Overview

The Unified AI Configurations feature introduces a centralized location for managing all AI-related attribute configuration settings in Akeneo. This includes prompt-based enrichment, asset generation, and translation customization previously configured at the attribute or family level.

By consolidating AI settings into one dedicated area, users benefit from a streamlined, scalable, and more maintainable configuration experience.

Access the page via:
Settings → AI Configurations

Configuration Structure

The new AI configurations page enables users to define reusable configurations that can be applied via the Rules Engine or manually from the Product Edit Form. These can be for all attributes or specific to one attribute, as defined by you in the configuration.

Each configuration includes:

• General Information
  • A label and code for identification, and the status (enabled/disabled).
• Target & Source
  • Select a specific attribute (or apply to all)
  • Define which attribute values should be used for content generation.
    (Note: This does not apply to translations)
• PX Insights User Reviews (Optional)
  • Automatically includes relevant customer review data if PX Insights is enabled.
    (Contact your CSM for more details)
• Assets (Optional)
  • You can include assets to be considered for your configuration; you can also choose specifically which asset collections to consider. 
• Instruction Prompt
  • Choose a predefined prompt or define your own custom instruction.

Example of AI configuration page

 

 

Preview the content generated

You can preview the content generated by your prompt directly from the AI Configurations page. Select a product or a product model from your catalog and click Generate to see the output produced by the current configuration (instruction prompt, selected attributes, and optional assets). This way, you:

• Reduce back-and-forth to the product grid while testing prompts.
• Accelerate content generation by iterating quickly on instructions.
• Increase confidence by validating outputs before automating them in the Rules Engine.

How to use the preview

1. In your AI configuration, make sure a prompt has been typed/selected
2. If the Preview is not yet open, in the top right corner of the AI Configuration, click Preview

   

3. Select a channel and a locale
4. Select a product
   • Enter a UUID or a main identifier to select a product
   • Enter a UUID or a code to select a product model
5. Click Preview

Once a valid reference is provided, Akeneo fetches the matching record and runs your prompt using the configuration’s settings and any selected asset collections. The generated result is displayed instantly in the preview panel.

Example Configurations

1. SEO Metadata Generation

Example Configuration:

• Label: SEO Meta Generator
• Code: seo_meta_generator
• Target Attribute: seo_metadescription
• Context Attributes: product_name, short_description
• Assets: Not included
• PX Insights: Enabled
• Prompt:

#Role:
Act as an expert SEO copywriter and senior product marketing manager with over 15 years of experience. Your specialty is crafting copy that is both user-engaging and perfectly optimized for search engines.

#Context:
For the current product, you will receive a set of attributes that describe it. Do not invent any information that does not come from this data.

#Tasks:
Your first task is to analyze these attributes to identify the following essential information:
* The product's name
* Its key features and benefits
* Its target audience
* The primary keywords

Once this information is identified, generate 3 distinct SEO meta descriptions. Each description must strictly adhere to the following criteria:
1. Character Limit: Do not exceed 155 characters, including spaces.
2. Keyword Integration: Seamlessly and naturally integrate the primary keywords you have identified.
3. Benefit-Oriented: Clearly communicate a key benefit for the identified target audience.
4. Call to Action (CTA): End with a clear and compelling call to action.

#Example of Expected Output:
* Input attributes provided:
** Product name: "Pro LED Architect Desk Lamp"
** Main web category : "Lighting > Desk Lamps"
** Unique Selling Proposition: "Articulated metal arm, 3 brightness levels, low-consumption LED bulb included, integrated USB port for device charging."
** Target audience: "professionals, students, remote workers"
** SEO primary keywords: "LED desk lamp", "architect lamp"
* Generated meta description:
"Need an LED desk lamp? Our Pro architect lamp with a USB port is ideal for pros and students. Brighten your workspace. Order yours today!"

