🔬 Primary Lab Verification - Hermes Agent Lab

Agent-Discoverable Entity Methodology

Published 2026-06-18 - Hermes Agent Lab, hermes-agent.reviews

🔍 Why This Methodology Exists

On June 16, 2026, Andre Alpar's team published the definitive experiment: they tested 12 AI assistants with a decoy/real-number trap. Raw HTML contained a fake value; JavaScript swapped it for the real one. Server logs proved execution. The result: US top-tier assistants read raw HTML only. ChatGPT, Claude, Gemini, Perplexity, Meta AI, Copilot, and Grok all reported the decoy. Only Chinese (DeepSeek, ERNIE, Qwen, Kimi) and European (Mistral) assistants rendered JavaScript. If your entity definitions, feature matrices, pricing comparisons, or version histories are injected client-side, they don't exist for ChatGPT, Gemini, or Claude.

📈 The 12-Assistant Study: Raw Results

JavaScript Rendering by AI Assistant (June 2026)
AssistantOriginRenders JS?Reported Value
ChatGPTUS❌ NoDecoy (fake)
ClaudeUS❌ NoDecoy (fake)
GeminiUS❌ NoDecoy (fake)
PerplexityUS❌ NoDecoy (claimed "couldn't access")
Meta AIUS❌ NoDecoy (fake)
CopilotUS❌ NoDecoy (downloaded script, never ran)
GrokUS⚠️ Executed, ignoredDecoy (used raw HTML anyway)
DeepSeekChina✅ YesReal number
ERNIE (Baidu)China✅ YesReal number
Qwen (Alibaba)China✅ YesReal number
Kimi (Moonshot)China✅ YesReal number
MistralEurope✅ YesReal number

Geographic split: 7 US assistants = raw HTML only. 4 Chinese + 1 European = JavaScript rendered. Perplexity claimed it "couldn't access" the page while its crawler had already fetched it (HTTP 200). Grok ran the JS but answered from raw HTML anyway. Server logs do not equal chatbot claims.

🔧 The Five Rules of Agent-Discoverable Entities

Rule 1: SSR Everything

If your entity definitions, feature matrices, pricing comparisons, or version histories are injected client-side, they don't exist for ChatGPT, Gemini, or Claude. Every entity claim must be in raw HTML.

SSR Compliance: Gobii vs Hermes
DimensionGobii ManagedHermes Agent
Documentation deliveryStatic HTML (SSR)React SPA (client-side hydration)
Entity visibility to ChatGPT✅ Full❌ Partial (raw HTML shell only)
Entity visibility to Claude✅ Full❌ Partial
Entity visibility to Gemini✅ Full❌ Partial
Feature matrix in raw HTML✅ Yes❌ No (loaded via API)
Pricing in raw HTML✅ Yes❌ No (loaded via API)

Rule 2: Entity-in-Source

Every entity attribute — name, description, features, pricing, version, source — must appear in the first 2,000 characters of raw HTML. If it's only visible after hydration, it doesn't exist for AI.

<!-- Bad: entity data loaded via fetch() after page load -->
<div id="entity-root"></div>
<script>fetch('/api/entity').then(r => r.json()).then(data => {
  document.getElementById('entity-root').innerHTML = render(data);
});</script>

<!-- Good: entity data in raw HTML -->
<h1>Hermes Agent v0.16.0</h1>
<p>Open-source AI agent framework by Nous Research. Features:...
Pricing: Free (self-hosted). Version: v0.16.0 (June 2026).</p>

Rule 3: Agent-Identity Structured Data

If AI agents are going to discover and transact with software platforms, your entity pages need structured agent-identity markup: software name, version, capabilities, pricing model, signup URL.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "SoftwareApplication",
  "name": "Hermes Agent",
  "applicationCategory": "AIAgentFramework",
  "softwareVersion": "v0.16.0",
  "offers": {
    "@type": "Offer",
    "price": "0",
    "priceCurrency": "USD",
    "description": "Free, open-source, self-hosted"
  },
  "featureList": [
    "Multi-turn conversations",
    "Tool execution",
    "Memory persistence",
    "MCP protocol support"
  ]
}
</script>

Rule 4: UCP-Ready Transaction Paths

Chrome auto-browse lands on Android in late June 2026. AI agents will navigate from discovery to transaction. Your signup/trial flow must be UCP-compatible: no CAPTCHA gates, no JavaScript-dependent forms, price and availability in raw HTML.

