categories are: billing, technical, account access, and general inquiry.”
Input Data
Provide the actual data to be processed. This is ICIO’s distinguishing component — a clearly demarcated section that contains the raw material the AI should work with. By separating input data from instructions and context, you prevent the model from confusing data content with operational directives.
“Ticket #1: Hi, I cannot log into my account since yesterday. My email is user@example.com and my phone is +44 20 7946 0958. Ticket #2: I was charged twice for my subscription. Contact me at billing@example.com or +33 1 42 68 53 00. Ticket #3: The app crashes when I try to export reports. Reach me at tech@example.com.”
Output Indicator
Specify exactly what the result should look like. Define the format (table, JSON, CSV, bullet list), structure (column names, nesting), and any quality criteria. The Output Indicator eliminates guesswork about how the processed data should be presented.
“Return a markdown table with columns: Ticket ID, Email, Phone, Issue Category, Confidence Level (High/Medium/Low). Sort by ticket ID. If no phone number is found, enter ‘N/A’ in that field.”
See the Difference
How separating data from instructions produces precise, structured results
Mixed Prompt
Here is some sales data: Q1 $45K, Q2 $52K, Q3 $38K, Q4 $67K. Can you analyze this and tell me about the trends? Also our target was $50K per quarter.
A narrative paragraph mixing observations about growth with general business advice, no clear structure, and no distinction between trend analysis and target comparison. The format is unpredictable and varies with each attempt.
ICIO Prompt
Instruction: Analyze quarterly sales performance against target and identify trends.
Context: Our quarterly sales target is $50K. We need to identify which quarters underperformed and whether there is a growth or decline trend.
Input Data: Q1: $45,000 | Q2: $52,000 | Q3: $38,000 | Q4: $67,000
Output Indicator: A table showing Quarter, Revenue, Target, Variance ($ and %), and a Status column (Above/Below Target). Follow with a 3-sentence trend summary.
A clean table with all five columns, accurate variance calculations, and a concise three-sentence trend analysis. Consistent format every time. Always verify AI-calculated financial figures against your actual accounting data.
Natural Language Works Too
While structured frameworks and contextual labels are powerful tools, LLMs are exceptionally good at understanding natural language. As long as your prompt contains the actual contextual information needed to create, answer, or deliver the response you’re looking for — the who, what, why, and constraints — the AI can produce complete and accurate results whether you use a formal framework or plain conversational language. But even in 2026, with the best prompts, verifying AI output is always a necessary step.
ICIO in Action
See how data separation produces precise, structured outputs
Instruction: Categorize these employee satisfaction survey responses by sentiment (positive, neutral, negative) and identify the top three themes.
Context: These are anonymous responses from our Q4 2025 pulse survey. The company recently went through a reorganization. We need to present findings to the leadership team with actionable themes, not just raw sentiment counts.
Input Data: Response 1: “Love the new team structure, my manager is great.” Response 2: “Workload has increased significantly since the reorg, no additional support.” Response 3: “Uncertain about my career path after the changes.” Response 4: “Collaboration between teams has improved a lot.” Response 5: “Too many meetings, not enough time for actual work.” Response 6: “The new tools are helpful but training was insufficient.”
Output Indicator: A table with columns: Response ID, Sentiment, Primary Theme. Below the table, list the top 3 themes with frequency count and a one-sentence recommendation for each.
The Input Data is clearly separated from the analysis instructions, preventing the AI from treating survey content as operational context. The Context about the reorganization helps the AI interpret responses accurately (e.g., “the changes” refers to the reorg). The Output Indicator demands both structured data (table) and actionable insights (recommendations) — exactly what a leadership presentation needs. Always have HR professionals review AI-analyzed survey data to ensure sentiment classifications are accurate and recommendations are appropriate.
Instruction: Analyze these application log entries to identify the root cause of the intermittent 500 errors our users are reporting.
Context: We run a Python Flask application on AWS ECS. The errors started appearing after our latest deployment on Monday. The application connects to a PostgreSQL RDS instance. We suspect either a database connection issue or a new code regression.
Input Data: [2025-12-15 14:23:01] ERROR: ConnectionPool exhausted, max_connections=20, active=20, waiting=15. [2025-12-15 14:23:02] ERROR: Request timeout after 30s for /api/users/search. [2025-12-15 14:25:15] WARN: Slow query detected: SELECT * FROM users WHERE last_login > ... (12.4s). [2025-12-15 14:25:16] ERROR: ConnectionPool exhausted, max_connections=20, active=20, waiting=23. [2025-12-15 14:30:00] INFO: Health check passed. [2025-12-15 14:32:44] ERROR: 500 Internal Server Error on /api/users/search.
