Core Concepts

Inception Agents is built on three interconnected systems: Detection, Optimization, and Measurement. Together, they make sure that when an AI agent evaluates your product or service, it sees the most relevant, well-structured content possible, and that you can see what it asked and how far it got.

This page explains what each system does and introduces the key primitives you will encounter throughout the platform.

The Three Systems

┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│  Detection   │────▶│ Optimization │────▶│ Measurement  │
│              │     │              │     │              │
│  Identify    │     │  Shape what  │     │  Record what │
│  agent visits│     │  agents see  │     │  they asked  │
└─────────────┘     └──────────────┘     └──────────────┘
       ▲                                        │
       └────────────────────────────────────────┘
          You decide what changes, from the record

1. Detection identifies when an AI agent visits your site, which platform it belongs to, and what it is looking for.
2. Optimization serves the right content in the right format for that specific agent and intent.
3. Measurement records each agent conversation with your site’s agent tools: what it asked, exactly what it was told, and how far it got.

Measurement is what closes the loop: the record shows what agents asked and where your answers fell short, so you know what to change next.

Detection

The detection layer identifies AI agent traffic and classifies it by platform, agent type, and inferred intent.

What It Covers

• 40+ distinct AI agents across 7 platforms (OpenAI, Anthropic, Google, Microsoft, Amazon, xAI, Perplexity)
• Crawler bots (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, BingBot)
• Real-time browsing agents (ChatGPT Browse, Copilot, Perplexity Search)
• API-based agents (Shopify Agentic API consumers, ACP/UCP clients)
• Embedded agents (Rufus within Amazon, Gemini within Google Shopping)

How It Works

Detection runs at the edge, before your origin server processes the request. For human visitors, the overhead is sub-2ms — effectively invisible. When an agent is detected, the request is routed through the optimization layer. When a human is detected, the request passes through untouched.

Detection uses a combination of:

• User-Agent analysis — primary signal for known crawlers
• TLS fingerprinting — identifies headless browsers and API clients
• Behavioral patterns — request cadence, header signatures, navigation patterns
• IP range correlation — validated against known agent infrastructure

Every detection event is logged with platform, agent type, requested URL, and timestamp.

Optimization

The optimization layer transforms your content for agent consumption. It operates across three tiers, each targeting a different stage of agent evaluation.

The Three-Layer Content Architecture

Each layer is generated and maintained automatically. You can override or customize any layer from the dashboard.

Per-Platform Tailoring

Different agents process content differently. The optimization layer accounts for this:

• OpenAI agents receive content emphasizing structured product attributes and direct answers
• Anthropic agents receive longer-form content with detailed comparisons and evidence
• Perplexity agents receive citation-friendly content with clear source attribution
• Google agents receive schema.org-heavy markup with entity relationships
• Microsoft Copilot receives API-compatible product catalog data
• Amazon Rufus receives competitive positioning signals and availability data
• Grok receives content enriched with recency signals and social proof

The platform selects the appropriate variant based on detection output. You do not need to manage per-platform content manually.

Measurement

When an AI agent uses your site’s agent tools (your MCP endpoint, your agent endpoints or the tools on your pages), each visit is recorded as one conversation in your workspace. Recording is on from day one for new workspaces.

What Each Conversation Holds

• The agent’s question, in its own words — contact details are masked when it is recorded
• Exactly what your site answered, and whether it was a real answer
• How far the agent got — found, understood, chosen, acted, delivered
• How it ended — for example, the agent got real answers and left without acting, or it reached the confirmation step and the person did not say yes

Conversations are kept 365 days by default, and an erasure request removes them.

What You Do With It

• Replay any conversation turn by turn, in Agent Intelligence > Conversations
• See every question agents ask your site, how often, and whether your site had a real answer, including from buyers who never became leads
• Start sales where the research ended — when an agent completes a request, the lead arrives with What their agent asked first and a link to Replay the conversation
• Ask your own Claude what came in today, through the Inception Agents connector

Plain page crawls are not conversations, so the record covers agents that use your site’s agent tools, not every agent that fetches a page.

Key Primitives

llms.txt

llms.txt is a standardized file — analogous to robots.txt — that provides AI agents with a machine-readable description of your site. It lives at the root of your domain:

https://example.com/llms.txt

The file is written in markdown and contains:

• Site identity and purpose
• Product or service catalog summary
• Key pages and their descriptions
• Structured links for deeper exploration
• Contact and support information

Inception Agents generates and maintains your llms.txt automatically based on your site content. You can also edit it directly.

A more detailed version is available at /llms-full.txt for agents that perform deep-dive analysis.

See the llms.txt for the full specification and customization options.

Agent Card

The Agent Card is a JSON file at /.well-known/agent.json that declares your site’s capabilities to AI agents. Think of it as a machine-readable business card:

{
  "name": "Example Store",
  "url": "https://example.com",
  "description": "Premium outdoor gear and apparel",
  "capabilities": {
    "commerce": true,
    "acp": true,
    "product_feed": "https://example.com/feeds/products.json"
  },
  "contact": {
    "support": "support@example.com"
  },
  "llms_txt": "https://example.com/llms.txt"
}

Agents that support the .well-known/agent.json standard will read this file to understand what your site offers and how to interact with it programmatically.

See Agent Card for an example and the fields it declares.

Content Variants

A content variant is an alternative version of a page or product description optimized for a specific agent intent. Inception Agents generates five variant types:

The platform selects the appropriate variant based on the detected agent’s intent signal. Variants are generated from your existing content — you do not need to write them manually.

See the Content Variants for customization and manual overrides.

Commerce Protocols

Commerce protocols give shopping agents your catalog in the formats their platforms read. Inception Agents serves the manifests and product feeds for four of them; purchases complete on your own store.

With a protocol active, an agent can read your prices, availability and product details in the format its platform expects, and link the shopper to your product page.

See Structured Data for the commerce protocols supported.

Next Steps

• How Agents Evaluate — learn what each AI platform prioritizes
• Quickstart — take your website live, step by step
• Agent Traffic — read your agent visits and what was served to them