2. Data Quality Audit

Example Configuration:

• Label: Data Quality Audit
• Code: data_quality_audit
• Target Attributes: All
• Assets: Included
• PX Insights: Disabled
• Prompt:

#Role:
You are an expert PIM consultant specializing in data quality and enrichment. Your goal is to help e-commerce teams improve their product data to increase conversion rates and reduce end-customer support inquiries.

#Task:
Carefully analyze the provided product data and identify all data quality issues based on the criteria listed.
For each product, provide a summary of issues in a bulleted list.

#Criteria for analysis
1. Missing or empty required attributes: Identify any key attributes that are empty or missing.
2. Format inconsistencies: Check for mismatches in currency, date formats, units of measurement, etc.
3. Logical clashes: Find contradictions in the data (e.g., a product's shipping weight is greater than its net weight, an 'eco-friendly' product made with non-recyclable materials).
4. Asset & attribute discrepancies: Note mismatches between information in the product assets (like images or documents) and the attribute values (e.g., an image shows a blue shirt, but the 'color' attribute is "Red").
5. Anomalous or outlier values: Detect values that are far outside the expected range (e.g., a price of $0.01, a weight of 500 kg for a t-shirt).

#Output format:
For each issue found, create a new line starting with a bullet point for the following details:
* Attribute/Asset: [The name of the attribute or asset with the issue]
* Problem: [A brief, clear description of the problem]
* Recommendation: [A specific, actionable step to correct the issue]

#Example:
##Product Data Input:
{
"identifier": "AKN_TS_105",
"family": "T-Shirts",
"values": {
"name": "Men's V-Neck Tee",
"description": "A comfortable 100% cotton v-neck t-shirt. Available in Red.",
"price-USD": "25.00",
"price-EUR": null,
"main_image": "v-neck-tee-blue.jpg",
"materials": "100% Cotton",
"sustainability_eco_friendly": "Yes",
"sustainability_materials": "Polyester Blend",
"weight-kg": "50",
"release_date": "2024/03/15"
}
}

##Expected Output for Example (different attributes):
* Product: AKN_TS_105
* Attribute: price-EUR
* Problem: The price for the EUR currency channel is missing.
* Recommendation: Add the correct price in EUR based on the current conversion or pricing strategy.
* Asset: main_image
* Problem: The image filename v-neck-tee-blue.jpg and description "Available in Red" suggest a color mismatch.
* Recommendation: Verify the correct product color. Update either the main image to a red shirt or the description to "Blue".
* Attribute: sustainability_eco_friendly & sustainability_materials
* Problem: There is a logical clash. The product is marked as "eco_friendly" but the material is listed as "Polyester Blend," which contradicts the description's "100% Cotton" and is less eco-friendly.
* Recommendation: Correct the sustainability_materials and description to be consistent. If it's cotton, update the materials attribute. If it's a blend, re-evaluate the "eco_friendly" tag.
* Attribute: weight-kg
* Problem: A weight of 50 kg is an extreme outlier for a t-shirt.
* Recommendation: Re-weigh the product and update the attribute to a realistic value, likely around 0.15 - 0.25 kg.
* Attribute: release_date
* Problem: The date format YYYY/MM/DD is non-standard.
* Recommendation: Change the date format to the standard YYYY-MM-DD for consistency.

3. Tone-Controlled Translation

Example Configuration:

• Label: Brand Tone Translation
• Code: brand_tone_translation
• Target Attributes: All localizable text fields
• Assets: Not included
• PX Insights: Disabled
• Prompt:

#Role:
You are the Head of Global Content for "Atelier Solène," a luxury fashion house. You are the ultimate custodian of our brand's voice, responsible for ensuring every word reflects our heritage of elegance and modern sophistication.

#Context:
Atelier Solène is built on three pillars: Timeless Elegance, Artisanal Craftsmanship, and Modern Sensibility. Our audience is discerning, globally-minded, and appreciates the story behind the product. They are not looking for a simple garment; they are investing in a piece of art. Your language must convey this value.

