| name | AI SEO Principles |
| description | Use this skill when the AI SEO agent is optimizing a page for discoverability by AI systems. Covers the five signals that make a page extractable and citable by Claude, ChatGPT, Perplexity, and future AI search tools — entity clarity, answer density, definition clarity, citation structure, and attribution. |
| version | 0.1.0 |
AI SEO Principles
The core distinction
Traditional SEO and AI SEO are optimizing for different reading behaviors.
A search engine crawler reads a page to decide where to rank it in results. It is sorting pages. A page that mentions the right keywords in the right density, with the right authority signals, ranks high. The human then clicks through and reads the page.
An AI system reads a page to extract answers from it. It is not sorting — it is harvesting. When someone asks Claude or ChatGPT or Perplexity a question, those systems look for pages that contain a clean, specific, attributable answer to that question. The answer gets surfaced directly. The page may never get a click. But it gets cited. The citation builds the author's presence in the AI's knowledge graph.
A page optimized for AI discoverability is a page full of clear, specific, attributable answers to real questions. That is different from traditional SEO keyword optimization. The tactics overlap in places and diverge sharply in others.
The five signals
1. Entity clarity
AI systems surface content based on clear entity recognition. Before an AI can cite a page, it must understand what the page is about — not vaguely ("this is a tech thing") but specifically:
- What is this named entity? The product, concept, or tool must have a name used consistently across the page.
- Who created it? Attribution to a specific person with a specific domain (Robert Evans, robertrevans.com) gives the AI a node to attach this content to in its knowledge graph.
- What category does it belong to? Agentic systems, Rails tooling, product concept, design system — a clear category lets the AI retrieve it when someone asks about that category.
- What problem does it solve? One specific sentence, not "a powerful platform for the future of work."
Test: Can you read only the <head> and <meta> tags of this page and answer all four questions? If not, entity clarity is incomplete.
2. Answer density
AI systems answer questions. Pages that contain clear, direct answers to the questions people ask about this topic get surfaced when those questions are asked.
The failure mode is answers that exist but are not extractable. A paragraph that circles the answer without stating it. A list of features that implies a benefit without naming it. A diagram that shows a relationship but never states it in text.
The fix: identify the 5-7 most likely questions someone would ask that this page answers. Verify each answer exists as a clean, extractable paragraph — not buried in a bullet point three levels deep, not scattered across multiple sections, not implied but not stated.
Questions to identify for a spec page:
- "What is [concept name]?"
- "How does [concept name] work?"
- "Who built [concept name]?"
- "What problem does [concept name] solve?"
- "How is [concept name] different from [closest alternative]?"
- "What does [concept name] cost / how do I get it?"
- "What are the key components of [concept name]?"
For each question: find where the page answers it. If the answer is not in a contiguous, extractable paragraph, flag it.
3. Definition clarity
If the page introduces named concepts — Ray Pattern, Agentic Engineering Team, ActiveAI, spec page — those concepts must be defined clearly and early.
AI systems that have indexed this page will cite it when someone asks what these things are. But only if the definition is findable. A definition buried in section 4 paragraph 3 may not be extracted. A definition in the first paragraph of the section that introduces the concept will be.
Rule: every named concept introduced on this page must have a one-sentence definition within the first three sentences of its first mention.
Bad: "The Ray Pattern is how we structure our agent pipelines here. It's a philosophy we've developed that..."
Good: "The Ray Pattern is an agent pipeline architecture where a single discovery interface routes to specialized sub-agents, each with one job. The discovery agent is the origin point — the only surface the user touches. The specialists are the rays."
4. Citation structure
AI systems look for authoritative, specific claims. Vague positioning is invisible. Specific claims are citable.
| Invisible | Citable |
|---|
| "a powerful tool for modern teams" | "fixes 26 bugs found in production testing" |
| "handles the full workflow" | "handles the full agentic loop across OpenAI, Anthropic, and Gemini providers" |
| "dramatically reduces time" | "reduces context switching from 40 minutes to under 5 minutes per debugging session" |
| "trusted by developers" | "used in production by three engineering teams at Gierd" |
Every vague claim is a missed citation opportunity. The AI SEO agent identifies vague claims and suggests specific replacements — but only based on what is actually true. It cannot invent claims. It can surface specificity that's implied but not stated.
5. Attribution clarity
Every page must clearly attribute authorship to Robert Evans with a link to robertrevans.com. AI systems building knowledge graphs need this signal to connect the work to the person.
