A Devastating Study Finds Your JSON-LD Is Almost Useless. Here’s the Fix.
You're implementing JSON-LD all wrong. A shocking academic study found it's almost useless on its own. Here’s the foundational fix AI engines are actually looking for.
OpenFound Team
Content Team
You’ve spent months adding JSON-LD to every corner of your website. You’ve been told it’s the golden ticket for AI visibility. Now for the bad news: on its own, it’s probably doing nothing.
That’s not an opinion. It’s the conclusion of a startling 2026 academic study on AI-driven search. Researchers conducted a controlled experiment and found that simply adding JSON-LD structured data to standard HTML pages had a “negligible” impact on the accuracy and completeness of answers generated by AI systems. The measured effect size was a minuscule 0.18—a statistically insignificant bump for all your hard work.
For years, the SEO industry has been shouting from the rooftops: “Add Schema! Add JSON-LD!” But this study exposes a fatal flaw in that advice. Most brands are meticulously applying a finishing coat of paint to a house with no foundation. The AI can’t see the paint because it can’t even recognize the house. This is the core of Generative Engine Optimization (GEO), and it's what separates winning brands from the ones disappearing from AI overviews.
The Real Problem: Structure Is Prioritized Over Keywords
The issue isn't that JSON-LD is intrinsically useless. The problem is that it’s being treated as a magic bullet. As Google’s own shift in February 2026 showed, its AI systems now prioritise structure over keywords. But what does “structure” actually mean?
It doesn’t mean a block of JSON-LD script copy-pasted at the bottom of a page. It means the entire document is built with clear, machine-readable semantics from the ground up.
"“If your templates still lack solid HTML semantics, don’t jump straight to JSON-LD: fix the foundations first.” - Andrea Volpini, Search Engine Journal"
This is where most websites fail. In a rush to create visually appealing sites, especially with component-based frameworks like React or Vue and WordPress themes, developers often create a “div soup”—a mess of generic
Semantic HTML is the foundational layer for AI content understanding; JSON-LD is the semantic accelerator.
This isn't just theory. A recent audit of 100 small business WordPress sites found that many themes prioritize aesthetics over clear HTML structure, making key details invisible to AI. For example, a bakery’s address might be in a pretty image in the footer but lack proper LocalBusiness schema or even basic semantic tags. As a result, it never appears in “best bakeries near me” AI results. Fixing the foundation is the first and most critical step in any modern AIO strategy at OpenFound.
The 2-Layer Fix: How to Make Your Content AI-Visible
To get cited and seen in 2026 and beyond, you need to stop thinking about JSON-LD as an add-on and start thinking about semantics as your core infrastructure. This requires a two-layer approach.
Layer 1: The Semantic HTML Bedrock
Before you write a single line of JSON-LD, you must ensure your HTML is impeccably structured. This is the non-negotiable foundation. It tells AI what each part of your page is.
- Wrap your content correctly: Is your blog post wrapped in an
tag? Is your navigation in a - Enforce heading hierarchy: Use only one
per page for the main title. Follow it with,, and so on. Never skip levels (e.g., anfollowed by an). AI uses headings to understand the document's outline. - Use specific tags: Use the
tag with adatetimeattribute for dates (e.g.,). Wrap images and charts inwith a. - Audit your components: For those using React, Vue, or Svelte, “Frameworks give us power. Semantic HTML keeps us grounded.”. Look inside your reusable components. Is that
Cardcomponent just a series ofs, or does it use,, and other semantic elements?
Getting this right isn’t just an AIO play. With the European Accessibility Act (2025) and upcoming WCAG 3.0 standards, semantic HTML is becoming a legal requirement. It future-proofs your work against regulatory change and makes your site usable for everyone.
Layer 2: The JSON-LD Accelerator
Once your semantic bedrock is in place, now you can add JSON-LD. This script acts as an executive summary for the AI, confirming and clarifying the structure you’ve already built. This is how you get results.
The aforementioned audit of WordPress sites proved this model. After fixing basic HTML, implementing just three key schema types (LocalBusiness, Service, and FAQPage) improved AI understanding by an average of 35%. The JSON-LD worked because the foundation was solid. For more ideas check out the OpenFound blog.
- Be Specific and Relevant: Don’t just add
Organizationschema. If it’s a blog post, useArticleorBlogPosting. If it’s a recipe, useRecipe. If it’s a product page, useProductand nestOfferandReviewschemas within it. - Match Visible Content: The arXiv study on structured data highlighted a critical ethical point: the data in your JSON-LD must describe the exact same information visible to human users on the page. Any mismatch is a red flag for AI and can erode trust.
- Use Google’s Recommended Format: Google prefers JSON-LD placed in a
The Takeaway: Stop Decorating, Start Building
The reason your JSON-LD is having a negligible impact is that you’re treating it like a decoration. AI overviews are not built on decoration; they are built on robust, machine-readable architecture.
The 2026 study was not an indictment of JSON-LD. It was an indictment of how it’s being used. When the researchers combined JSON-LD with more advanced agentic retrieval systems and properly structured entity pages (the C6 condition in their experiment), performance skyrocketed. This proves the model: Solid HTML Foundation + Targeted JSON-LD = AI Visibility.
Stop blindly chasing schema. Fix your foundations first. Treat semantic HTML as the bedrock of your content strategy and JSON-LD as the accelerator. That is how you stop being invisible to AI and start winning in the new era of generative search.
What is the difference between semantic HTML and JSON-LD?
Semantic HTML (like
,
,
Does JSON-LD still matter for AIO after this study?
Yes, absolutely. The study didn't say JSON-LD is useless; it said it's useless on its own when applied to poorly structured pages. When layered on top of a solid semantic HTML foundation, JSON-LD is a powerful accelerator that significantly improves how AI understands and trusts your content.
What are the most important schema types for AI?
This depends on your business, but common high-impact types include Article/BlogPosting for content, FAQPage for Q&A sections, HowTo for guides, LocalBusiness for physical locations, and Product (with nested Offer and Review) for e-commerce. A recent audit showed implementing just LocalBusiness, Service, and FAQPage improved AI understanding by 35% for small businesses.
How can I check my website's semantic HTML?
You can start by viewing your page source ('View > Developer > View Source' in Chrome) and looking for semantic tags like
,
,
, and a logical heading (
,
) structure. For a deeper analysis, use browser developer tools to inspect the DOM tree and identify areas that rely heavily on generic
tags instead of semantic ones.
Why is semantic HTML also important for accessibility?
Semantic HTML is crucial for accessibility because it provides context for assistive technologies like screen readers. Tags like
Continue reading
Your Robots.txt Is Creating An AI Misinformation Crisis
New data reveals a terrifying truth: by blocking AI crawlers, reputable brands are accidentally feeding LLMs a diet of pure misinformation. Here is the fix.
AI Ignores 75% of Websites. The Fix Is A 4-Hour Technical Sprint.
A shocking 75% of sites have zero Schema.org metadata, making them invisible to AI. Learn the 4-hour technical fix to get cited by Gemini and Perplexity.
AI Doesn't Trust Your Website. It Trusts This Architecture Instead.
78% of tech leaders are rebuilding their sites for AI. While you're optimizing keywords, they're re-architecting their sites to be trusted sources for Generative Engines.