Case Study: Implementing schema.org / JSON-LD for GEO (AI Search) in B2B and eCommerce

Case Studies Software Development Google Cloud & Vertex AI eCommerce and Online B2B Data and Business Intelligence

INTRODUCTION

SEO optimization in 2026 (Web 3.0) means more than just keywords and link building. It requires semantic clarity through graphs: for Google, but also for AI search agents (AI Overviews, ChatGPT, Claude, etc.). This phenomenon is known as GEO (Generative Engine Optimization) or AI Search, and the challenge lies in server-side generation, not WordPress plugins.

OPTI implemented schema.org / JSON-LD for three projects: a B2B portal (opti.ro) and two e-commerce websites, treating the schema as a reusable semantic graph, generated server-side, validated, and continuously monitored.

Within two months, results included an increase in non-brand queries, faster and more stable indexing, the appearance of breadcrumbs in SERPs, and the first AI Overviews on e-commerce.

CHALLENGES

Without a schema.org implementation, the three websites had slow indexing (mostly for long-tail queries), had few AI summaries, and heavy content creation had lower-than-expected ROI.

1. Classic SEO vs. GEO: same content, different asks

In classic SEO, a good title and a coherent H1 help. In GEO, AI systems look for who is speaking (publisher/author), what the topic is (product/service/article), what evidence exists (data/links/entities/offers), and how they connect semantically (graph).

2. Identifier (ID) consistency and deduplication

Schema must be free of contradictions. If an organization appears in hundreds of places, it must be the same entity. OPTI.ro uses a stable core https://www.opti.ro#organization, https://www.opti.ro#website and links the rest of the pages back to it.

3. Multilingual capabilities and AI context

When a website has RO/EN versions, it must explicitly declare its language versions to avoid ambiguity. OPTI.ro uses inLanguage with ro-RO and en-US across all pages.

4. E-commerce data is abundant but fragile

Prices, stocks, product lists, and sometimes even product details frequently change. The schema must be continuously updated to match Google's requirements.

"Schema.org is the semantic link between the website and search engines or AI. The difference between a good and a poor implementation is graph consistency: the exact same entities, the same IDs, clear relationships, and verifiable data."

Daniel Curculescu, Software Engineer

SOLUTION

GRAPH-FIRST APPROACH

We treated the schema as a reusable semantic graph, implementing it across three layers for both B2B and the two e-commerce websites:

Brand graph with stable entities
- Comprehensive Organization: name, legalName, alternateName, logo, contactPoint, served regions, sameAs etc.
- WebSite: publisher, inLanguage, actions, subscribe etc.
- WebPage for Homepage: description, images, publisher.
- Essential for GEO: identifying the source and how it can be verified.
Offer graph featuring services and products
- Service pages modeled as Service (see OPTI.RO): additionalProperty highly useful for AI as it compacts technical capabilities into a standardized format; subjectOf pointing to case studies, an explicit verification mechanism (the service is backed by real examples).
- Software product pages as SoftwareApplication: offers with commercial packages, featureList, rating; plus HowTo and FAQPage which allow AI agents to deliver semantic answers.
Editorial graph for authority and citations
- NewsArticle: author, publisher, media (Images, Video/PDF).
- TechArticle: mentions, about, proficiencyLevel.
- Hierarchical links: hasPart / isPartOf (with sub-articles).
- CreativeWorkSeries for article series or guides. Example OPTI.ro: AI Architecture for B2B 2026.
- Articles are linked to entities, products, and services.
- Series become semantically navigable.
- Subchapters are part of a larger whole and vice versa.

Treating schema as a semantic graph (not isolated markup) is the difference between a functional implementation and one with real impact. When entities are connected, consistently identified, and server-side generated, AI can verify and cite the information, just as a careful human reader would.

ARCHITECTURE: FOUR REPEATABLE STEPS

The schema.org / JSON-LD implementation followed four repeatable standard steps across all three projects:

Data modelling
- Entity inventory: defining what is an Organization, Service, Product/Book, Article/TechArticle, Series. Establishing what relationships exist: publisher, author, mentions, subjectOf, isPartOf.
Server-side JSON-LD generator
- Google recommends JSON-LD as a format because it is easier to implement and maintain. The schema is generated on the fly from real data (CMS and product catalog), not written manually.
Templates per page type
- Creating templates for a complex website: homepage, service pages, product pages, category pages, internal search pages, and editorial content pages.
Validation, QA, monitoring
- Syntactic and semantic validation (ensuring marked content exists and is visible), compliant with Google's structured data policies. Search Console tracking (errors, coverage, enhancements), regression testing, and continuous optimization.