#Tone of Voice Guidelines:

##Do:
Be Evocative & Sensory: Use rich, descriptive adjectives. Instead of "nice fabric," describe the "sumptuous feel of pure silk" or the "structured drape of Italian wool."
Emphasize Craftsmanship: Highlight the skill and heritage behind each piece. Use terms like "meticulously crafted," "hand-finished," and "expertly tailored."
Be Elegant & Aspirational: The tone should be sophisticated and elevated, but remain clear and inspiring.

##Don't:
Avoid Overly Casual Language: No slang, contractions (don't, it's), or overly familiar phrases.
Avoid Technical Jargon: Instead of "interwoven high-denier polyester," prefer "a resilient, finely woven textile."
Avoid Clichés & Hard-Selling: Never use phrases like "must-have," "best-seller," or "get yours now." Our brand's appeal is implicit.

#Mini Glossary (Brand-Approved Terms):
When you encounter these French terms, you must use the approved English translation and context.
* "Savoir-faire": Always translate as "Artisanal Craftsmanship" or "Expert Artistry."
* "Pièce maîtresse": Translate as "Statement Piece."
* "Prêt-à-porter": Translate as "Ready-to-Wear."
* "Atelier": Use the French word "Atelier" directly, as it is part of our brand's DNA. Do not translate as "workshop."

#Task:
Translate the following product description fields from French to English, perfectly adhering to the role, context, tone of voice guidelines, and glossary provided.

#Example:
##French Locale (Input):
* Product Name: Robe "Solène" en Soie
* Short Description Découvrez la pièce maîtresse de notre nouvelle collection prêt-à-porter. Une robe qui illustre notre savoir-faire unique.
* Materials: 100% Soie

##English Locale (Expected Output):
* Product Name: Silk Dress "Solène" 
* Short Description: Introducing the statement piece of our new Ready-to-Wear collection. This dress is a testament to our expert artistry.
* Composition: 100% Silk

4. Set simple or multi-select values based on Asset

Example Configuration:

• Label: Set Color Scheme /w Asset for Shirt Family
• Code: set_color_scheme_w_asset_for_shirt_family
• Target Attributes: color (simple select)
• Assets: Packshot
• PX Insights: Disabled
• Prompt:

#Role:
You are an AI-powered PIM Data Specialist. Your function is to automate the tagging of product attributes based on visual data from images.

#Context:
You are enriching product data for a large e-commerce website. Accurately identifying the base color of apparel is critical for creating effective search filters and improving the customer's ability to find products. An error in color tagging leads directly to a poor customer experience.

#Task:
Analyze the provided product image(s). Identify the single, dominant base color of the main garment shown. Your response MUST be one of the exact values from the "Allowed Colors" list below, and nothing else.

#Instructions:
* Focus only on the garment's most prevalent, solid color.
* Ignore small logos, patterns, prints, graphics, or stitches of a different color.
* In your assessment, digitally isolate the product from its background and any props.

Prompt Guidelines for Generating Product Content with AI

When creating product content using AI models like ChatGPT, it’s important to provide clear and detailed instructions to ensure the generated descriptions align with your goals. Follow these guidelines to craft effective prompts that will produce precise, engaging, and high-quality product content.

1. Identify Key Features

List the essential features of the product, such as specifications, benefits, and unique selling points (USPs). Highlight what makes the product stand out compared to similar items in the market.

2. Specify the Target Audience

Consider who will be reading the product description. Tailor your prompt to resonate with your target audience's needs, interests, and preferences to create relevant content.

3. Choose the Tone of Voice

Define the tone and voice of the description. This can range from formal to informal, enthusiastic to neutral, and can include voices like professional, friendly, or whimsical. Being specific about the tone and voice helps maintain consistency in your product content.

4. Provide Context

Offer context about the content’s purpose and intended use. This helps ensure that the AI-generated content aligns with the overall goals and objectives of your project.

5. Include Relevant Technical Specifications

If applicable, provide important technical details like dimensions, material, and compatibility. These are especially important for tech products, appliances, and gadgets to ensure accurate descriptions.

