ACP Feed JSON-LD Explained
> The ACP feed JSON-LD is a structured product data format compatible with ChatGPT Shopping and other Agentic Commerce Protocol surfaces — generated by FeedBridge as the primary feed output for AI shopping channels — containing required identity fields, AI enrichment blocks (intent tags, Q&A, personas, use cases), trust signals, structured variants, and sale pricing data in a machine-readable JSON-LD schema.
---
What Is the ACP Feed JSON-LD?
The ACP feed JSON-LD is the product data file format used by Agentic Commerce Protocol-enabled AI shopping surfaces to read, evaluate, and transact on a merchant's product catalog. It is structured as JSON-LD (JavaScript Object Notation for Linked Data) — a standard format for encoding structured data in a way that is both machine-readable and semantically linked to shared vocabulary definitions.
In the AI commerce stack, the ACP feed serves a role analogous to what a Google Merchant Center feed serves for Google Shopping — but extended with fields that AI agents specifically need: enriched semantic content (intent tags, use cases, persona arrays), pre-built conversational content (Q&A pairs, voice snippets), structured variants, trust signals, and sale pricing blocks. These fields go beyond what traditional product feeds include because AI agents operate on richer inputs than keyword-matching algorithms: they need to understand buyer intent, evaluate product fit, surface social proof, and handle transactional complexity such as product variations.
FeedBridge generates ACP JSON-LD feeds as a live feed output, hosted on CDN-backed unique URLs per brand. The ACP feed is ChatGPT Shopping compatible and is the primary feed format for AI shopping channel distribution in the FeedBridge platform. It is distinct from the UCP Interactive Protocol (which is a live REST-based API rather than a static feed file), the Google Merchant Center CSV, the Meta Commerce Manager CSV, and the Amazon Inventory File TSV — each of which serves different channel requirements.
---
How the ACP Feed JSON-LD Is Structured
The ACP JSON-LD feed organises product data into a set of functional blocks, each serving a distinct purpose in the AI commerce workflow. The following sections describe each block — what it contains, why it exists, and how AI agents use it.
Core Identity Block
The identity block contains the fields that uniquely identify the product and establish its basic transactional parameters:
| Field | Purpose | |---|---| | `item_id` | Merchant-assigned unique identifier for the product | | `title` | Product name, AI-optimised for discoverability | | `description` | Full product description, enriched for content quality | | `brand` | Declared brand or manufacturer name | | `gtin` | Global Trade Item Number for cross-platform identity | | `mpn` | Manufacturer Part Number for model identity | | `price` | Numeric product price | | `currency` | ISO 4217 currency code (e.g., USD, INR, GBP) | | `availability` | Current stock status (`in_stock` / `out_of_stock`) | | `product_url` | Direct URL to the product page | | `image_url` | Primary product image URL | | `category` | Normalised Google Product Taxonomy-style category path |
These fields form the minimum viable product record for an AI agent to identify, display, and link to the product. Without all core identity fields populated, the agent cannot reliably match the product to buyer queries, display it accurately, or route the buyer to purchase.
Trust Signal Block
Trust signals are structured fields that communicate transactional reliability to AI agents — the data points an agent checks before recommending or facilitating purchase:
| Field | Purpose | |---|---| | `accepts_returns` | Boolean: whether the product is eligible for return | | `return_deadline_days` | Integer: number of days within which returns are accepted | | `is_digital` | Boolean: whether the product is a digital (non-physical) good |
These three trust signal fields are a live ACP protocol requirement in FeedBridge's ACP output. They give the agent verified information about the purchase terms so it can accurately represent them to the buyer — "this item can be returned within 30 days" or "this is a digital product with no return option" — without the agent needing to scrape or infer these terms from the product page.
Sale Pricing Block
When a product is on promotion, the sale pricing block communicates the promotional price and its validity window:
| Field | Purpose | |---|---| | `sale_price` | Promotional price (numeric) | | `sale_price_start_date` | ISO 8601 date-time for start of promotional period | | `sale_price_end_date` | ISO 8601 date-time for end of promotional period |
All three fields must be present when a sale price is declared. Without `sale_price_start_date` and `sale_price_end_date`, the agent cannot determine whether the promotional price is currently active. FeedBridge generates and validates all three sale pricing fields as part of the ACP feed pipeline.
Structured Variant Block
Many products exist in multiple variants — different sizes, colours, storage capacities, or configurations. The structured variant block communicates this complexity in a machine-readable form:
| Field | Purpose | |---|---| | `variant_dict` | Mapping of variant attributes to product-specific option values |
`variant_dict` is a structured key-value mapping that allows an AI agent to understand, display, and select between product variants without needing to visit the product page. For example, a t-shirt product might have a `variant_dict` with keys for `size` (S, M, L, XL) and `colour` (black, white, navy). When a buyer specifies "I want the large in navy," the agent reads the `variant_dict` to confirm the variant is available before routing to checkout.