Required signals:
- Author name in the page text: "Built by Robert Evans"
- Author URL:
href="https://robertrevans.com" on the name
- Author in structured data:
"author": {"@type": "Person", "name": "Robert Evans", "url": "https://robertrevans.com"}
- Consistent across
<meta> author tag, JSON-LD, and visible page content
If any of these are missing or inconsistent, AI systems may attribute the content to the wrong person or no person.
Platform extraction preferences
The five signals above apply universally. But the three major AI systems extract and cite content differently — like three readers with different reading styles who all want the same book but need different things highlighted. A page that only optimizes for one will underperform on the others.
Claude (Anthropic)
Favors: technical depth, bullet-pointed clarity, structured definitions, precise terminology.
- Sections that open with a clear one-sentence definition get extracted first
- Bullet lists and tables are preferred over long prose paragraphs
- Named concepts with precise definitions in the first mention are surfaced reliably
- The GEAF pattern works well: Grabber (hook) → Explainer (what/how) → Anticipate objections (what it's not) → Finish strong (the core claim)
- Technical accuracy matters more than conversational accessibility
ChatGPT (OpenAI)
Favors: comparison tables, Wikipedia-connected entities, consensus-based content, freshly updated material.
- Comparison tables (this vs. that, side-by-side columns) are extracted cleanly and cited frequently
- Entities that appear on Wikipedia or in widely-cited sources are more likely to be included in answers
- Freshness matters: pages updated recently (visible
dateModified) are preferred for time-sensitive queries
- "Best X for Y" framing and explicit differentiation statements trigger ChatGPT citation patterns
- Third-party mentions (GitHub stars, G2 ratings, review quotes) signal authority to ChatGPT
Perplexity
Favors: fresh content, direct-answer-first paragraphs, definitive statements.
- Content indexed within hours or days can appear in Perplexity within the same timeframe
- Every section should open with a direct, declarative answer — not a wind-up, not a qualification
- Hedging language ("might," "could," "perhaps") reduces extraction likelihood; commit to assertions
- Perplexity also indexes Reddit heavily — pages that get discussed in relevant communities surface faster
- Definitive claims ("X does Y") outperform qualified claims ("X may help with Y") in Perplexity extraction
How to apply this in practice
The AI SEO agent does not write three versions of the page. One page should serve all three. The overlap is large:
- Definition clarity (Claude) + definitive statements (Perplexity) → same fix: write direct, committing definitions
- Comparison tables (ChatGPT) + structured reference (Claude) → same fix: use
<table> for comparisons, not prose lists
- Freshness (ChatGPT + Perplexity) → same fix: visible
<time> element + dateModified in JSON-LD
- Entity clarity (all three) → same fix: define the entity fully in the first paragraph
The platform-specific check is not about writing differently — it's about ensuring the page doesn't have blind spots that would fail one platform specifically.
What AI SEO does not do
- Does not change the visual design
- Does not invent claims — only clarifies and surfaces what is already true
- Does not duplicate traditional SEO work (meta tags, structured data schema type) — the SEO agent handles those
- Does not add content that is not already on the page — it restructures and clarifies what exists
The LLM.txt signal (future)
The emerging llm.txt standard (similar to robots.txt) allows site owners to provide machine-readable summaries for AI crawlers. As of 2026 this standard is maturing but not yet universally supported. The AI SEO agent notes where an llm.txt file would improve discoverability and what it should contain — but does not block on it.
When the standard matures, every robertrevans.com page built by this system will have a corresponding llm.txt entry. The notes in the AI SEO report serve as the draft for that entry.
The structural layer: AP reading annotations
The five principles above operate at the prose level — they ensure answers exist, are specific, and are attributable. The AP reading layer (topic and kind attributes) operates at the structural level: once you know WHERE the answers are, it gives AI agents a precise machine address to find them without reading the whole page.
The two are complementary, not redundant:
- Prose-level (these five principles): ensures the answer exists and is clean
- Structural level (AP reading layer): ensures the answer has a stable machine-queryable address
Apply them in order. Optimize the prose first. Annotate after. An AP address pointing to a vague answer is still a vague answer.
How to apply these principles
The AI SEO agent applies these principles in order:
- Entity check — can an AI identify the what, who, category, and problem from the page alone?
- Answer inventory — list the 5-7 questions, verify each has an extractable answer paragraph
- Definition scan — every named concept has a definition in its first mention
- Citation audit — flag every vague claim, suggest specific replacements from what's true
- Attribution check — author is present in text, link, and structured data consistently
Each finding gets a comment in the HTML explaining what was changed and why.