6. Mention Constraints or Special Instructions

Outline any specific constraints or requirements for the description. For example, if certain phrases, keywords, or compliance guidelines need to be included, specify them clearly in the prompt.

7. Request Examples or Comparisons

If necessary, ask the AI to include comparisons or examples to make the product description more relatable. Comparing the product to familiar items or scenarios can help customers understand its value more easily.

8. Clarify the Purpose of the Content

Specify where the generated content will be used, whether for an online store, catalog, or marketing material. This will help shape the content to best suit the intended platform or audience.

9. Review and Refine the Prompt

Before submitting your prompt, proofread it to ensure clarity and completeness. You can test your prompt with several products in the Product Edit Form before automating it through the Rules Engine.

Example Prompt:

Product: Portable, rechargeable hand warmer "WarmPal"
Target Audience: Outdoor enthusiasts and commuters in cold climates
Tone: Friendly and informative
Voice: Casual, yet knowledgeable
Key Features:
- Compact design
- 12-hour battery life
- Fast-charging capability
- Eco-friendly materials
- Suitable for hiking, commuting, and winter outdoor activities

Technical Specifications:
- Dimensions: 3x2 inches
- Weight: 4 ounces
- Material: Recycled plastic
Special Instructions: Avoid technical jargon and make the description accessible to non-experts.

By following these guidelines and using the provided example as a template, you can create detailed and effective prompts to generate product content that aligns with your goals and resonates with your target audience.

Execution Options

1. Manual Execution via Product Edit Form

• Navigate to a product in the grid
• Locate the Ask AI button next to the target attribute
• Select Generate from configuration
• Choose the desired AI configuration
• The generated value will be populated immediately in the interface.

2. Automated Execution via Rules Engine

• Go to Settings → Rules
• Create or edit a rule
• Define product conditions as needed
• Add an action: Generate attribute value with AI
• Select the appropriate AI configuration from the list

Limits & Considerations

To ensure optimal performance and reliability, the AI enrichment features have a few important limitations:

Supported Asset Types

The following file types are supported for asset-based AI generation:

• Images: JPG, JPEG, PNG, WEBP, TIFF, and single-frame GIFs
• Documents: PDF

Both media filles and media links are supported.

Asset Limits

Each product can include:

• Up to 5 images (maximum 32MB in total)
• 1 PDF (maximum 20MB, first 20 pages analyzed)

Note: Larger files or additional assets beyond these limits will be skipped during processing.

Asset Collections

• Only assets stored in collections that include a “main media” attribute are supported.
• When you generate or process images through AI in the PIM, they are automatically resized to ensure optimal display and performance.
  • The system detects the longest side of the image (width or height).
  • If the longest side is greater than 1024 pixels, the image is proportionally resized so that its longest side is exactly 1024 pixels.
  • The aspect ratio is always preserved. The image is never distorted.

Examples:

• An image of 2000 × 1000 px generated by AI will be resized to 1024 × 512 px.
• An image of 800 × 600 px will not be modified, since its longest side is already below 1024 px.

Note: This resizing only applies to images created or processed by AI. Images uploaded directly into the PIM are not resized.

Processing Order

When multiple assets are available, the AI will prioritize the first images and/or PDF found in the selected asset collections. For more detailed extraction, you can provide the `asset_collection` to take into consideration. Click here for more information on how to organize assets

Attribute Option Limit

For select-type attributes (simple or multi-select) used as AI outputs:

• A maximum of 500 attribute options are supported
• Any options beyond this limit will be ignored during generation
• This limit is imposed by prompt size constraints from our AI provider.

Fair Usage and Billing

There are currently no hard usage limits, but we plan to introduce customer-specific usage limits to prevent misuse.

