The Prompt field in the Configure Attribute panel is where you tell the AI exactly what you want it to generate. A well-written prompt dramatically improves output quality—vague instructions lead to vague results, while specific instructions lead to consistent, usable data.
The Power of Specificity
Your prompt is the conversation between you and the AI. The more context and clarity you provide, the better the output.
Bad prompt: "Write a description"
Good prompt: "Write a 2-3 sentence product description for {{product_title}} by {{brand}}, highlighting the key material ({{material}}) and intended use. Keep the tone professional and avoid superlatives."
The good prompt tells the AI:
- Exactly how long (2-3 sentences)
- What product (using {{product_title}})
- Who makes it ({{brand}})
- What to focus on (material and intended use)
- What tone to adopt (professional)
- What to avoid (superlatives)
Use {{ }} Tags to Inject Product Data
The {{ }} syntax lets you reference other attributes in your prompt. When the AI processes your prompt, it substitutes the actual product values.
Common tags:
{{product_title}}— the product name{{brand}}— manufacturer or brand name{{material}}— material composition{{color}}— primary color{{size}}— dimensions or size category{{price}}— MSRP or list price{{description}}— existing description
Example: "Extract the warranty period from the source document for {{product_title}}. If no warranty is mentioned, respond with 'Not specified.'"
How to insert tags:
- Type
@in the Prompt field to open the attribute selector, then choose the attribute you need - Or click the "Insert Value" button below the Prompt field and select from the list
Using {{ }} tags ensures the AI has product-specific context and can tailor output to each product rather than generating generic text.
Tell the AI What NOT to Include
Just as important as telling the AI what to do is telling it what to avoid.
Example prompt: "Generate a short product description for {{product_title}}. Focus on material, durability, and use cases. Do not include price, warranty information, or marketing hype."
Negative instructions help the AI stay focused and prevent common issues like:
- Including irrelevant information
- Being too promotional or vague
- Repeating information already in other fields
- Generating content that's too long or too short
Format and Tone Matter
Be explicit about format expectations:
- Length: "2-3 sentences," "under 100 characters," "bullet points"
- Tone: "professional," "casual and friendly," "technical"
- Style: "avoid superlatives," "use active voice," "include specific measurements"
Example: "Create a bullet-point list of 3-4 key features for {{product_title}}. Use active voice and be specific about measurements or capabilities. Keep each point under 15 words."
Use the "Edit with AI" Button to Iterate
Once you've written your first prompt, you don't have to get it perfect. Click the "Edit with AI" button to let Claude help you refine it.
The Edit with AI workflow:
- Write your initial prompt
- Click "Edit with AI" below the Prompt field
- Claude suggests improvements—more specific language, added context, better structure
- Accept or modify the suggestion
- Run a test generation to see how the updated prompt performs
This is especially useful when you're unsure about phrasing or want to add {{}} tags you might have missed.
Good vs. Bad Prompts: Examples
| Scenario | Bad Prompt | Good Prompt |
|---|---|---|
| Color detection | "What color is this?" | "Based on the product images and title {{product_title}}, identify the primary color. Return only the color name, nothing else." |
| MSRP extraction | "Get the price" | "Extract the MSRP or list price for {{product_title}} from the source document. Return only the numeric value (e.g., 29.99). If not found, return 'Unknown.'" |
| Category assignment | "What category?" | "Based on {{product_title}}, {{description}}, and product images, assign the primary product category. Choose from: Home & Garden, Sports & Outdoors, Electronics. Return only the category name." |
| Material specs | "Describe the materials" | "List the primary materials used in {{product_title}} in the following format: Material Type: [name], Percentage: [%], Durability: [high/medium/low]. Be specific based on images and source docs." |
Common Mistakes to Avoid
1. Being too vague
- ❌ "Create a title"
- ✅ "Create a search-friendly product title for {{product_title}} that includes the material and primary color. Keep it under 60 characters."
2. Forgetting {{ }} tags
- ❌ "Describe the product"
- ✅ "Describe {{product_title}} by {{brand}} in 1-2 sentences, emphasizing {{material}} and durability."
3. Asking for too many things at once
- ❌ "Write descriptions, extract specs, find the category, and suggest keywords"
- ✅ Create separate attributes, each with a focused prompt
4. Not setting expectations for edge cases
- ❌ "Extract the warranty period"
- ✅ "Extract the warranty period from the source document. If no warranty is stated, respond with 'Limited Warranty (check with manufacturer).'"
5. Using subjective language
- ❌ "Make it sound nice and describe the quality"
- ✅ "Write a 2-3 sentence description highlighting durability, material, and intended use. Use professional tone. Avoid marketing language."
Next Steps
Once you've mastered prompt writing, the next step is to use Acceptable Values to constrain AI output to specific, approved values for categorical data. See Using Acceptable Values to Constrain AI Output.