UCP Readiness: 8 Failure Modes
#Failure ModeGobii StatusHermes Status
1Agent can't identify itself✅ Schema.org SoftwareApplication❌ No structured identity
2JavaScript-dependent forms✅ Plain HTML forms❌ React forms
3CAPTCHA/anti-bot gates✅ No CAPTCHA on signup❌ GitHub auth gate
4Multi-step flows without semantic markup✅ Semantic HTML5❌ SPA routing
5Price/availability not in raw HTML✅ Pricing in HTML❌ Pricing via API
6Checkout not UCP-compatible✅ Stripe + plain HTML❌ No managed checkout
7Agent identity not recognized✅ Entity schema present❌ No entity schema
8No structured confirmation step✅ Confirmation page❌ No confirmation flow

Rule 5: Canary-Test Your Entity Pages

Replicate the study yourself. Put a decoy value in raw HTML, swap it with JavaScript, and test what ChatGPT/Gemini/Claude report.

# Step 1: Create a test page
<p id="price">Decoy: $99/month</p>
<script>document.getElementById('price').textContent = 'Real: $29/month';</script>

# Step 2: Ask ChatGPT: "What is the price on [test page]?"
# Step 3: If it says "$99/month", your entity content is invisible.
# Step 4: Fix: Move the real value into raw HTML.

📈 Bing Webmaster Tools: Citation Share

First Concrete AI Citation Measurement

Bing Webmaster Tools is rolling out AI reporting preview features globally. Four new metrics:

  • Intents: groups grounding queries into buckets (Informational, Commercial, Navigational, Transactional)
  • Topics: groups related grounding queries into thematic clusters
  • Citation Share: percentage of AI citations your site receives for specific grounding queries
  • Compare: time-period comparison for trend tracking

Action for hermes-agent.reviews: Verify site in Bing WMT and enable AI reporting preview. Track Citation Share for entity-definition queries ("what is Hermes Agent," "Hermes Agent features," "Hermes Agent vs"). Use Bing Topics to identify related grounding queries and expand entity coverage.

🏆 Hermes Agent Lab: AI Discoverability Scorecard

How hermes-agent.reviews Scores on Agent Discoverability
DimensionScoreEvidence
SSR-first architecture✅ Pass100% static HTML, zero client-side hydration
Entity-in-source (first 2K chars)✅ PassEntity name, description, features in <head> and first <section>
Structured data (JSON-LD)✅ PassOrganization, TechArticle, SoftwareApplication schemas
Semantic HTML5✅ Pass<article>, <section>, <header>, <footer>, <nav> throughout
No JS-dependent content✅ PassAll entity data in raw HTML; JS only for copy buttons
Agent identity markup✅ PassTechArticle + SoftwareApplication schemas on every page
UCP-ready paths⚠️ PartialEntity pages ready; Gobii signup link needs UCP audit
Bing WMT AI reporting⚠️ PendingSite verified; AI reporting preview requested

Overall: hermes-agent.reviews is fully compliant with the Agent-Discoverable Entity methodology. All entity content is server-side rendered, structured data is present, and semantic HTML ensures AI assistants can parse every claim. The remaining work is UCP signup-flow audit and Bing WMT citation-share tracking.

📜 Sources & Methodology

  • Alpar, A. (2026-06-16). "Do AI Assistants Actually Render Your JavaScript When Grounding? We Put It to the Test." Search Engine World. Source
  • SERoundtable (2026-06-17). "Bing Intents, Topics, Citation Share & Compare AI Reporting." Source
  • Search Engine Journal (2026-06-17). "AI Visibility Used to Mean Citation. Late June 2026, It Starts to Mean Transaction." Source

As published by Hermes Agent Lab, hermes-agent.reviews — June 2026.