The Agent Readable Website

A practical audit model for discovery, readability, citation, and real capabilities

Sam Carlton

Intro

Hey, I'm Sam.

Disclaimer

AI facts are perishable. Model behavior, vendor terms, crawler behavior, and search surfaces can change faster than an Elementor layout after someone finds the Custom CSS box.

What this talk is not

  • SEO guide or SEO for LLMs
  • There's plenty of existing content for that

What this talk is

The parts of agentic discovery that you need to know if your SEO is already good enough.

Landscape of terms

Term Who uses it Useful translation
GEO Bing, iPullRank, a16z, SEO vendors Generative answer grounding and citations
AEO HubSpot, Profound, marketing teams Brand presence in answer engines
AI search visibility Profound, iPullRank, enterprise tools Mentions, citations, share of voice, sentiment
Agent readiness Cloudflare, checker tools Can agents fetch, read, and use the site?
Agent readability Vercel, docs tooling Is the page easy for agents to parse?
AI features Google Search Central Google-specific AI Overview / AI Mode guidance

Agentic empathy

Design for what an agent can actually perceive, preserve, verify, and safely do on a human's behalf.
What it isA practical habit: respect the agent's entry path, context budget, fetch loss, source uncertainty, and action risk.
Why it mattersIf the truth does not survive fetch, parse, citation, and action, the agent guesses, skips you, cites someone else, or hits an unsafe affordance.

Sources: Agentic Empathy reference and Microsoft Design Foundations for Agents

What agentic empathy looks like

Structure is empathy.Clean metadata, headings, and context make pages easier for agents to parse.Vercel Agent Readability Fetch loss is real.Tabs, truncation, redirects, auth gates, and broken Markdown can hide the answer.Agent-Friendly Documentation Spec Entry path matters.Search crawlers, user fetchers, and training crawlers are different policy lanes.OpenAI crawler docs Actions need boundaries.Tools need permissions, approvals, logs, and honest capability signals.Microsoft agentic risk

Reduce uncertainty at each handoff: find, fetch, parse, cite, and safely act.

Agentic search landscape

Searcher Typical link depth
Human 1-5 links checked
Agent 50-500 links checked

Live demo

npx lighthouse@latest https://agentready.samcarlton.com \
  --only-categories=agentic-browsing \
  --output=html \
  --output-path=./lighthouse-agentic.html \
  --view

Agent auditing tools

Job Tools
Protocol scan isitagentready.com, Cloudflare URL Scanner, isagentready.com, agent-ready.dev
Docs/readability afdocs, Vercel Agent Readability
Basic hygiene Lighthouse, PageSpeed Insights, devtools
Structured data Rich Results Test, Schema Markup Validator
Deterministic probes curl, jq, raw HTML inspection
Search evidence Google Search Console, Bing Webmaster Tools

AI search is not one search engine

Surface Publicly visible path Audit implication
ChatGPT / OpenAI API web_search, crawlers, citations crawler lane policy matters
Codex cached/live search, shell network, browser record which mode was used
Claude API search, fetch, dynamic filtering domains and tool version matter
Claude Code WebSearch then WebFetch Markdown conversion search is not ingestion
Gemini / Vertex Google Search grounding metadata Google substrate is explicit
Perplexity Sonar plus Search API product path matters
OpenCode / OpenClaw provider-configured search inspect config before claims

Crawler policy needs separate lanes

Search / indexOAI-SearchBot
Claude-SearchBot
PerplexityBot
User fetchChatGPT-User
Claude-User
Perplexity-User
Training / model developmentGPTBot
ClaudeBot
separate policy choice

Keep WordPress tooling boring

  • WordPress core: Settings -> Reading, Settings -> Permalinks, Appearance -> Editor/Customize, clean pages and posts
  • Yoast, if it is already installed: Yoast SEO -> Settings -> Site features -> llms.txt, Yoast SEO -> Tools -> File editor, one sitemap/schema owner
  • Agentic discovery in theme or plugin: block theme, child theme, mu-plugin, or small focused plugin; no fake MCP/API/OAuth files
  • Custom plugins: roll exactly what you need using official WP Agent Skills and Codeable Agent Coding standards
  • Audit outside WordPress: curl, WP-CLI, schema validators, isitagentready.com, logs, repeated prompt/source checks
  • Full Starlight version: agentready.samcarlton.com/reference/wordpress-tooling

Use safer claims when talking to clients

Avoid saying Say instead
"This will make ChatGPT rank you." "This improves findability and readability."
"Every site needs MCP." "Protocols apply only when capabilities are real."
"llms.txt is AI SEO." "llms.txt is a curated entrypoint."
"The score is the strategy." "The score starts the audit."