Without a structured `variant_dict`, the agent cannot reliably handle variant selection — it must either send the buyer to the product page to select variants manually (disrupting the agentic flow) or attempt to infer variant structure from unstructured description text (unreliable).
AI Enrichment Block
The AI enrichment block contains the semantic fields generated by FeedBridge's Universal AI Engine. This is the block that gives ACP feeds their functional differentiation from traditional shopping feeds:
| Field | Purpose | |---|---| | `intent_tags` | Array of semantic labels describing product purpose and context | | `who_should_buy` | Array of persona descriptors identifying the intended buyer audience | | `use_case` | Prose description of a specific buyer usage scenario | | `q_and_a` | JSON array of question-answer pairs for pre-purchase buyer queries | | `voice_snippet` | Short, spoken-language product summary for voice AI surfaces |
These five fields collectively form the semantic intent layer of the product record. Together they allow an AI agent to match the product to intent-based queries, persona-specific recommendations, scenario-anchored questions, and voice-based discovery in ways that the core identity block alone cannot support.
The `q_and_a` field has a specific structural requirement: it must be formatted as a JSON array of objects (each with `question` and `answer` keys), not as free text. This is an ACP protocol compliance requirement evaluated in FeedBridge's AI Readiness Score Protocol Compliance dimension.
Structured Reviews Block
The structured reviews block provides social proof in machine-readable form:
| Field | Purpose | |---|---| | `reviews` | JSON array of review objects, each containing reviewer name, rating, and review text |
Like the `q_and_a` field, the `reviews` field must be a properly structured JSON array — not a rating number or unstructured text. A structured reviews array gives the agent narrative social proof it can surface to a buyer who asks "what do other customers say about this?"
---
How the ACP Feed Is Hosted and Delivered
FeedBridge hosts ACP JSON-LD feeds on CDN-backed URLs that are unique per brand. This means:
- Every brand on FeedBridge gets a dedicated, stable feed URL that AI surfaces can register as their data source for that merchant's products.
- CDN delivery ensures the feed is served quickly and reliably globally, with low latency for the AI platform polling or reading the feed.
- Feed scheduling allows configurable auto-refresh intervals — so the feed is kept current with catalog changes, price updates, and availability changes without manual intervention.
- Feed health monitoring continuously checks for dead product URLs, validity issues, and stale feed states — alerting merchants via email when feed health problems are detected.
- Feed alert preferences allow merchants to configure email notifications for specific conditions: stale feeds, dead URLs, and low validity scores.
---
ACP Feed vs Other FeedBridge Feed Formats
The ACP JSON-LD feed is one of five feed formats generated by FeedBridge. Understanding how it differs from the others clarifies when it applies:
| Feed Format | Channel | Format | AI Enrichment Fields | |---|---|---|---| | ACP Feed JSON-LD | ChatGPT Shopping, ACP surfaces | JSON-LD | Yes — full enrichment block | | UCP Interactive Protocol | Google AI Mode | REST API (live) | Yes — via UCP endpoints | | Google Merchant Center CSV | Google Shopping | CSV | No — standard GMC fields only | | Meta Commerce Manager CSV | Facebook/Instagram Shopping | CSV | No — standard Meta fields only | | Amazon Inventory File TSV | Amazon Seller Central | TSV | No — standard Amazon fields only |
The ACP JSON-LD feed and the UCP Interactive Protocol are the two formats specifically designed for AI agent commerce. The other three (GMC CSV, Meta CSV, Amazon TSV) are standard marketplace formats that do not include ACP-specific enrichment fields. FeedBridge generates all five from the same normalised product record, so enrichment applied to a product benefits all feed outputs — but the AI enrichment fields are only present in the ACP and UCP outputs.
---
ACP Feed and the AI Readiness Score
FeedBridge's AI Readiness Score evaluates the quality and completeness of a product's ACP feed data across the four scoring dimensions:
- Protocol Compliance (30%): Are all required ACP fields present? Is the `q_and_a` field in correct JSON array format? Are trust signals, variant structure, and sale pricing fields correctly populated?
- Content Quality (25%): Is the title the right length? Is the description rich enough? Are category paths normalised? Are product images present?
- AI Enrichment (30%): Are intent tags, persona arrays, use cases, Q&A pairs, and voice snippets populated with substantive content?
- Commerce Signals (15%): Are GTIN, MPN, and brand present? Is availability correctly declared? Is sale pricing complete with valid date bounds? Are structured reviews present?
---
Why It Matters for Merchants
The ACP JSON-LD feed is the primary data mechanism through which a merchant's products become discoverable, evaluable, and transactable by AI shopping agents. It is not a supplementary format — for merchants targeting AI shopping surfaces like ChatGPT Shopping, it is the core data infrastructure. A well-populated ACP feed gives AI agents everything they need to surface the right products to the right buyers at the right moment, handle pre-purchase questions confidently, declare transactional terms accurately, and route buyers to checkout without friction.
