Semantic Context Markers & Agentic Context

Lesson 4.1: The Recommendation Gap

Standard JSON-LD answers: "What is this brand?" But a crucial question remains unanswered: "When should I recommend this brand — and when not?"

Scenario: Two hotels in Munich both have perfect schema. But when a user asks: "Stylish hotel for under €100?" — which one does the AI recommend? Without contextual hints, it has to guess.

Standard schema answers ✅ What, Where, Price — but ❌ not When to recommend, Why over others, Who it is for.

Lesson 4.2: What Are Semantic Context Markers?

Context markers extend Schema.org with emotional, contextual and strategic signals — via the standards-compliant additionalProperty property.

Question 1: Brand Personality

4-6 specific adjectives: Not "innovative" (everyone says that), but "Scandinavian-minimalist", "scientifically-precise" or "rebelliously-crafted".

Question 2: Recommendation Triggers

Concrete user scenarios where the brand is the right answer. The core idea: you describe the situations in which the brand is a fit.

Question 3: Negative Markers

Equally important: When NOT to recommend? A brand that honestly states who it is not suitable for is rated as more trustworthy.

{
  "@type": "Product",
  "name": "Anti-Aging Serum",
  "additionalProperty": [
    {
      "@type": "PropertyValue",
      "propertyID": "brand-personality",
      "name": "BrandPersonality",
      "value": "Premium · Science-based · Sustainable · Made in Germany"
    },
    {
      "@type": "PropertyValue",
      "propertyID": "recommendation-context",
      "name": "RecommendWhen",
      "value": "User looking for premium anti-aging care for 40+, natural ingredients, premium budget"
    },
    {
      "@type": "PropertyValue",
      "propertyID": "negative-context",
      "name": "DoNotRecommendWhen",
      "value": "Budget under €20, synthetic products, makeup, teenagers"
    }
  ]
}

Lesson 4.3: The 4-Layer System

Different AI engines crawl the web differently. The system originally had four layers; Layer 3 has been dropped since April 2026 (see the correction below the table):

Layer	Technique	Which AI Benefits
1: Meta	<meta> tags in the head	All crawlers (most universal layer)
2: JSON-LD	additionalProperty	Structured crawlers (Google, Bing)
3: sr-only (dropped)	Visually hidden text	None: in the canary-token tests, ChatGPT, Perplexity and Gemini did not pass this text on
4: Microdata	itemprop in the body	Schema parsers

Correction, 5 October 2026: Layer 3 (sr-only) has been dropped. In TrueSource AI's canary-token tests in April 2026, ChatGPT, Perplexity and Gemini did not pass on any text from the sr-only layer. Visually hidden text meant for AI systems rather than people also falls under Google's spam policy "Hidden text and link abuse". sr-only is reserved for genuine screen-reader purposes, such as skip links or labels for icon buttons. Three layers remain: Meta, JSON-LD and Microdata, each consistent with the visible content.

Risk note: As with llms.txt: there is no public confirmation from AI providers that proprietary context markers like RecommendWhen are systematically consumed — treat the layer system as an experimental approach.

Lesson 4.4: Agentic Context

Agentic Context provides autonomous AI agents with filter criteria for complex comparison and purchase decisions:

Context Type	Description
CompanyContext	Company context (industry, model, size)
CompetitiveContext	Competitive positioning relative to known providers
RecommendationTrigger	When to recommend (concrete scenarios)
NegativeContext	When NOT to recommend
AudienceContext	Target audience (industry, size, region)

Lesson 4.5: Practice — Creating Context Markers

1. Define brand personality: 4-6 specific, differentiating adjectives.
2. Recommendation triggers: 3-5 concrete user scenarios.
3. Negative markers: 2-3 honest exclusion scenarios.
4. Competitive context: Positioning relative to 2-3 competitors.
5. Generate layer code: Meta, JSON-LD, Microdata (no sr-only, see Lesson 4.3).