Understanding the Full Prompt Display

Summary

To help you get the most value from AI Configurations in the PIM, we strongly recommend following the guides below:

 

What Is the Full Prompt?

The Full Prompt is a transparency feature that shows you exactly what instructions are being sent to the AI model when generating content. It combines all the different instruction layers into a single, readable view, helping you understand how your AI Configuration works behind the scenes.

When you preview an AI Configuration, you can view the complete prompt in a dedicated Full Prompt tab. This feature is particularly valuable for:

  • Debugging: Understanding why the AI produces certain outputs
  • Optimization: Identifying redundancies or conflicts in your instructions
  • Learning: Seeing how Akeneo structures prompts for different attribute types
  • Auditing: Documenting the exact instructions used for compliance or quality assurance
GenAI Full Prompt

How the Full Prompt Is Constructed

The Full Prompt is assembled from multiple components that work together to guide the AI model. Here's how these components are combined:

1. System Instructions

This is the foundational layer that defines the AI assistant's role and core behavior. It includes:

  • The assistant's identity as a "product information manager assistant"
  • General guidelines for handling product information, images, and documents
  • Fallback behavior (using the __NULL__ keyword when generation isn't possible)
  • Core principles like preferring incomplete answers over no answer

Example System Instructions:
"You are a product information manager assistant, and you will be given information about a product and a task to perform. Your specific job is to generate a text response to the task, using the available product information, documents and images if provided."

 

2. User Message

The User Message contains the specific task instructions and is composed of three sub-sections:

a) Task Description

  • Identifies the target attribute (code and labels in multiple languages)
  • Lists numbered instructions for completing the task
  • Specifies the output language based on the selected locale
  • Includes attribute-specific constraints (e.g., maximum character length for text attributes)

b) Custom Instructions (Optional)

This section appears only when you've defined custom instructions in your AI Configuration's Instruction Prompt field or when youve defined custom intructions at the Family level. It's wrapped in <CUSTOM_INSTRUCTIONS> tags and explicitly tells the AI to prioritize these instructions over general guidelines when conflicts arise.

Example Custom Instructions:

Additional task-specific instructions from the user:
<CUSTOM_INSTRUCTIONS>
You are an assistant writing product marketing copy in the style of 
Braun's brand voice.

Write a product marketing description for the product described below.

Tone and style guidelines:
- Sound calm, precise, and human-centred.
- Use simple, clear language with short sentences and minimal adjectives.
- Lead with function, then explain the concrete benefit in everyday life.
</CUSTOM_INSTRUCTIONS>
 

c) Product Information

Contains the actual product data that the AI will use, including:

  • Product title
  • Product family
  • Attribute values from selected source attributes
  • Asset references (images, PDFs) when enabled

Example Product Information:

<PRODUCT_INFO>
- **Title**: Braun BN0095 Wrist watch
- **Other information**:
    - Brand: Braun (website: braun.com)
    - Color: Silver
    - Material: Stainless steel
</PRODUCT_INFO>
 

Complete Prompt Structure Example

Here's how a complete Full Prompt looks when assembled:

# System Instructions

You are a product information manager assistant, and you will be given 
information about a product and a task to perform...

# User Message

Your task is to generate text content for a specific product attribute.

Target attribute information:
- Identifier code: "long_description"
- Descriptive labels: Long description (en_US)

Instructions:
1. Read through the product information thoroughly inside <PRODUCT_INFO>...
2. Look for any mentions of the target attribute...
3. Generate appropriate text content for this attribute...
4. Use the custom user instructions to help you generate the text content...
5. Provide your response in the following language: English (United States).

Additional task-specific instructions from the user:
<CUSTOM_INSTRUCTIONS>
[Your custom instructions appear here]
</CUSTOM_INSTRUCTIONS>

Now, carefully examine the following product information:

<PRODUCT_INFO>
[Product data appears here]
</PRODUCT_INFO>
 

Accessing the Full Prompt

The Full Prompt display is available when previewing AI Configurations:

  1. Navigate to an AI Configuration in edit mode
  2. Scroll to the Preview AI Generated Content section
  3. Select a product to preview
  4. Click the System Instructions or Debug tab in the preview results

Why the Full Prompt Is Useful

Transparency and Control

By seeing exactly what instructions are sent to the AI, you maintain full control and understanding of your AI Configuration's behavior. There are no "hidden" instructions or mysterious behaviors.

