Delectable AI AI-Ready Catalog · Generative Engine Optimization
— Generative Engine Optimization for grocery

The next basket isn't searched.
It's cited.

When a shopper asks ChatGPT "what's a heart-healthy weeknight dinner for four?", the products in the answer aren't picked from a search index. They're picked from whichever brand emitted conversational attributes the LLM could chunk, verify, and cite. Delectable AXP turns every retailer SKU into a cite-able entity — at parity with the brands that already buy this surface area at Amazon.

Research foundation
Princeton & Georgia Tech · KDD 2024
Measured citation lift
Up to +40% Share of Model
Engines we feed
ChatGPT · Perplexity · Gemini · Claude · Google AI Mode
01 · The problem

LLMs don't browse. They chunk.

The Princeton/Georgia Tech KDD 2024 study quantified what every brand discovers eventually: AI answer engines don't crawl the page like Googlebot. They pull 40–60-word "chunks" of structured, machine-readable claims, and they cite whichever ones they can verify across sources. A product page written for human eyes returns nothing. A product entity emitted as conversational attributes returns citations.

Traditional PDP markup

Marketing prose. Invisible to LLMs.

Brand-voice paragraphs, hero images, bullet-list features. Indexable by Googlebot, ignored by RAG-based answer engines.

<!-- product page · marketing copy -->
<h1>Heart-Healthy Olive Oil</h1>
<p>Bring the taste of the
   Mediterranean to your kitchen
   with our premium cold-pressed
   extra-virgin olive oil. Perfect
   for drizzling, dipping, or
   finishing your favorite dishes.</p>
<p>Available in 16oz and 25oz.</p>
✗ Not cited · 0 chunkable claims · 0 verifiable specs
AI-Ready · emitted by AXP

Structured chunks. Cite-able by every major LLM.

Same product, emitted as MerchantListing JSON-LD plus conversational attribute feeds. Verifiable specs, pre-chunked Q&As, machine-readable variants.

// /.well-known/merchant payload
{
  "@type": "MerchantListing",
  "name": "Extra-Virgin Olive Oil 25oz",
  "healthClaims": ["heart_healthy", "low_sat_fat"],
  "variant_option": {
    "size": ["16oz", "25oz"],
    "acidity_pct": 0.3
  },
  "question_and_answer": [{
    "q": "Is it heart-healthy?",
    "a": "Yes — <3g sat fat / tbsp, FDA qualified."
  }]
}
✓ Cited · 6 chunkable claims · verified across 3 sources
02 · The four conversational attribute types

Four shapes the LLM actually reads.

We don't invent these formats. They're the structured shapes Google's Merchant Center already accepts, the JSON-LD schemas every major search engine indexes, and the comparison-table shape LLMs prefer for matrixed reasoning. AXP emits all four, for every SKU, automatically — from the same Food HyperGraph that builds the Perfect Cart.

01 / SCHEMA

MerchantListing JSON-LD

schema.org

Authoritative product entity embedded as JSON-LD in the PDP. Captures price, availability, claims, certifications — the canonical machine-readable shape every major engine indexes.

→ The base entity for cross-engine consensus.
02 / FEED

question_and_answer

merchant.google.com

Pre-chunked Q&A pairs — 40–60-word answers to the questions shoppers ask. Mimics the exact format LLMs cite from. Generated from FAQ data, support tickets, and Food HyperGraph relationships.

→ The most cited surface in agentic answers.
03 / FEED

variant_option

merchant.google.com

Structured variant matrices: size, pack count, dietary certification, regional sourcing. Lets the LLM filter "gluten-free under 10oz, sodium <140mg" without scraping prose.

→ Powers attribute-precise recommendations.
04 / SURFACE

compare.matrix

self-published

Self-comparison tables on the PDP — this SKU vs three alternates — in the table shape LLMs prefer for matrixed reasoning. Includes private-label vs national-brand parity rows.

→ Wins the agentic shootout in the answer.
03 · Live explorer

Pick a SKU. See what we emit. See what ChatGPT cites.

Six representative SKUs. Click one to see the conversational-attribute payload AXP emits, alongside a simulated ChatGPT response that cites the product when a shopper asks an unbranded query. Every cite is grounded in the structured payload — no hallucinations, no guesswork.

Conversational-attribute payload · emitted by AXP

Loading…


        
Cited in the agentic answer
04 · The new metric

From Share of Voice to Share of Model.

The retailer who indexes well for "best low-sodium pasta" on Google still loses the basket if ChatGPT, Perplexity, and Gemini don't cite their SKUs in the agentic answer. Tracking citation rate across major LLMs — Share of Model — is the new equivalent of share of voice for the agentic-checkout era.

+40%
Lift in citation rate from structuring product data into conversational attributes.
Princeton + Georgia Tech · KDD 2024
60% of intent
Of all shopping discovery is projected to be mediated by AI agents within 24 months.
eMarketer · Capgemini 2025
5engines
Read the same /.well-known/merchant manifest. One emission. Cooperative-wide reach.
OpenAI · Perplexity · Google · Anthropic · Gemini
openai.chatgpt perplexity.shop google.ai-mode gemini.shopping anthropic.claude

One emission. Every agent. Every cooperative member.

The same AXP brain that builds the Perfect Cart for each retailer also publishes their cooperative-wide conversational-attribute manifest. Private label gets cite-able. Regional SKUs get discoverable. The middleman tax on agentic discovery disappears.