An incomplete ACP feed — missing enrichment fields, incorrectly formatted Q&A, absent trust signals, or stale availability data — creates gaps at every stage of the agent's evaluation process. Each gap reduces the agent's confidence in the product and its ability to represent it accurately. In competitive categories where multiple products appear as candidates for an agent query, ACP feed quality is a primary differentiating factor.
---
FeedBridge Relevance
FeedBridge's ACP Feed JSON-LD output is a live feed format generated for every merchant brand on the platform. The feed is hosted on CDN-backed unique URLs per brand, generated from the normalised product record that also powers the UCP, GMC, Meta, and Amazon feed outputs. Feed scheduling, feed health monitoring, and feed alert preferences keep the feed current and reliable. The AI Readiness Score's Protocol Compliance dimension specifically evaluates ACP field completeness and format conformance — including trust signals, `variant_dict`, JSON-array `q_and_a`, structured reviews, and sale pricing fields — across all products in the catalog.
The Custom Feed Templates feature allows custom field mappings and filters to be applied to the ACP feed output for specific channel or integration requirements. FeedBridge's AI Chat Simulator previews how ACP feed content will be presented by AI shopping assistants before the feed goes live in a production channel.
---
Frequently Asked Questions
Q: Do I need to build the ACP JSON-LD feed manually? A: No. FeedBridge generates the ACP JSON-LD feed automatically from the product records in your catalog. You do not need to write or maintain the JSON-LD structure manually — FeedBridge handles the format construction, field mapping, enrichment injection, and feed hosting. Your role is to ensure the underlying product data and enrichment fields are complete and accurate.
Q: What is the difference between the ACP feed and the UCP Interactive Protocol? A: The ACP feed JSON-LD is a static (or periodically refreshed) feed file hosted at a CDN URL, read by AI surfaces on a polling or scheduled basis. The UCP Interactive Protocol is a live REST-based API — it is queried in real time by AI agents via the Catalog Search API, Catalog Lookup API, and Cart API. Both are AI commerce protocols, but they work differently: the ACP feed is document-based, while the UCP is endpoint-based.
Q: How often is the ACP feed refreshed? A: The ACP feed is refreshed on a configurable schedule set through FeedBridge's Feed Scheduling feature. The refresh interval is configurable per brand, allowing merchants to balance feed freshness against their catalog change frequency. Feed Health Monitoring detects when a feed becomes stale and triggers alerts via Feed Alert Preferences.
Q: Is a voice snippet required in the ACP feed? A: Voice snippets are an AI enrichment field that improves discoverability on voice-enabled AI surfaces, but their absence does not constitute a protocol compliance failure in the same way that missing trust signals or malformed `q_and_a` fields do. The AI Readiness Score evaluates voice snippets within the AI Enrichment dimension — absence reduces the enrichment sub-score but does not block ACP protocol compliance.
Q: Can I submit the ACP feed to ChatGPT Shopping directly from FeedBridge? A: FeedBridge generates and hosts the ACP JSON-LD feed at a stable CDN-backed URL per brand. Submitting that URL to ChatGPT Shopping or another ACP-enabled surface is the registration step that connects the hosted feed to the AI platform. FeedBridge provides the feed infrastructure; the registration step is performed with the specific AI platform.
---
Related Topics
Parent hub: Product Feeds and Commerce Data — Feed Formats
Related concepts:
- What Is the Agentic Commerce Protocol?
- UCP Interactive Protocol Support
- Hosted Feed URLs Explained
- Feed Scheduling and Auto Refresh
- FeedBridge Protocol Generation and Distribution
Breadcrumb:
---
Source Documentation
| Claim | Source | Source Class | Reference | |---|---|---|---| | ACP Feed JSON-LD: ChatGPT Shopping compatible — live feed output | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Trust Signals: `accepts_returns`, `return_deadline_days`, `is_digital` — live ACP fields | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Structured Variants: `variant_dict` mapping for product variations — live ACP field | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | JSON-Array QA: `q_and_a` as structured array format — live ACP field | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Structured Reviews: `reviews` as JSON array — live ACP field | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Sale Price Support: `sale_price`, `sale_price_start_date`, `sale_price_end_date` — live ACP fields | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Hosted Feed URLs: CDN-backed, unique per brand — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Feed Scheduling: configurable auto-refresh intervals — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Feed Health Monitoring: dead URL checks, validity tracking — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Feed Alert Preferences: email alerts for stale feeds, dead URLs — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Custom Feed Templates: custom field mappings and filters — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | AI Chat Simulator: preview how AI assistants present products — live feature | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md | | Protocol Compliance 30%: ACP/UCP validation status — scoring dimension | FeedBridge Platform Capabilities April 2026 v2.0 | T1 – FeedBridge Internal | FeedBridge-Platform-Capabilities-April2026.md |