Features for measurable visibility
Seryvon is an open-source web visibility audit tool that measures whether a site can be found, understood, cited and chosen across five pillars: SEO, GEO, GSO, AEO and ASO. It combines a website crawler, explicit scoring rules and reports that connect findings to measured signals. The open-core release is public; the managed product is coming soon.
Core features
- Five-pillar scorecards: review SEO, GEO, GSO, AEO and ASO together, with separate scores and coverage.
- Asynchronous crawling: discover pages and extract technical, editorial and structured signals.
- Deterministic scoring: calculate results from the same signals and configuration using inspectable rules.
- BYOK citation tracking: configure your own provider keys for supported language-model connectors.
- Report workflows: use JSON, HTML and Markdown output, with PDF generation also present in the API.
- Self-hosted evaluation: use the CLI or Docker workflow documented in the repository.
Inspect what the crawler actually collected
The audit orchestrator combines discovery, page extraction and external signals before scoring. This matters when a finding depends on access: a missing signal may reflect unavailable content or a restricted crawl rather than an editorial omission. Review the crawled pages before treating any site-level conclusion as comprehensive.
For a documentation site, a useful evaluation is to inspect a representative set of pages, identify repeated metadata or structural findings and check those findings against the actual templates. Seryvon supplies audit evidence; deciding how to change the templates remains a development task.
Investigate citations with explicit settings
The source includes OpenAI, Anthropic and Gemini connectors alongside a Perplexity connector. Citation workflows expose provider-dependent observations, not a universal answer to whether a brand is visible across every AI service. Model, prompt, retrieval mode and repetition settings affect what is measured.
Use a citation dry run to inspect the planned measurement before sending paid requests. A brand mention is different from a linked source citation. A response generated without web retrieval should not be interpreted as proof that an engine fetched your current page. Provider access, quotas and errors must remain part of the interpretation.
Prioritize findings and compare carefully
The action-plan implementation orders warning and critical findings using impact, severity and estimated effort. That makes it useful for preparing a review backlog, but the priority value is not a forecast of traffic or revenue.
The comparison implementation checks measurement profiles and supports strict, intersection and descriptive modes. For example, comparing runs with different active connectors may require restricting the comparison to shared criteria. Changing scoring thresholds can make a numerical comparison inappropriate.
Current status
Comparison is present in the open-core source, including API and dashboard entry points. Managed hosting, continuous monitoring and collaboration should not be assumed available from that code. The managed product is coming soon.
Coverage also affects the current scoring calculation: missing measurements are not estimated, but reduced coverage can lower a pillar score. Review the measurement limitations before using results as acceptance criteria. Static agent checks do not demonstrate that an agent completed a real transaction.