Debugging and Optimization

If the AI generates unexpected content, reviewing the Full Prompt helps you:

  • Identify conflicting instructions between system defaults and your custom instructions
  • Spot missing or ambiguous guidance
  • Understand how product data is structured and presented to the AI
  • Verify that attribute constraints (like character limits) are properly communicated

Learning and Best Practices

The Full Prompt serves as a learning tool, showing you how Akeneo engineers structure effective AI prompts. You can:

  • Learn professional prompt engineering patterns
  • Understand how to properly scope and structure instructions
  • See how context (product info, images) is formatted for optimal AI comprehension
  • Apply these patterns to your own custom instructions

Documentation and Compliance

For organizations with strict content governance, the Full Prompt provides:

  • A complete audit trail of AI instructions
  • Documentation for regulatory compliance
  • Evidence of responsible AI usage
  • Training materials for new team members

How System Prompts Are Prioritized

When multiple instruction layers exist, they follow this priority hierarchy:

Priority Layer Description
1 (Highest) Custom Instructions Your AI Configuration's custom prompt defined in the Instruction Prompt field
2 Task-Specific Instructions Attribute type requirements (max length, output format, constraints)
3 (Lowest) System Instructions Core behavioral guidelines that apply to all generations

The Full Prompt explicitly tells the AI: "If there are conflicts between these custom instructions and the general instructions above, prioritize the custom instructions." This ensures your specific requirements always take precedence.

 

Differences Across Attribute Types

The Full Prompt structure adapts based on the target attribute type:

  • Text attributes: Include character length constraints and HTML formatting guidelines
  • Select attributes: Include available options and selection rules
  • Number/Metric attributes: Include unit constraints and numerical format requirements
  • Reference Entity attributes: Include entity structure and relationship rules

The System Instructions section remains consistent, while the Task Description adapts to each attribute type's specific needs.

Best Practices

When using the Full Prompt display:

Practice Why It Matters
Review Before Publishing Always check the Full Prompt before enabling an AI Configuration to ensure instructions are clear and complete
Look for Conflicts Check if your custom instructions contradict system instructions, and adjust if necessary
Verify Product Data Ensure the <PRODUCT_INFO> section contains the attributes you expect
Test Edge Cases Preview with products that have minimal data to see how the AI handles missing information
Document Custom Logic Use the Full Prompt as documentation when training team members on your AI Configurations

Common Use Cases

Debugging Unexpected Output

Scenario: Your AI Configuration generates descriptions that are too long, even though you specified "keep it brief" in your custom instructions.

Solution: Check the Full Prompt to see if:

  • Your "brief" instruction is clear enough (e.g., specify "under 100 characters" instead)
  • The attribute's max_length constraint is properly set and communicated
  • There are conflicting instructions asking for "comprehensive" or "detailed" content

Training New Team Members

Scenario: A new merchandiser needs to understand how your brand voice AI Configuration works.

Solution: Use the Full Prompt as a teaching tool:

  1. Show them how System Instructions define the AI's base behavior
  2. Explain how your Custom Instructions layer on brand-specific requirements
  3. Demonstrate how Product Information flows into the generation process
  4. Review the priority hierarchy to clarify what happens when instructions conflict

Troubleshooting

Issue Possible Cause Suggested Fix
Custom instructions not appearing Instruction Prompt field is empty Verify that you've saved custom instructions in the AI Configuration's Instruction Prompt field
Product data missing Source attributes not configured Check that you've selected the correct source attributes in the "Data to take into account" section
Conflicting instructions Custom and system instructions contradict Remember that custom instructions take priority. Adjust your custom prompt to be more specific about desired behavior

Next Steps

After understanding the Full Prompt feature:

Want to find out more? Take a look at these related articles