Skip to main content

OpenAI Product Feeds

Overview​

Catalog Delta Files provide a machine-readable snapshot of only the changes made since the last publish event. These files are ideal for feeding external AI platforms such as OpenAI without requiring full catalog re-ingestion.

Typical AI use cases include:

  • Semantic search indexing
  • AI shopper agents
  • AI merchandising agents
  • AI-driven recommendations
  • Conversational product discovery

How Delta Files Are Generated​

Delta files are generated automatically whenever:

  • A catalog is published
  • A partial republish occurs
  • Products, prices, availability, or attributes change

Each delta file contains:

  • Product IDs
  • Change type (create, update, delete)
  • Updated attributes
  • Relationships
  • Availability and pricing references

Delta files are generated alongside full publish artifacts and follow the same distribution model.

High-Level Data Flow​

  1. A catalog publish or update is triggered.
  2. Elastic Path generates a Catalog Delta File.
  3. Your ingestion service retrieves the delta file.
  4. The delta data is transformed into OpenAI Product Feed format.
  5. The transformed data is sent to OpenAI.
  6. OpenAI updates its product index using only the changed records.

Accessing Catalog Delta Files​

Delta files are accessed using the same Catalog Publishing process as full exports.

From the Commerce Manager:

  1. Navigate to: Product Experience Manager → Catalogs
  2. Select a catalog.
  3. Publish the catalog.
  4. Retrieve the generated delta file from the publishing artifacts.

For automated pipelines, delta files should be fetched via your existing catalog publishing automation.

Mapping Elastic Path Delta Data to OpenAI Product Feeds​

Each delta entry should be mapped into the OpenAI Product Feed schema.

Example Mapping​

Elastic Path FieldOpenAI Product Feed Field
product_idid
nametitle
descriptiondescription
skusku
priceprice
availabilityavailability
imagesimages
attributesattributes
delete_flagdeleted

Handling Change Types​

Your ingestion pipeline must handle the three delta operations correctly.

Create​

  • Insert a new product record into the OpenAI product index.

Update​

  • Replace only the modified fields.
  • Retain embedding continuity when possible to avoid re-vectorizing unchanged content.

Delete​

  • Remove the product from the OpenAI index.
  • Prevent future AI results from referencing inactive products.

Example Delta Processing Workflow​

  1. Poll catalog publish events or webhooks.
  2. Download the delta file.
  3. Validate schema and version.
  4. Normalize fields.
  5. Convert to OpenAI Product Feed format.
  6. Send to OpenAI ingestion endpoint.
  7. Log processing status and failures.

Best Practices​

  • Process delta files immediately after publish to minimize AI drift.
  • Maintain idempotency in your ingestion service.
  • Log every processed delta for audit and replay.
  • Version your field mappings to protect downstream AI pipelines.
  • Backfill with a full catalog export on schema changes.

Common Use Cases Enabled​

  • AI-powered site search with minute-level freshness.
  • Conversational shopping assistants with accurate availability.
  • AI-driven upsell and cross-sell.
  • Real-time AI merchandising insights.
  • Agentic commerce workflows using product-level reasoning.

Summary​

By using Catalog Delta Files as a native feed for OpenAI Product Feeds, Elastic Path enables:

  • Near real-time AI synchronization
  • Lower ingestion and token costs
  • Faster AI index updates
  • More accurate AI-driven commerce experiences

This bridges the gap between transactional commerce systems and real-time AI execution.

Documentation​

Ask External AI