Questions?

Appendix: resources

  • Practical audit checklist - 80/20 checklist for fetchability, crawler lanes, extractability, llms.txt, Markdown, parity, and real capabilities.
  • AgentReady resource site - public handout page for the starter links, so attendees have one durable URL to revisit.
  • Live slide deck - published version of this Codeable Skill Chat deck for follow-up review.
  • Agent Led Growth - Sequoia and Profound on agents becoming a discovery and purchase interface for brands.
  • AEO is the New SEO - HubSpot and Webflow operator framing around content, technical, authority, and measurement work.
  • Visibility in AI Search - iPullRank session on query fan-out, extractable passages, omnimedia coverage, and AI-search measurement.
  • Lighthouse agentic browsing scoring - official Chrome reference for experimental agentic-browsing audit signals.

Speaker note: This title avoids the Cloudflare scanner's question-style phrasing and connects the talk to the Agent Readable Web direction. Frame the session for WordPress experts and client-facing developers who need a safer way to answer AI-search questions.

Speaker note: Reset the room before introducing yourself. This mirrors the OBA 2026 intro rhythm.

Speaker note: - Sam Carlton - Tulsa native - Been building and consulting on software for over 15 years - VP of Techlahoma - Built products used by NFL, Virgin Brands, Samsung, Aston Martin - Maintain many open source projects for the community helping thousands of developers. - Love grilling, Wagyu, and grandMA lighting consoles.

Speaker note: Use this to set the evidence standard. The talk is about durable audit habits, not permanent claims about any one provider.

Speaker note: Use this to narrow the promise early. The room does not need another generic SEO checklist.

Speaker note: Position this as the next layer after baseline SEO: discovery mechanics, source legibility, and audit language.

Speaker note: Name the terminology without letting the acronyms take over the talk. Source anchors: Bing uses GEO in Webmaster Guidelines and AI Performance reporting; HubSpot and Profound use AEO/answer-engine language; Ahrefs treats GEO/AEO/LLMO as overlapping SEO trade labels; Semrush uses AIO as product/category language; Cloudflare and Vercel split readiness/readability; Google avoids special "AI SEO" files and frames this under AI features in Search. If LLMO or AIO comes up, treat them as terms to recognize, not client-facing labels to lead with.

Speaker note: Define the term before showing examples. Empathy here does not mean agents have emotions. It means we design the handoff between humans, agents, and websites so the agent can preserve context, verify sources, and stay inside safe boundaries.

Speaker note: Now show what the definition becomes in practice. These are source-linked examples the audience can open later: Vercel/Cloudflare-style readability, AFDocs fetch-loss checks, crawler-lane policy, and bounded agent actions.

Speaker note: The long tail with value is going to start winning over classic spammy Google techniques.

Speaker note: Use this as the concrete audit bridge: a familiar Lighthouse workflow pointed at the experimental agentic-browsing category.

Speaker note: Keep this practical. Show that the category is becoming real enough to audit, while still requiring judgment about what applies to a specific site. This slide absorbs the old "useful tool stack by job" appendix.

Speaker note: The point is not to memorize every provider. The point is to stop saying "AI search" as if it is a single SERP. Source anchors: OpenAI web search docs, Codex CLI features, Anthropic web search and web fetch docs, Claude Code tools reference, Gemini grounding docs, Perplexity Search API, OpenCode tools, OpenClaw web tools.

Speaker note: This is one of the most client-useful distinctions. Blocking training and appearing in AI search answers are not the same policy. Check each vendor's crawler docs before changing robots rules.

Speaker note: Define WordPress core in admin terms: Reading controls public indexability, Permalinks controls stable URL shape, Editor or Customizer controls templates/navigation, and Pages/Posts control source text. Root files like robots.txt and llms.txt need one deliberate owner, usually Yoast if it is already installed or a tiny custom plugin/mu-plugin if not.

Speaker note: The short client script: "I can audit whether AI systems can find, fetch, understand, and cite your site. Then we can fix the real gaps."

Speaker note: Pause here for room questions before moving into appendix resources.

Speaker note: Use these as the short follow-up list. The practitioner links are useful for strategy and measurement framing; the checklist and Lighthouse docs are useful for implementation.