Data Quality Best Practices
Product data quality directly affects how your items appear in ChatGPT and ultimately impacts conversion rates. This chapter consolidates ACP’s data quality requirements and recommendations.6.1 ID Stability
Product IDs and variant IDs must remain stable throughout their entire lifecycle.- ChatGPT tracks purchase history and user preferences by product ID
- Changing an ID is equivalent to creating an entirely new product; all prior data is lost
- ID instability degrades the product’s ranking and frequency in AI recommendations
6.2 Description Format
Provide at least one description format. Plain text is recommended:- Accurately reflect the product’s actual features
- Include key specifications (dimensions, weight, materials, etc.)
- Avoid marketing hyperbole
- Be easy for AI systems to parse and retrieve
6.3 Variant-Specific Fields
The following fields should be set independently on each Variant, not only at the Product level:| Field | Purpose | Rationale |
|---|---|---|
title | Variant title | Distinguishes different specifications |
url | Variant page URL | Links directly to the specific variant page |
description | Variant description | Describes specification-specific differences |
media | Variant images | Shows images for the corresponding color/style |
availability | Stock status | Different variants may have different inventory |
price | Price | Different variants may have different prices |
variant_options
Usevariant_options to label user-facing specification dimensions:
6.4 URL Conventions
Add Tracking Parameters
Includeutm_medium=feed in all product URLs to track traffic originating from ACP:
Use Durable Public URLs
Seller URLs must:- Be durable public URLs
- Not contain expiring session tokens or temporary parameters
- Remain accessible over the long term
URL Encoding
All URLs must be valid and properly encoded:%20 and handle special characters per URL encoding standards.
6.5 Prohibited Content
The following product categories must not be submitted through ACP:| Category | Description |
|---|---|
| Adult content | Pornographic or sexually explicit material |
| Alcohol / Tobacco / Gambling | Alcoholic beverages, nicotine products, gambling services |
| Weapons | Firearms, ammunition, regulated bladed weapons |
| Prescription drugs | Medications requiring a doctor’s prescription |
| Unauthorized financial services | Unlicensed financial or investment products |
| Illegal goods | Any products violating applicable laws |
| Deceptive products | False advertising, counterfeits, misleading merchandise |
6.6 Data Freshness
Keep inventory and pricing data up to date:| Data Type | Recommended Update Frequency |
|---|---|
| Stock status | Real-time or hourly |
| Price changes | Immediately upon change |
| New product listings | Push via API immediately |
| Product delistings | Update immediately or remove in the next daily snapshot |
| Promotions | Create in advance; set automatic expiration |
6.7 Image Quality Guidelines
| Requirement | Details |
|---|---|
| Resolution | High-resolution with clearly visible product details |
| Background | Solid white or neutral background preferred |
| Angles | Primary image shows the product front; supplementary images show multiple angles |
| Format | JPEG or PNG; URLs must be publicly accessible |
| Consistency | Maintain a consistent style across images of the same product |
| No text overlays | Avoid adding promotional text on product images |
6.8 Quality Checklist
Before submitting product data, verify each item:- All IDs are stable and unique
- Every Product has at least one Variant
- Every Variant has a title, price, and availability status
- Prices use minor units (cents, not dollars)
- At least one description format is provided
- URLs are valid and properly encoded
- URLs include the
utm_medium=feedtracking parameter - Image URLs are publicly accessible
- No prohibited-category products are included
- Variant-specific fields are set independently on each Variant
Next chapter: Chapter 7: Testing and Validation — Small-batch testing, format validation, and common error troubleshooting