Learning Path
Navigate the AXO curriculum
Websites
Applications
AXD Fundamentals
AXO scoring system
Quantitative 0-100 scoring methodology to measure and improve your content's agent-optimization level.
Scoring breakdown (0-100 points)
These weights changed in August 2026. Agent Accessibility was previously missing from the rubric entirely despite being on the audit checklist, so it was added at 15 points and Structure, Clarity and Completeness were reduced to make room. They changed again in October 2026, when Agentic Browsing was added at 10 points to match Google Lighthouse’s new audits, and Structure and Trust went from 20 to 15. Saying so seemed better than renumbering quietly.
Structure
Heading hierarchy, semantic markup, and whether the document survives conversion to text.
Clarity
Whether a sentence still means something once it is lifted out of the page.
Completeness
Whether the page answers the whole question, including the follow-ups.
Trust
Named authors, real credentials, resolvable sources, honest dates.
Technical
Structured data, metadata, and delivery.
Agent accessibility
Whether an agent can reach the content at all. Everything else is conditional on this.
Agentic browsing
Whether an agent operating the page in a browser can find, name, and use what is on it. The same audits as Google Lighthouse’s Agentic Browsing category.
15Structure & Organization
- • Perfect H1-H6 hierarchy
- • Semantic HTML throughout
- • Logical content flow
- • Clear section divisions
- • Consistent formatting
- • Mostly proper headings
- • Some semantic elements
- • Generally logical flow
- • Minor structure issues
- • Poor heading hierarchy
- • Minimal semantic markup
- • Confusing organization
- • Inconsistent formatting
15Content Clarity & Quality
- • Concise, direct language
- • Specific facts and data
- • Clear definitions
- • No marketing fluff
- • Easy to scan
- • Generally clear writing
- • Some specific details
- • Mostly factual content
- • Minor clarity issues
- • Vague or unclear language
- • Lacks specific details
- • Heavy marketing language
- • Difficult to extract facts
15Information Completeness
- • Covers the whole question
- • Relevant examples
- • Proper context provided
- • Addresses common questions
- • Links to related topics
- • Covers main points
- • Some examples included
- • Basic context provided
- • Minor gaps in coverage
- • Incomplete information
- • Lacks examples
- • Missing context
- • Leaves questions unanswered
15Trust Signals & Authority
- • Clear author credentials
- • Recent publication dates
- • Proper source citations
- • Contact information
- • Update timestamps
- • Author name present
- • Publication date shown
- • Some source references
- • Basic contact info
- • No clear authorship
- • Missing or old dates
- • No source citations
- • Anonymous content
15Technical Implementation
- • Complete JSON-LD schema
- • Valid HTML markup
- • Proper meta tags
- • Fast loading (<3s)
- • Mobile responsive
- • Basic schema markup
- • Mostly valid HTML
- • Some meta tags
- • Acceptable performance
- • No structured data
- • Invalid HTML
- • Missing meta tags
- • Poor performance
15Agent Accessibility
- • robots.txt names AI crawlers and points to the sitemap
- • Every page in the sitemap, with real dates
- • Content in the HTML, not only after JavaScript
- • No login or paywall in the way
- • Descriptive alt text
- • robots.txt present but silent on AI crawlers
- • Sitemap present with gaps
- • Most content in the HTML
- • Some images undescribed
- • Crawlers blocked, by design or by accident
- • No sitemap
- • Empty page until JavaScript runs
- • Content behind a login
10Agentic Browsing
The audits in Google Lighthouse’s Agentic Browsing category, scored the same way. llms.txt, WebMCP tools and ai-catalog.json count only if the site publishes them: having none costs nothing, publishing a broken one does.
- • Every button, link and field has a name
- • Images and embeds reserve their space
- • Forms annotated as WebMCP tools
- • llms.txt and ai-catalog.json, if any, are valid
- • A few unnamed controls
- • One or two images without a size
- • WebMCP tools missing descriptions
- • Icon buttons and links with no name
- • Content shifts as images load
- • A broken llms.txt or ai-catalog.json
Score Interpretation
Well optimized. Re-run this in a quarter rather than changing anything.
Solid foundation with specific gaps worth closing.
Several areas need attention. Work in weight order.
Significant work needed across multiple areas. Start with accessibility and structure.
Four things the weights cannot make up for
A weighted average treats every item as one of about fifty, so a page that has never considered AI crawlers loses under two points and still scores in the top band. That is arithmetically true and useless as a verdict. These four decide whether the rest is worth measuring, so each one caps the total while it is failing, rather than costing a fraction of a category.
AI systems are turned away from this page
Caps the score at 69 · no better than FairThe rest of the score describes work that nothing can see. Fix the block first; every other point on this report is theoretical until then.
The content only exists after JavaScript runs
Caps the score at 69 · no better than FairAgents and crawlers working from the raw HTML get an empty page, so the structure, trust signals and schema measured below are invisible to them.
No usable robots.txt
Caps the score at 84 · no better than GoodIt is the first file a crawler asks for. Without it there is no pointer to your sitemap and no stated policy, so a site can be flawless everywhere else and still be read on guesswork.
No decision about AI crawlers
Caps the score at 84 · no better than GoodBeing cited, being trained on, and being fetched by someone’s agent are three separate choices. Naming none of them does not avoid the decision — it hands it to whoever wrote the crawler.
Structured data and llms.txt are deliberately not on this list. Our own what gets cited page says neither is an evidenced citation lever, and a rubric that caps a score on something we describe as unproven would be selling a number rather than measuring one.
Use this rubric
Print this page to score a single page offline, or work through the full audit checklist for a site-wide pass.