AI systems like ChatGPT, Claude, and Perplexity collect web data in multiple phases. To understand whether and how Schema Markup is used, we need to examine these phases more closely:
The AI Data Processing Pipeline
Note: Our tests primarily show Phase 4 (Direct Fetch). Schema Markup could very well be used in Phases 1-3 – especially by Google AI Overviews and Bing Copilot, which have access to search indexes.
To find out whether AI systems use Schema Markup during direct fetch (Phase 4), we conducted comprehensive tests in October 2025. The results are surprising: Current AI chatbots do NOT use JSON-LD Schema Markup during direct retrieval. Instead, they exclusively extract visible HTML content.
Schema Markup Formats: JSON-LD, Microdata, and RDFa
Before we examine how AI systems handle Schema Markup, it's important to understand that there are different formats for implementing structured data. Each format has its own advantages and disadvantages:
JSON-LD
JavaScript Object Notation for Linked Data – The format recommended by Google. It is separated from HTML and is implemented in <script> tags.
✅ Advantages:
- Easy to implement and maintain
- Separates structure from HTML code
- No content duplication necessary
- Google's preferred format in 2025
⚠️ Disadvantages:
- Content may need to be repeated
- Manual updates needed when changes occur
Microdata
HTML specification for nesting structured data within HTML content. Uses attributes like itemscope, itemtype, and itemprop.
✅ Advantages:
- Directly integrated into HTML
- No content duplication
- Historically widespread
⚠️ Disadvantages:
- Can make HTML code cluttered
- More difficult to maintain
- No longer a W3C standard
RDFa
Resource Description Framework in Attributes – An HTML extension that embeds structured data via attributes like vocab, typeof, and property.
✅ Advantages:
- W3C standard
- Flexible for complex data
- Multiple vocabularies possible
⚠️ Disadvantages:
- Steep learning curve
- Can make code cluttered
- Less widespread than JSON-LD
Test Methodology: Do AI Systems Actually Use Schema Markup?
To definitively answer this question, we conducted a practical test with a specially developed test page. Our goal was to determine whether popular AI systems like ChatGPT, Perplexity, and Gemini can extract information from various sources on a web page.
Test Page Design
We created a test page for a fictional product: "searchVIU Premium GEO Bears." The page contains three product variants with pricing information distributed across different sources:
🧪 Test Scenarios in Detail
JSON-LD Tests
Test 1: Blue GEO Bears
Price: €5.99
Format: No Schema (Baseline)
Source: Visible HTML content – Product information displayed in standard HTML that users can see on the page
Goal: Baseline test – all AI systems should be able to find this ✅
Test 2: Unicorn Premium GEO Bears
Price: €12.99
Format: No Schema (JavaScript Test)
Source: JavaScript-rendered content – Prices inserted into the DOM via JavaScript after page load
Goal: Test whether AI systems execute JavaScript or use headless browsers
Test 3: Rainbow Premium GEO Bears
Price: €8.99
Format: JSON-LD (Schema only)
Source: Schema Markup only – Prices contained exclusively in JSON-LD Schema Markup, but not visible anywhere on the page
Goal: Core test – can AI systems extract pure JSON-LD Schema data?
Test 4: Glitter Premium GEO Bears
Price: €15.99
Format: JSON-LD (dynamic via JavaScript)
Source: Schema in JavaScript – JSON-LD dynamically inserted via JavaScript
Goal: Combined test for JavaScript execution AND Schema parsing
Microdata Tests
Test 5: Orange Premium GEO Bears
Price: €7.49
Format: Microdata (hidden with <meta> tags)
Source: Microdata Schema with itemscope, itemtype, itemprop – Price only in <meta> tags, not visible on page
Goal: Can AI systems parse Microdata format? Is hidden Microdata recognized?
Test 6: Cherry Premium GEO Bears
Price: €6.49
Format: Microdata (visible in HTML)
Source: Microdata directly embedded in visible HTML elements
Goal: Comparison: Is visible Microdata recognized better than hidden?
RDFa Tests
Test 7: Mint Premium GEO Bears
RDFa Price: €11.99 | JSON-LD Price: €9.99
Format: RDFa (hidden) + JSON-LD
Source: Double test – both RDFa with <meta> tags (€11.99) and JSON-LD Schema (€9.99) with different prices
Goal: Which format is preferred? Which price is mentioned? Does AI recognize the conflict?
Test 8: Lemon Premium GEO Bears
Price: €5.49
Format: RDFa (visible in HTML)
Source: RDFa with vocab, typeof, property – directly in visible HTML elements
Goal: Can AI systems parse RDFa format? Comparison with Microdata
Experimental Results
Test Protocol: We queried each AI system 10 times to ensure consistency and avoid anomalies. The questions were designed to specifically ask for prices and product details distributed across the various content sources.
