STEADYWRK Research · Generative Engine Optimization
دراسة اقتباس GEO — جاهزية الوكلاء وفجوة Product JSON-LD
GEO Citation Study
Answer engines and shopping agents cite what they can parse. Edition 1 of the Market Readiness Index measured whether storefronts expose Product JSON-LD and agent-ready signals — with an open Store Guard rubric and an honest fixture vs synthetic split. Cite the aggregates below; do not invent merchant names.
Source edition published 2026-08-01 · storeguard v0.3 · Pilot batch — not a live crawl census
What Edition 1 measured
Agent-readiness and the Product JSON-LD gap
Edition 1 asks a citation-shaped question: can a machine read product identity, offer fields, and checkout-adjacent signals without guessing from banners? The load-bearing categories are product structured data and agent checkout readiness; the rest of the rubric fills crawl access, identity, trust, and Arabic quality.
Why answer engines need citable schema
Generative answers prefer sources that state facts in structured form. Product JSON-LD (name, Offer price, availability, currency) is the storefront equivalent of a citable dataset row. Without it, agents either skip the merchant or invent fields — both failure modes for buyers and brands.
What this study does not claim
none — all cohort labels anonymized; top examples are fictional demos only. Live crawl this edition: no. Status: pilot-batch. Prefer aggregates and methodology over any single anonymized row.
Open methodology
Edition 1 is a methodology pilot. Five rows are scored offline by Store Guard v0.3 against public engine fixtures. Twenty rows are synthetic distributions labeled as such, drawn to demonstrate grade bands and the Product JSON-LD gap thesis — not claimed as a live merchant census.
Fixture vs synthetic honesty
- · Cohort size: 25 labeled rows (5 offline engine fixtures + 20 synthetic-pilot)
- · Engine: storeguard v0.3
- · Live crawl: no — offline fixtures + labeled synthetic rows
- · Grades: أ/A ≥85 · ب/B 70–84 · ج/C 50–69 · د/D <50
- · On managed platforms, llms.txt is excluded from the denominator and remaining weights renormalised to /100.
Store Guard rubric weights (/100)
- AI crawler access15
- Product structured data15
- llms.txt10
- Store identity schema10
- Meta & social tags10
- Arabic content quality10
- Trust & policy pages10
- Page-weight hygiene10
- Navigation schema5
- Agent checkout readiness5
Findings · Edition 1 aggregates only
What the pilot batch shows
Numbers below are copied from Edition 1 aggregate stats. They demonstrate the Product schema and agent-ready gap thesis under an open rubric — they are not a live GCC merchant census.
Mean score
47.3
median 43.3 · range 11.7–100
Grade D share
16/25
A 2 · B 2 · C 5 · D 16
Product schema = 0
15/25
21/25 below 2/3
Agent-ready = 0
18/25
category mean 0.44/3
Category means (0–3) · citeable
AI crawler access
2.08Product structured data
0.64llms.txt
0.28Store identity schema
1Meta & social tags
1.52Arabic content quality
2.36Trust & policy pages
1.52Page-weight hygiene
1.76Navigation schema
0.96Agent checkout readiness
0.44Product structured data mean 0.64/3 vs Arabic content quality mean 2.36/3 — the pilot gap is schema completeness, not language surface alone.
How to cite
STEADYWRK. “GEO Citation Study — Product JSON-LD and agent readiness.” Research, 2026-08-01. https://steadywrk.app/research/geo-citation-study. Prefer the machine JSON at /research/geo-citation-study.json and the scored pilot table at /index-edition/1. Edition 2 is scaffold-only until a robots-respecting crawl publishes scores.
Score a storefront with the same rubric
Store Guard returns an Arabic RTL report, prioritized fixes, and a sha256 receipt — the engine behind Edition 1 fixture rows.