Output Indicator: Return: (1) Root cause summary in 2-3 sentences, (2) Evidence chain linking specific log entries to the diagnosis, (3) Three prioritized fix recommendations with urgency level (critical/high/medium).
Log data is text-heavy and contains patterns that could confuse an AI if mixed with instructions. ICIO’s Input Data section keeps the logs cleanly separated. The Context provides the deployment timeline and architecture details needed for accurate diagnosis. The Output Indicator structures the response as a diagnostic report with evidence chain — not just a guess. Always have your engineering team verify AI-suggested root causes against actual system behavior before implementing fixes.
Instruction: Transform these raw meeting notes into structured action items with assigned owners and deadlines.
Context: This was a product roadmap meeting with engineering, design, and product teams. The quarter ends March 31. Any unassigned items should be flagged for the next meeting. Team leads are: Sarah (Engineering), Mike (Design), Lisa (Product).
Input Data: “Sarah mentioned the API refactor needs to happen before the mobile launch. Mike said the new onboarding screens are 80% done, needs final review from Lisa. We agreed to push the analytics dashboard to Q2. Lisa wants user testing for the checkout flow done by mid-February. Someone needs to update the public roadmap page. The performance benchmarks Sarah ran show we need to optimize the search endpoint before launch.”
Output Indicator: A numbered list of action items. Each item should include: Description, Owner (or “Unassigned - flag for next meeting”), Deadline, Priority (P1/P2/P3), and Status (New/In Progress/Blocked).
Raw meeting notes are messy, informal, and full of implicit context. ICIO’s Input Data preserves the notes exactly as they were captured, while the Context provides the organizational structure (team leads, quarter end date) needed to correctly assign owners and set deadlines. The Output Indicator transforms chaotic notes into a clean, actionable format. Without this separation, the AI might miss implicit assignments or invent deadlines. Always review AI-extracted action items with meeting attendees to confirm accuracy.
When to Use ICIO
Best for data processing, transformation, and analysis tasks
Perfect For
Pulling structured information from unstructured text — extracting entities, categorizing content, tagging data, and parsing documents.
Converting data from one format to another — CSV to JSON, meeting notes to action items, raw feedback to structured reports, logs to diagnostic summaries.
Processing multiple items through the same operation — analyzing a list of customer reviews, translating a batch of strings, or scoring a set of applications.
When you need the same output structure every time — ICIO’s Output Indicator ensures reproducible formatting across repeated runs with different input data.
Skip It When
Writing blog posts, brainstorming ideas, or generating narratives — these tasks do not have “input data” to process. Use CO-STAR or RACE for content creation.
When the AI’s persona and domain expertise are the primary driver of quality. ICIO has no Role component — frameworks like RACE or RISEN handle role-based tasks better.
Complex questions requiring decomposition or chain-of-thought reasoning. ICIO is optimized for data operations, not reasoning processes. Use Self-Ask or Chain-of-Thought.
Use Cases
Where ICIO delivers the most value
Financial Data Analysis
Process revenue figures, expense reports, and financial metrics with clear separation between the numbers and the analytical instructions.
Document Processing
Extract structured information from contracts, invoices, resumes, and other documents where the input is unstructured text requiring specific output formatting.
Feedback Classification
Categorize customer reviews, survey responses, and support tickets by sentiment, topic, urgency, and other dimensions with consistent output tables.
Log Analysis
Diagnose application errors, identify patterns in system logs, and extract actionable insights from technical log data with structured diagnostic output.
Translation and Localization
Process text for translation with explicit context about audience, register, and domain terminology — keeping source text cleanly separated from translation instructions.
Data Migration
Transform data between formats and schemas — restructuring CSV exports, normalizing inconsistent fields, and mapping old data models to new ones with exact output specifications.
Where ICIO Fits
ICIO bridges ad-hoc data requests and programmatic processing pipelines
ICIO’s structure makes it ideal for creating reusable prompt templates. Because the Input Data is explicitly separated from everything else, you can create a template with fixed Instruction, Context, and Output Indicator sections, then swap in different Input Data each time. This is the foundation of scalable AI workflows — the same processing logic applied consistently to varying data. If you find yourself writing similar prompts repeatedly with different data, convert your pattern into an ICIO template.
Related Techniques & Frameworks
Explore complementary approaches to structured prompting
Build Your ICIO Prompt
Structure your next data processing prompt with all four ICIO components or find the right framework for your task.