Transparent scoring
How the AI agent readiness scan works
Can AI Use My Site? runs a point-in-time technical review of public website signals that can affect discovery, retrieval, interpretation, and reliable interaction. It is a diagnostic—not a promise that any particular AI product will visit, cite, recommend, or transact with a site.
AI search
Can an automatic search crawler discover and retrieve the public page?
User-directed access
Can a fetcher or browser agent reach and understand the page for a person?
Model development
Does the site express a separate policy for model-training or development use?
What the scanner requests
The scanner starts with the public URL you submit. Within a fixed request budget, it may request the homepage, robots.txt, llms.txt, declared or common sitemap locations, selected links referenced by llms.txt, and a small set of emerging well-known files. It does not log in, submit forms, buy products, or change the scanned website.
What is checked
Access and discovery
robots.txt directives, selected AI-related user agents, sitemap discovery, HTTP availability, and obvious bot-specific access differences.
Machine-readable meaning
Valid JSON-LD, useful Schema.org types, page titles, descriptions, headings, landmarks, links, form labels, and accessible names.
Content delivery
Whether meaningful content appears in the initial HTML or the page resembles a JavaScript-only shell.
Optional and emerging signals
llms.txt, selected experimental agent files, detected tool declarations, and optional public Chrome UX layout-shift data.
How scoring works
Each applicable scored check occupies one stable scoring slot. A passing result earns that slot; warnings and failures do not. Informational findings and checks that cannot be measured are excluded from the denominator. The displayed percentage is the number of passed scored checks divided by all applicable scored checks.
- Pass: the tested signal was found and met the scanner rule.
- Warning: the signal is incomplete, ambiguous, or worth manual review.
- Fail: the tested signal was missing or did not meet the rule.
- Information: useful context that does not affect the score.
How evidence earns weight
A fashionable signal does not automatically belong in the score. Checks are classified by the strength of the evidence behind them.
Web standard
A recognized or widely established web mechanism with deterministic behavior.
Vendor-documented control
A first-party instruction for the crawler or product that consumes it.
Proposal or experiment
A credible emerging idea whose adoption or effect is not yet broadly settled.
Scanner heuristic
A useful inference that requires careful wording and false-positive monitoring.
Important limitations
AI systems differ in their crawlers, retrieval methods, policies, capabilities, and respect for site directives. The scan is mainly based on fetched HTML and public protocol files; it is not a full browser automation test and does not reproduce every AI agent. Results can also change when the scanned site, its CDN, or its security rules change.
Use the findings as a prioritized engineering checklist, then verify important fixes with the relevant vendor documentation, accessibility testing, structured-data tools, server logs, and real user journeys.
Saved reports and privacy
Saved reports are private and marked not to be indexed by default. They expire under the configured retention period, and the management link in the report email can permanently delete a report earlier. Read the privacy notice for details.