OpenAI Product Feed · 2026-02-05

OpenAI Product Feed Requirements Update: Clear Breakdown of New, Renamed, Changed, Removed, and Unchanged Fields

OpenAI has significantly updated its product feed specification for ChatGPT Shopping. This article provides a clear, field-by-field breakdown of what is new, renamed, changed, removed, and unchanged, with the exact new requirements for each attribute.

picture of Jordan STYCZEN
Written by Jordan STYCZEN
Last updated on 2026-02-05
...

OpenAI recently updated its Product Feed Specification used for ChatGPT Shopping and AI-powered commerce experiences.

This update is not incremental.
It is a structural clarification of how products must be described so that AI systems can index, understand, compare, make them appear in ChatGPT and potentially sell them.

To remove ambiguity, this article lists all specification changes, grouped by status:

  • New fields
  • Renamed fields
  • Changed fields (same name, new logic or validation)
  • Removed fields
  • Unchanged fields (still required or recommended)

Each field is listed, with its new requirement explicitly stated.


1. New Fields

These fields did not exist previously and are now part of the official specification.

  • accepts_exchanges (Optional)
  • accepts_returns (Optional)
  • dimensions_unit (Optional – required if any dimension is provided)
  • is_digital (Optional)
  • item_weight_unit (Optional – required if weight is provided)
  • listing_has_variations (Required)
  • marketplace_seller (Optional)
  • return_deadline_in_days (Optional)
  • sale_price_start_date (Optional – ISO 8601 date)
  • sale_price_end_date (Optional – ISO 8601 date)
  • shipping_price (Optional – structured format: country:region:service_class:price)
  • store_country (Required – ISO 3166-1 alpha-2)
  • target_countries (Required – list of ISO 3166-1 alpha-2)
  • variant_dict (Optional – JSON object with string values)

These additions introduce explicit signals around eligibility, geography, fulfillment, trust, and variants, removing the need for AI systems to infer intent.


2. Renamed Fields

These fields already existed but were renamed.
Only the new name is valid. Legacy names are not aliases.

  • item_id (Required, replaces id)
  • image_url (Required, replaces image_link)
  • url (Required, replaces link)
  • group_id (Required when variants exist, replaces item_group_id)
  • is_eligible_search (Required, replaces enable_search)
  • is_eligible_checkout (Required, replaces enable_checkout)
  • additional_image_urls (Optional, replaces additional_image_link)
  • model_3d_url (Optional, replaces model_3d)
  • video_url (Optional, replaces video_link)
  • review_count (Optional, replaces product_review_count)
  • star_rating (Optional, replaces product_review_rating)
  • reviews (Recommended, replaces raw_review_data)
  • store_star_rating (Optional, replaces store_review_rating)
  • return_deadline_in_days (Optional, replaces return_window)

Renames are strict.
Using the old field names does not produce warnings, the data is ignored.


3. Changed Fields (Same Name, New Logic or Validation)

These fields keep the same name but now follow stricter rules, new formats, or additional dependencies.

  • brand (Now strictly required)
  • gtin (Optional, but now logically linked with mpn)
  • mpn (Optional, but now logically linked with gtin)
  • height (Optional, now requires unit context)
  • length (Optional, now requires unit context)
  • width (Optional, now requires unit context)
  • weight (Optional, now requires item_weight_unit)
  • material (Optional, stricter text validation)
  • product_category (Optional, stricter taxonomy validation)
  • q_and_a (Recommended, now a JSON list of { "q": string, "a": string })
  • sale_price (Optional, must be strictly lower than price)
  • seller_privacy_policy (Required if is_eligible_checkout is true)
  • seller_tos (Required if is_eligible_checkout is true)

These changes introduce cross-field logic, not just format checks.


4. Removed Fields

These fields are no longer part of the specification and should not be used.

  • applicable_taxes_fees
  • inventory_quantity
  • shipping
  • sale_price_effective_date (replaced by sale_price_start_date + sale_price_end_date)
  • product_review_count
  • product_review_rating
  • raw_review_data

Using removed fields has no effect and may create false confidence.


5. Unchanged Fields (Still Important)

These fields remain valid, with unchanged requirements, and continue to play a critical role.

  • availability (Required)
  • availability_date (Required if availability=pre_order)
  • title (Required)
  • description (Required)
  • price (Required)
  • image_url (Required)
  • seller_name (Required / Display)
  • seller_url (Required)
  • return_policy (Required)
  • color (Optional)
  • condition (Optional)
  • age_group (Optional)
  • age_restriction (Recommended)
  • delivery_estimate (Optional)
  • dimensions (Optional)
  • expiration_date (Optional)
  • geo_price (Recommended)
  • geo_availability (Recommended)
  • item_group_title (Optional)
  • offer_id (Recommended)
  • pickup_method (Optional)
  • pickup_sla (Optional)
  • pricing_trend (Optional)
  • related_product_id (Recommended)
  • relationship_type (Recommended)
  • size (Recommended – apparel)
  • size_system (Recommended – apparel)
  • store_review_count (Optional)
  • popularity_score (Recommended)
  • unit_pricing_measure / base_measure (Unchanged)
  • warning / warning_url (Unchanged)

“Unchanged” does not mean “less important”.
These fields remain the foundation of product understanding.


6. What This Update Really Means

This update confirms a clear direction:

Product feeds are becoming formal contracts for AI eligibility, not best-effort datasets.

AI systems:

  • do not infer missing data
  • do not guess intent
  • do not warn when data is ignored

If a field is missing, misnamed, or invalid, the product may simply not exist in the AI layer.


7. Feed Auditing Is Now a Requirement

With this level of complexity, manually keeping a feed compliant is no longer realistic.

SharpFeed’s backend logic has already been fully adapted to the new OpenAI product feed specification, including:

  • all new fields
  • all renames
  • all removals
  • all cross-field dependencies
  • all logical validations

SharpFeed is ready to audit your feed against the latest requirements and clearly show what needs to be fixed.

⚡ Final takeaway

The OpenAI product feed update is not just a spec change.
It is a structural shift toward agentic, rule-based commerce.

Understanding the changes is step one.
Validating your feed against them is step two.

Product feeds become a true infrastructure and a pillar for understanding an e-commerce catalogue for LLM.

Sources :
OpenAI Product Feed Specs