Configuring an Attribute for AI Enrichment

This is the core article for setting up attributes in Merchkit. You'll walk through the Configure Attribute panel, which is where you tell Merchkit's AI exactly what to generate.

Opening the Configure Attribute Panel

1. Go to Workspace → Attributes
2. In the attributes table, click any attribute name (or the row)
3. The Configure Attribute panel slides open on the right side

[SCREENSHOT: Attributes table with one attribute row highlighted, Configure Attribute panel visible on the right side]

Understanding the Configure Attribute Panel

The panel contains all the settings you need to configure an attribute. Let's walk through each one:

Use AI Toggle

Located at the top of the panel, the Use AI toggle controls whether Merchkit's AI will generate or enrich values for this attribute.

• Toggle ON — The AI will generate or enrich this attribute when you run enrichment
• Toggle OFF — This attribute is a reference field only; the AI ignores it

For attributes you want the AI to fill in, turn this on.

[SCREENSHOT: Configure Attribute panel showing Use AI toggle in the ON position]

Attribute Name

The name of your attribute (e.g., short_description, brand, care_instructions). You can edit this if needed.

Field Type

The data type that this attribute stores. Options include:

• Single Line Text — Short text values (names, SKUs, colors)
• Number — Numeric values (prices, inventory counts)
• Image — Product images or image references
• Formula — Computed values based on other fields
• Reference Lookup — Links to a single related record
• Collection Reference — Links to multiple related records
• Record Reference — Links to a single record in another entity type

For AI enrichment, you'll most commonly use Single Line Text or Number. For a detailed breakdown of each field type, see Field Types.

Product Data: Tell the AI What to See

The Product Data section shows you which product fields the AI can access when generating this attribute. By default, Merchkit includes title and description, but you can customize this.

Click Product Data to expand it and see which fields are selected. You'll see something like:

✓ title
✓ description
✓ price
  vendor
  tags

Checked fields are sent to the AI as context. Unchecked fields are not.

Why does this matter? The more relevant context you give the AI, the better its output. For example:

• If you're generating a "brand" attribute, make sure the product description is included (it often mentions the brand)
• If you're generating "size_range", include price (expensive items may come in more sizes)
• If you're generating something unrelated to price, you can uncheck it to reduce noise

Select the fields that are most relevant to this attribute.

[SCREENSHOT: Product Data section expanded, showing checkboxes for title, description, price, vendor, tags, etc.]

Image Source: When You Need the AI to Look at Pictures

The Image Source section tells the AI which product images to analyze when generating this attribute.

You'll only use this if your attribute requires image analysis. Examples:

• "product_color" — the AI looks at the image to detect color
• "has_packaging" — the AI looks for packaging in the image
• "style_category" — the AI analyzes the visual style

If your attribute doesn't need images (e.g., "SEO title", "material"), leave Image Source unchecked.

Click Image Source to expand and select which product images to include (primary image, all images, etc.).

[SCREENSHOT: Image Source section showing options like "Primary Image", "All Images", etc.]

Prompt: Your Instruction to the AI

This is the most important field. The Prompt is your instruction to the AI—it tells the AI exactly what to generate.

Example prompts:

Generate a concise product title under 60 characters based on the product description and title.

Based on the product description and images, identify the primary material. Return only one material name.

Analyze the product image and determine if it shows packaging. Return 'Yes' or 'No'.

Using dynamic references: You can reference other attributes in your prompt using {{ }} tags. For example:

Generate an SEO-friendly meta description for {{product_title}} under 160 characters.

When the AI runs, {{product_title}} will be replaced with the actual product title.

To insert a reference:

• Type @ in the prompt field, and a dropdown of available attributes will appear
• Or click the Insert Value button below the prompt field to browse and select an attribute

[SCREENSHOT: Prompt field with @ symbol typed, showing attribute dropdown]

[SCREENSHOT: Insert Value button and resulting attribute selector popup]

Acceptable Values: Constrain What the AI Can Return

The Acceptable Values field lets you restrict the AI's output to a specific list of values. Use this when you want to ensure consistent, controlled outputs.

Enter comma-separated values. For example:

Small, Medium, Large, XL

Red, Blue, Green, Black, White, Other

When you set acceptable values, the AI will only return one of those options. This is useful for:

• Size attributes (where you want standardized sizes)
• Category or type attributes (where options are fixed)
• Status attributes (approved, rejected, needs_review)

Leave this blank if you want the AI to have freedom to generate any value.

[SCREENSHOT: Acceptable Values field populated with comma-separated list]

Note: For detailed guidance on using acceptable values effectively, see Using Acceptable Values to Constrain AI Output.

Edit with AI Button

Click Edit with AI to let Merchkit's AI help you refine your prompt. Describe what you want to change, and the AI will suggest improvements.

For example:

• "Make this prompt simpler"
• "Add a requirement for the output to be under 50 characters"
• "Make sure it prioritizes the product images"

[SCREENSHOT: Edit with AI button highlighted]

The Relationship Between These Settings

Here's how all these pieces work together:

1. Product Data + Image Source = What the AI sees (its context)
2. Prompt = What the AI does (your instruction)
3. Acceptable Values = What the AI is allowed to return (constraints)

A complete attribute configuration might look like:

In this example, the AI sees the product details and primary image, follows your instruction, and can only return one of the 11 color options you specified.

Saving Your Configuration

After you've configured all the fields:

1. Click Save Changes at the bottom of the panel
2. Merchkit will save your configuration
3. The attribute is now ready for AI enrichment

[SCREENSHOT: Configure Attribute panel with Save Changes button visible at the bottom]

Common Configuration Issues

Issue: "The AI is giving me output that doesn't match my Acceptable Values"

• Make sure your Acceptable Values are spelled exactly as you want them
• Re-run enrichment on this attribute after saving changes
• Check that the prompt isn't asking for something outside your acceptable values

Issue: "The AI is including irrelevant information in its output"

• Reduce the Product Data fields — uncheck fields that don't help
• Make your prompt more specific (e.g., "Return only the primary material, nothing else")
• Add Acceptable Values to constrain the output

Issue: "The AI needs to see product images but I haven't selected an Image Source"

• Open Configure Attribute, expand Image Source, and select which images to include
• Re-run enrichment

What's Next?

• To write better prompts: Writing Effective AI Prompts
• To use acceptable values strategically: Using Acceptable Values to Constrain AI Output
• To learn about field types in detail: Field Types
• Ready to run enrichment? Running Your First Enrichment