Data Privacy and Security

All data is processed through OpenAI’s Enterprise API:

• Your data is not used to train AI models
• No data is stored after processing
• All data is transmitted securely over encrypted channels

Understanding AI Limits & Realistic Use Cases

Generative AI in Akeneo PIM is a powerful tool for accelerating product enrichment, but it is not a search engine, a fact-checker, or a magic fix for incomplete data. It works best when you provide it with high-quality, relevant data from your catalog and clear, structured instructions.

To get the most accurate and useful results, it’s important to understand what AI can and cannot do within the PIM environment.

What AI Can Do

• Generate new product content (text, number, select values) based only on the data available in your PIM.
• Analyze attributes and assets you specify in the “Data to take into account” section.
• Understand and use:
• Product name (label attribute)
• SKU / Identifier
• Any listed attributes you explicitly include
• Uploaded or linked assets (e.g., images, PDFs)
• Follow structured prompts to produce content that fits your tone, brand, or formatting goals.
• Be reused safely and consistently across products once a configuration is validated.
   

What AI Cannot Do

AI cannot access or infer information outside your PIM data. Specifically:

Setting Realistic Expectations

Generative AI works best as an assistant, not a replacement. It requires context, structure, and good data to perform well.

Here are a few principles to guide your use:

Clarity beats complexity

• A short, well-focused prompt is better than a long, overbearing one.
  (If you ask it to “generate content like Amazon listings using tone from my marketing site,” but that data isn’t in your PIM, the result will be poor.)

AI cannot create what it doesn’t know

• If attributes are empty or missing, the model won’t invent accurate information. It can only use what exists.

Good prompts follow a framework

• We recommend using the C.R.A.F.T. method:
  • Clearly: State exactly what you want
  • Relevant: Use only meaningful attributes
  • Adapted: Match tone and style to your brand
  • Focused: Avoid combining multiple tasks in one prompt
  • Testable: Preview and iterate before automation

Iterate before automating

• Use the Preview tool to test prompts on sample products. Once confident, deploy via the Rules Engine for large-scale automation.

AI output ≠ final truth

• Always review AI-generated content for accuracy, compliance, and tone. 
• You are the expert. AI assists. It does not replace your judgment.
   

Example of a Poor Prompt

Prompt:
“Generate engaging product descriptions using my website data and the latest market trends. Make sure it fits our brand voice and lists features from our marketing brochure.”

Data to Take into Account:
None

Use Assets within product data:
No

This will fail because:

• The AI cannot access your website or brochures.
• “Brand voice” is undefined in the PIM.
• No product attributes were provided as context.

 Security Overview

Akeneo takes your data security and privacy seriously. All product enrichment, whether generated from attributes, from assets (images, PDFs), or translations, is governed by robust security standards and privacy controls. We use OpenAI Enterprise, ensuring that the data we provide isn't utilized in model training. This data flows exclusively within their servers, returning directly to us without storage or reuse. We share only precise information with OpenAI to ensure that we generate accurate content tailored to your requirements.

AI models used

We are currently using OpenAI's ChatGPT models:

Information shared with our AI provider

For content generation

• Product name: the value of the attribute used as a label
• Locale: the language of the generated content
• Values: the values of the product in the selected attributes to take into account
• Prompt: whether it is a pre-defined Akeneo prompt or your custom one that you have defined 

For content translation

• Source locale
• Target locale
• Value to translate
• Prompt: your custom one that you have defined 

For content rephrasing

• Value to rephrase
• Rephrase type 

Additional Security Measures

• Data is transmitted over secure, encrypted channels.
• No personal data or credentials are ever sent to the AI service unless explicitly included by the user.
• You have full control over which attributes, assets, and prompts are used and shared.
• Feature flags and permissions allow granular control of access and usage per user and per customer.

Compliance & Recommendations

• Always review AI-generated content before publishing for accuracy, compliance with company policy, and regulatory standards. You can use Workflows to do so.
• Do not include confidential or sensitive information in prompts or attribute values unless necessary and reviewed.
• If required, consult your legal or data protection team to ensure usage is in line with your organization’s privacy and data governance policies.