Architecture - schema.org JSON-LD implementation

Partial TechArticle schema example: OPTI.RO. See website

RESULTS

AFTER 2 MONTHS: B2B AND E-COMMERCE

Increase in non-brand B2B queries

OPTI.RO now ranks for queries like "HubSpot ERP integration", "Google Cloud partner Romania", "Vertex AI Search for commerce".

Fast and stable B2B indexing

Service pages and long-form content (guides / case studies) were indexed immediately, bringing in news/event-type traffic spikes.

+~70% Clicks and impressions in GSC

Driven by new breadcrumbs and rich product details appearing in SERPs, compared to the three months prior (with simultaneous UX improvements).

Increase in listing page indexing

A significant drop in unindexed or not-served pages. Deep categories and filters became stably indexed.

Appearance in AI Overviews

Structured data does not guarantee rich results or AI Overviews, but it enables AI search engines to correctly understand the website.

+~20% Purchase rate and first acquisition

Google Analytics: Scale-up on e-commerce site, with simultaneous UX improvements.

Google Search Console – review snippets and translated results (B2B site) — **Google Search Console: Availability of review snippets and translated results (B2B site)**

Google Search Console – clicks and impressions +70% (e-commerce site) — **Google Search Console: Clicks and impressions grew by ~70% compared to the three months before (E-commerce site, with simultaneous UX remake)**

Google Analytics – purchase rate +20% (e-commerce site) — **Google Analytics: Scale-up: purchase rate and the rate of first acquisition grew by ~20% (E-commerce site, with simultaneous UX remake)**

TECH AND METHODOLOGY

Services: Technical SEO audit, schema design (graph-first), JSON-LD implementation, validation, regression testing, Search Console / indexing monitoring, internal guidelines and team training.
Tech Stack: Schema.org, JSON-LD, Google Search Console, Rich Results Test, schema validators, CMS platforms, e-commerce (nop, custom), Server-side scripting (Node.js, PHP), automated QA.
Methodologies and best practices: JSON-LD fully compliant with Google's structured data recommendations. The schema strictly reflects visible content, in line with Google's policies.
Standards: ISO 9001 (quality), ISO 27001 (information security).

Quick Questions

What is GEO and why does it matter for my website?

GEO (Generative Engine Optimization) means optimizing your site to be understood and cited by AI systems (AI Overviews, ChatGPT, Claude etc.), not just classic search engines. Schema.org / JSON-LD is the core technical element of this optimization.

Why is server-side schema generation important?

Server-side generation ensures the schema is stable, easy to test, and always synchronized with real data (CMS, catalog). WordPress plugins or client-side solutions can introduce inconsistencies or delays.

How long until the first visible results?

In all three projects, the first effects (faster indexing, breadcrumb appearance, growth in non-brand impressions) were visible within 4–8 weeks of implementation.

Does schema.org guarantee appearance in AI Overviews?

No. Structured data does not guarantee rich results or AI Overviews, but it enables AI search engines to correctly understand the site and significantly increases the probability of being cited.

What is the TLDR (conclusion)?

What technologies and methodologies are involved?

Technologies: Schema.org, JSON-LD, Google Search Console, Rich Results Test, schema validators, CMS platforms, e-commerce, Server-side scripting, automated QA
Methodologies: Technical SEO audit, schema design (graph-first), JSON-LD implementation, syntactic and semantic validation, regression testing, Search Console / indexing monitoring, internal guides and team training

How can I find out more?

Article written by

Daniel Curculescu

Data Engineer (Google Cloud Certified Professional). CRM Automation, AI, mobile.

See on LinkedIn →

70% Traffic Growth via UX Refactoring and Technical SEO

Case study on SEO and UX optimization for e-commerce.

AI-Ready Infrastructure (Web, Email, Sales)

How to prepare your digital infrastructure for the AI era.

AI Integration Guides for B2B Workflows in 2026

Practical guides for implementing AI in B2B processes.

Want your website to be AI-ready?

→ Easy to understand and widely cited in modern search results.

Let's talk

Case Study: Implementing schema.org / JSON-LD for GEO (AI Search) in B2B and eCommerce

NEW March 2026

INTRODUCTION

CHALLENGES

SOLUTION

ARCHITECTURE: FOUR REPEATABLE STEPS