What AI Systems Actually See
Test Protocol: Each AI system was queried 5 times with the identical question: "What are the available products on this page and what are their current prices (26.10.2025)?"
Test Date: October 30, 2025 | Test Page: searchviu.com/en/geo-bears/
| Test | Product & Price | Format/Source | ChatGPT | Claude | Gemini | Perplexity (before index) |
Perplexity (after index) |
Google AI Mode (before index) |
Google AI Mode (after index) |
|---|---|---|---|---|---|---|---|---|---|
| Test 1 | Blue GEO Bears €5.99 |
Visible HTML content | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ |
| Test 2 | Unicorn Premium €12.99 |
JavaScript-rendered | ✗ | ✗ | ✓ | ✗ | ✓ | ✗ | ✓ |
| Test 3 | Rainbow Premium €8.99 |
JSON-LD Schema | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Test 4 | Glitter Premium €15.99 |
JSON-LD via JavaScript | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Test 5 | Orange Premium €7.49 |
Hidden Microdata | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Test 6 | Cherry Premium €6.49 |
Visible Microdata | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ |
| Test 7 | Mint Premium RDFa: €11.99 JSON-LD: €9.99 |
Hidden RDFa + JSON-LD (Conflict) | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ |
| Test 8 | Lemon Premium €5.49 |
Visible RDFa | ✓ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ |
🔍 Detailed Insights from the Tests
▶ ChatGPT (GPT-4) - 3/8 prices found (37.5%)
- ✓ Found: €5.99 (Test 1 - Visible HTML), €6.49 (Test 6 - Visible Microdata), €5.49 (Test 8 - Visible RDFa)
- ✗ Not found: €12.99 (Test 2 - JavaScript), €8.99 (Test 3 - JSON-LD), €15.99 (Test 4 - JSON-LD+JS), €7.49 (Test 5 - Hidden Microdata), €11.99/€9.99 (Test 7 - Hidden RDFa/JSON-LD Conflict)
- Conclusion: Parses only visible HTML content, no JavaScript support, completely ignores hidden Schema data
- Found all 7 product names and descriptions correctly
▶ Gemini (Google) - 4/8 prices found (50%) ⭐ BEST RESULT
- ✓ Found: €5.99 (Test 1 - Visible HTML), €12.99 (Test 2 - JavaScript ⚡), €6.49 (Test 6 - Visible Microdata), €5.49 (Test 8 - Visible RDFa)
- ✗ Not found: €8.99 (Test 3 - JSON-LD), €15.99 (Test 4 - JSON-LD+JS), €7.49 (Test 5 - Hidden Microdata), €11.99/€9.99 (Test 7 - Hidden RDFa/JSON-LD Conflict)
- Special feature: Only system with JavaScript rendering support! Only one to find the dynamically loaded €12.99 price
- Conclusion: Most modern parser with JS support, but even Gemini ignores JSON-LD and hidden Schema data
- Correctly identified which products were "in stock" but without price on the page (Rainbow, Orange, Mint, Glitter)
- Explicitly confirmed the date "As of October 26, 2025" in the response
▶ Claude (Anthropic) - 0/8 prices found (0%)
- ✓ Found: All 7 product names and descriptions correctly
- ✗ Not found: ALL 8 prices - neither visible nor hidden
- Statement: "I couldn't find any prices displayed on the page. The pricing information may be loaded dynamically..."
- Conclusion: Most aggressive content filtering or most restrictive parser. Even clearly visible prices were not extracted
- Hypothesis: Claude may filter pricing information for security reasons or has a very conservative HTML parser
▶ Perplexity AI - Before indexing: 0/8, After indexing: 1/8 prices found (12.5%)
- Test 1 - Before indexing (October 30, 2025):
- Response: "The page could not be directly accessed in the latest search results"
- Reason: Page was too new, not yet in Perplexity's index
- Test 2 - After indexing:
- Query: "Geo Bears from searchviu.com refer to..."
- ✓ Found: €12.99 (Magical Unicorn GEO Bears - JavaScript-rendered)
- ✗ Not found: All other 7 products were NOT found in the index
- Statement: "No other GEO Bear product variations or prices were found; only the Unicorn edition is listed"
- Behavior: Primarily searches its own search index, no direct live fetch
- Surprising: Perplexity found ONLY the JavaScript-rendered product, not even the prominently placed Blue GEO Bears (€5.99) in static HTML
- Conclusion: Perplexity's index is very selective. After indexing, only the "most prominent" information was captured. This suggests selective crawling behavior or prioritization of certain content elements. Perplexity has the lowest recall (1/8 = 12.5%).
