A 100-point scoring framework across five pillars. Score each criterion Green, Amber, or Red, total the points, and you have a defensible picture of how ready your site is for the AI-native web.
The principle
Score the substrate, not the surface. A future-proof site is built on a stable, structured, verifiable layer that humans, AI agents and crawlers all share. Immersive design and per-visitor personalisation are a surface that sits on top of that layer — valuable for people, but presentation, not future-proofing. The Index scores the substrate; the surface is deliberately out of scope, with one guard (P3) to keep it from hiding the canonical layer from machines.
| Pillar | Points |
|---|---|
| Discoverability & Structure | 25 |
| Citeability & Answer Architecture | 15 |
| Authority, Trust & Provenance | 20 |
| Accessibility & Performance | 15 |
| Agent Surface & Observability | 25 |
| Total | 100 |
Self-check verifiable by eye or with free validators · Needs a tool needs measurement or ongoing monitoring
Run the free automated audit — it scores the structural criteria for any URL in seconds and flags the rest for review.
The Needs a tool criteria — the measurement and monitoring you can't do by eye — are covered by the deeper scans: the automated audit's Premium report adds Core Web Vitals and domain authority, and the GEO/AEO scan crawls your site to grade answer architecture, AI-crawler access and how you're represented in AI answers. Same criteria — only the depth of automation changes.
Where most marketing managers are still blind: the AI-specific discovery and structure signals beyond classic SEO.
Organization, WebSite, Article, BreadcrumbList, FAQPage and Person schemas present and validated.
llms.txt at root, AI-specific robots rules, noai/noimageai where appropriate.
Correct heading hierarchy, landmark roles, clean predictable DOM that LLMs parse accurately.
A recognised entity in the wider knowledge graph: consistent identifiers, sameAs links, and presence in authoritative sources (Wikidata, registries).
The heart of generative engine optimisation: being not just found but quoted — accurately and with attribution. Optimised for share-of-model, not rank.
Self-contained 80–120 word answer units near the top of key pages that lift cleanly without surrounding context.
Stable fragment IDs and deep links so an engine can cite a specific claim, not just the page.
datePublished/dateModified, visible last-reviewed/'as of' dates, changelogs, and canonical statements of fact engines can rely on.
AI engines prioritise verifiable, credible sources. As the web fills with AI slop, provable origin becomes the moat.
Machine-readable author schema with credentials, qualifications and verifiable experience.
Case studies with outcomes, dated testimonials, current privacy policy, external citations.
Clear human authorship, disclosure of AI-assisted content, and — increasingly — cryptographic provenance (C2PA content credentials, signed content) proving origin.
The shared substrate humans and machines both depend on — now a legal floor, not a differentiator. The immersive, personalised human surface sits on top of this and is intentionally not scored.
LCP, INP and CLS in green; fast TTFB; mobile-first rendering. Verifiable free via PageSpeed Insights / CrUX.
Contrast, alt text, keyboard nav, screen-reader support. Since the European Accessibility Act (in force 28 June 2025) this is a legal obligation for many commercial sites, not best practice.
A canonical, server-rendered content layer identical for every human, agent and crawler — personalisation and immersive UI sit on top without hiding or altering it.
The frontier and the fastest-rising weight: sites agents can use, that defend themselves in the agentic web, and that watch how they are represented.
A documented, callable surface — public API and/or MCP endpoint — so an agent can invoke functionality, not just read about it. Designed for Agent Experience: typed inputs, deterministic, idempotent, capabilities declared.
Rate limiting, scraper detection, noai directives, prompt-injection hygiene, and GDPR / EU AI Act-compliant handling of any data flowing through AI features.
Monitoring how AI assistants actually describe you — accuracy, citations, hallucinations — plus AI-referral analytics. You can't manage what you don't measure.
The criteria evolve as AI web standards do. The monthly Radar keeps your assessment current.
No spam. Unsubscribe anytime. We handle your email per our Privacy Policy.