What a Seryvon audit can and cannot tell you
A Seryvon audit is a structured inspection of evidence collected within a defined scope. Its scores help identify and prioritise issues. They do not establish a search engine's internal assessment, predict a citation probability or guarantee that an autonomous agent will complete a task.
A score describes measured criteria
The engine evaluates rules against collected signals. A high score means that the measured criteria performed well under that rule set. It does not prove that every page was crawled, every provider was queried or every commercial use case was tested. Read the coverage information alongside the number, especially when comparing sites with different configurations.
not_measured means that required evidence is unavailable; it remains relevant to coverage but is excluded from scoring. not_applicable means the criterion is not relevant in that evaluation and is excluded from eligibility as well. Neither should be rewritten as a successful check. A missing provider key, an unsuccessful request and a genuine negative observation require different explanations.
The crawl is a sample with boundaries
Crawl depth, page limits, robots rules, redirects, availability and discovery paths determine which URLs enter an audit. The default configuration limits a crawl to 300 pages and depth three; operators can change these values. A clean sampled result cannot certify a larger site. Record the actual scope and examine important templates that may not have been reached.
JavaScript rendering is optional. A static fetch may not contain the content a visitor sees after scripts run. Browser rendering can improve that view, but it does not reproduce every login state, location, consent choice or interaction. Inspect both raw and rendered content when an important finding depends on that distinction.
Readiness and observed visibility are different
Clear headings, accessible content and explicit entities can be inspected directly. Whether an AI service cites a page requires an observation from that service. A structured answer is not evidence that it appeared in an answer engine. ASO checks likewise assess inspectable agent-facing signals; they are not proof of a completed purchase, form submission or other real-world action.
Citation observations depend on the prompt set, provider, model, time and available search context. A domain absent from one response is absent from that sample, not from every possible answer. Repeat a consistent set of business-relevant prompts and retain the source responses before drawing a trend.
Reproducible scoring needs comparable inputs
The deterministic part is the calculation from fixed inputs, rules and configuration. Two fresh crawls can collect different evidence because the site or an external service has changed. Preserve the report, measurement profile and relevant artifacts to explain a result.
The comparison implementation distinguishes exact, compatible, intersection and incompatible profiles. Adding a connector can change coverage without improving the website. Use an intersection comparison when appropriate; if rules or thresholds change incompatibly, describe the two runs separately. See the scoring methodology for the calculation and comparison contract.
Turn findings into decisions carefully
Prioritisation should consider the affected page, evidence, business importance and implementation effort. Validate a critical recommendation before making a site-wide change. Then check the deployed result, repeat a comparable audit and track the intended business outcome separately. A higher audit score alone does not establish more qualified traffic, leads or revenue.
These boundaries follow the status definitions, crawl configuration and comparison implementation. Use the illustrative report to see how to communicate them, and the security guide before running shared infrastructure.