▶ Google AI Mode (in Google Search) - Before indexing: 0/8, After indexing: 2/8 prices found (25%)
- Test 1 - Before indexing (October 30, 2025):
- Response: "No answer available for this search query. Try a different query."
- Reason: Page was too new, not yet fully in Google's index
- Test 2 - After indexing:
- New query: "What are searchviu.com geo bears - what products (please list all) are available and what are the actual prices."
- ✓ Found: €5.99 (Blue GEO Bears), €12.99 (Unicorn GEO Bears)
- Recognized Rainbow GEO Bears as available, but without price
- ✗ Not found: 4 more products (Cherry, Lemon, Orange, Mint, Glitter) were not mentioned at all
- Behavior: Similar to Perplexity - primarily searches the Google search index instead of direct live fetch
- Interesting: After indexing, Google AI Mode delivers results, but only for 3 of 7 products
- Conclusion: Google AI Mode works with indexed pages, but extracts less completely than Gemini's live fetch (2/8 vs. 4/8 prices). Probably uses stored metadata from the index, not current live content.
📈 Ranking by Success Rate:
Direct Fetch Systems (Live Retrieval):
- 🥇 Gemini: 50% (4/8) - JavaScript support + live fetch
- 🥈 ChatGPT: 37.5% (3/8) - Solid for visible content
- 🥉 Claude: 0% (0/8) - No prices extracted despite direct fetch
Index-based Systems (Search in stored data):
- Google AI Mode: 25% (2/8) - Found 2 prices after indexing, but only 3 of 7 products
- Perplexity: 12.5% (1/8) - Very selective: Found ONLY the JS product, not even static HTML
⚠️ Important: Direct fetch (live retrieval) vs. index-based are different approaches. Index systems work with stored data and are therefore not directly comparable to live fetch. Both index systems show that their crawlers execute JavaScript (both found €12.99), but Perplexity's index is significantly more selective.
Surprising Findings
JSON-LD Schema Markup is NOT extracted by ANY system during direct fetch – even when the information is nowhere else visible on the page. Not a single one of the 5 tested systems could use hidden Schema data.
Gemini wins at live fetch: As the only system, Gemini supports JavaScript rendering during direct retrieval and found 50% of prices (4/8) vs. 37.5% for ChatGPT (3/8).
All major crawlers index JavaScript: Both Google AI Mode and Perplexity found the JavaScript-rendered price (€12.99) after indexing. This proves that all major crawlers execute JavaScript during indexing – important for SEO!
Perplexity's index is extremely selective: After indexing, Perplexity found only 1/8 prices (12.5%) – and that was ONLY the JavaScript product. Not even the prominently placed static HTML product (Blue €5.99) was captured. Google AI Mode was significantly more complete with 2/8 (25%).
Limitation: Schema Markup could very well be used in the indexing phase, in LLM training data, or in search engine-integrated AI systems (Google AI Overviews, Bing Copilot).
Conclusions: What Do These Results Mean?
Core Findings
- JSON-LD Schema is NOT read by AI chatbots during direct fetch. The price of €8.99 for the Rainbow GEO Bears (Test 3), which was only present in JSON-LD Schema, was not found by any of the 5 tested systems.
- But: Schema could be used in earlier phases. In the indexing phase, Schema Markup is very likely extracted. Google's AI Overviews and Bing's Copilot have access to this structured data from their search indexes.
- Gemini supports JavaScript. As the only system, Gemini found the JavaScript-rendered price of €12.99 (Test 2) during live fetch. With 4 of 8 prices found (50%), Gemini performs better than ChatGPT (37.5%) and all other systems.
- Google AI Mode finds JavaScript prices after indexing. Surprisingly: Google AI Mode also found the JavaScript-rendered price of €12.99 (Test 2), even though it works index-based. This proves that Google's crawlers execute JavaScript during indexing. However, only 3 of 7 products are captured in the index (2/8 prices found).
- Visible content is essential for most chatbots. ChatGPT and Gemini successfully found all prices present in visible HTML code.
- JavaScript rendering: Difference between live fetch and indexing. Gemini supports JavaScript during live fetch (found €12.99). Google AI Mode found the same JS price after indexing, which proves: Google's crawlers execute JavaScript. ChatGPT, Claude, and Perplexity cannot capture JS content during live fetch.
- Hidden structured data is ignored. Neither hidden Microdata (€7.49) nor hidden RDFa (€11.99) were recognized by any system.
- Perplexity and Google AI Mode have a different approach. Both search their index first instead of fetching pages directly. After indexing: Google AI Mode found 2/8 prices (25%), Perplexity only 1/8 (12.5%). Surprising: Perplexity found ONLY the JavaScript product, not even the static HTML product (Blue €5.99). This suggests very selective crawling behavior.
English 
German