How to Audit Your AI Readiness: A Step-by-Step Framework
ClickRadius Institute · July 6, 2026
An AI Readiness audit answers a bounded question: what setup evidence can be observed for this site at this time? It does not simulate a private engine decision, predict a citation, or blend external outcomes into a site score. The current ClickRadius model evaluates exactly five evidence families. SEO Health is reported beside readiness, and AI Presence is sampled separately.
Before you inspect anything: freeze the scope
Record the submitted URL, canonical host, capture time, scanner revision, and evidence source for every finding. Distinguish a page observation from a domain-level observation. If a source was not collected, failed, or cannot establish the fact, record unknown; do not turn missing evidence into a zero, a pass, or a diagnosis.
A defensible audit says what was inspected, what was observed, and what remains unknown. It does not claim to know why an external engine selected a source.— ClickRadius Institute
Keep AI Presence outside the readiness score
A buyer-question set belongs to a separate AI Presence sample. ClickRadius samples ChatGPT, Gemini, Perplexity, Claude, and Grok monthly and records what those engines returned for the disclosed questions. That sample can show observed mentions or citations at its capture time. It does not make AI Presence a sixth readiness family, and a later change does not by itself prove which intervention caused it.
The five AI Readiness evidence families
The current starting weights are Off-site Authority 28%, On-page GEO 26%, Schema Engine 18%, Entity / E-E-A-T 18%, and Crawler Access 10%. These are labelled starting estimates, not weights fitted to customer outcomes. Raw source prevalence, intervention leverage, and score weight are different quantities.
1. Off-site Authority — 28%
Inventory the independently controlled records available about the entity: relevant directories, review surfaces, syndication and social footprint, links or mentions, and entity records. Record collection coverage and distinguish a verified public record from a proposed placement. Off-site Authority has the largest current starting weight; that is not a measured causal share of citations.
2. On-page GEO — 26%
Inspect whether the submitted page contains clear, self-contained passages; specific and attributable claims; useful structure; and content that is not purely promotional. The Princeton-led GEO study supports evaluating quotations, statistics, and source citations in its benchmark setting. Do not convert that study into a promise for this page or prescribe arbitrary counts that the study did not establish.
3. Schema Engine — 18%
Parse the structured data and compare it with visible content. Check that types and properties are valid for the page, identity facts agree, and reviews, FAQs, authorship, dates, products, or services are not asserted only in markup. A validator result proves syntax and eligibility within its scope; it does not prove that an external engine crawled, indexed, or cited the page.
4. Entity / E-E-A-T — 18%
Inspect whether the publisher and authors are identifiable, credentials and ownership are stated where relevant, important claims are sourced, and first-party facts agree with independently controlled records. Record contradictions as findings. Do not infer that a cleaner entity record caused a sampled engine response.
5. Crawler Access — 10%
Inspect robots policy and whether important content appears in the public-origin response. Keep declared policy separate from observed access: an allowed user agent does not prove a crawler arrived, and a successful test fetch does not prove indexing or citation. If a JavaScript snippet supplies markup only after execution, label that client-side boundary; do not call it verified in raw public HTML.
Coverage and unknown are part of the result
Report which observations were present, absent, stale, failed, or unknown. A family with insufficient evidence should not become a confident zero, and small measured coverage should not be silently renormalized into a strong headline. When coverage is below the served threshold, insufficient data is the result. Preserve the score revision so later runs are compared only when the method is compatible.
Turn findings into governed, verifiable work
A score is a compact view of evidence, not the work order by itself. Prioritize concrete defects and dependencies: a malformed schema block before an enhancement to it; an unresolved canonical identity before copying that identity elsewhere; a blocked public-origin response before editing passages the scanner cannot read. Where evidence is incomplete or findings conflict, investigate before prescribing a change.
For every proposed effect, preserve the lifecycle:
- Authority: identify who may approve the exact revision and whether the connector can perform it.
- Provider receipt: record what the CMS, host, or third-party surface accepted.
- Independent reread: fetch the public origin again and compare the observed result with the approved revision.
- Client receipt: state what was proven, what remains unknown, and what was not attempted.
- Correction or withdrawal: preserve a governed path when the public result is wrong or no longer authorized.
A provider acceptance is not a public effect, and a public effect is not an engine outcome. Keep those receipts separate.
When to reassess
Reassess readiness when material evidence changes: after approved site work, a migration, a crawler-policy or CDN change, an entity rebrand, a connector change, or an off-site record correction. Compare only compatible score revisions and coverage. Keep the product’s monthly AI Presence sample on its own cadence; do not describe it as continuous monitoring or use it to overwrite readiness.
Frequently asked questions
What belongs in an AI Readiness audit?
The current model evaluates five evidence families: Off-site Authority, On-page GEO, Schema Engine, Entity / E-E-A-T, and Crawler Access. SEO Health and AI Presence are reported separately and are never blended into AI Readiness.
How is AI Presence different from AI Readiness?
AI Readiness reports setup evidence ClickRadius can inspect. AI Presence is a separate monthly sample of what ChatGPT, Gemini, Perplexity, Claude, and Grok returned for a disclosed question set. Neither score proves that ClickRadius caused an external engine outcome.
What should happen when audit evidence is missing?
Record the family or observation as unknown and report evidence coverage. Do not silently turn missing evidence into a zero, a pass, or a confident diagnosis. If coverage is insufficient to render readiness, the honest result is insufficient data.
Want the current evidence snapshot? Run the free AI Readiness Score for the submitted URL, then review the five family findings, evidence coverage, and unknowns before deciding what work is justified. See ClickRadius plans for governed ongoing work and separate monthly AI Presence reporting.