#06 Prompting Technique

Instructional Prompts

Give AI a recipe, not just a request.

Overview

What It Is

Instructional prompts break down tasks into explicit, sequential steps that guide AI through complex processes. Instead of asking AI to figure out what needs to be done, you tell it exactly how to do it, step by step.

Think of these as recipes for AI: just as a recipe specifies the exact sequence and method, instructional prompts prescribe how to achieve the result. This level of control produces consistent, predictable outputs — especially useful for standardizing workflows.

The Mechanics

Why It Works

AI processes instructions sequentially, activating patterns for each step before moving to the next. When you provide explicit instructions, you create a structured path through AI's pattern space that reduces ambiguity and increases consistency.

Each step becomes context for the next, building a coherent chain of pattern activations. AI can't plan ahead autonomously, but it excels at following explicit procedures where each action is clearly defined.

Decision Guide

When To Use It

Use When
  • Tasks requiring consistent outputs across multiple runs
  • Complex workflows where sequence matters
  • Quality control situations where deviations are problematic
  • Converting human procedures into AI-executable steps
  • Training others on a specific process or approach
Skip When
  • Simple tasks where instructions would be overkill
  • Creative work where rigid structure limits quality
  • Exploratory analysis where you want AI to find patterns
  • When you're unsure of the best approach yourself
  • Tasks where flexibility matters more than consistency
Copy & Paste

Ready-to-Use Templates

T01 Multi-Step Process Complex tasks with distinct phases
Follow these steps in order to [accomplish goal]: Step 1: [First action] - [Specific detail about how to do this] - [What to include/exclude] Step 2: [Second action] - [Build on Step 1's output] - [Specific requirements] Step 3: [Third action] - [Final transformation or synthesis] - [Quality checks] Provide your final output as [format].
Example

"Follow these steps to analyze this meeting transcript: Step 1: Extract all action items with owners. Step 2: Categorize by urgency (High/Medium/Low). Step 3: Format as a prioritized bullet list with owners and deadlines."

T02 Quality-Controlled Output When output quality and format are critical
Create [output type] following these requirements: Content Instructions: 1. [Content requirement 1] 2. [Content requirement 2] 3. [Content requirement 3] Format Instructions: - [Structural requirement 1] - [Structural requirement 2] Quality Checks: Before submitting, verify: ✓ [Criterion 1] ✓ [Criterion 2] ✓ [Criterion 3]
Example

"Create a LinkedIn post following these requirements: Content: 1) Start with a hook question, 2) Share one specific insight, 3) End with clear CTA. Format: Keep under 150 words. Quality Checks: ✓ No jargon ✓ One clear takeaway ✓ Professional tone."

T03 Conditional Instructions Tasks where approach depends on conditions
First, determine [classification criteria]. If [Condition A]: - [Specific action 1] - [Specific action 2] If [Condition B]: - [Alternative action 1] - [Alternative action 2] If [Condition C]: - [Third alternative action 1] - [Third alternative action 2] Then, [final common step for all conditions].
Example

"First, determine lead score. If score above 80: Schedule immediate call with senior rep. If score 50-80: Send personalized email sequence. If score below 50: Add to nurture campaign. Then, update CRM with next action date."

T04 Iterative Refinement Outputs needing progressive improvement
Create [output type] using this iterative process: Draft 1: [Initial requirements] - Focus on [core element] - Don't worry about [secondary elements] Draft 2: Improve Draft 1 by: - Adding [enhancement 1] - Refining [enhancement 2] Draft 3: Polish Draft 2 by: - Checking for [quality criterion 1] - Ensuring [quality criterion 2] Provide your final Draft 3 as the complete output.
Example

"Create a product description using this iterative process: Draft 1: List all features in bullet points. Draft 2: Convert each feature into a customer benefit with specific examples. Draft 3: Polish by checking tone is conversational and removing any jargon."

T05 Validation Checklist Critical outputs where errors have consequences
[Primary instruction for creating the output] Before finalizing, validate against this checklist: Content Validation: □ [Criterion 1] □ [Criterion 2] □ [Criterion 3] Format Validation: □ [Requirement 1] □ [Requirement 2] Compliance Validation: □ [Standard 1] □ [Standard 2] Only submit output after all checkboxes can be marked complete.
Example

"Write a legal disclaimer. Before finalizing, validate: Content — covers liability, warranty, jurisdiction. Format — under 150 words, plain language, no ambiguous terms. Compliance — matches standard template, reviewed for clarity."

Pro Tips

  • Number your steps: 'Step 1', 'Step 2' creates clear structure that AI follows reliably — numbered steps outperform vague sequences.
  • Use action verbs: Start each instruction with verbs like 'Create', 'Analyze', 'Extract', 'Convert' for unambiguous direction.
  • Include negative instructions: 'Do not include X' or 'Avoid Y' prevents common mistakes just as effectively as positive instructions.
  • Specify output format: Always end with 'Provide output as [format]' to control structure and avoid surprises.
  • Test and refine: Run your instructions once, see what breaks, then add clarifications. One refinement cycle usually locks in a reusable template.

Understanding the Fundamentals

These techniques work because of how AI actually processes language. Our Making AI Make Sense series breaks down exactly why — making techniques like this intuitive instead of mysterious. 34 videos, free on YouTube.

Watch on YouTube →

Need Help Applying AI to Your Business?

Templates teach the technique. We help you figure out where it actually fits in your work.

See Our Services Talk to Us

Eau Claire AI — Your path to AI, made clear.