Site Audit — User Guide | YoBench
How to use the Site Audit module in YoBench: site crawling, Lighthouse scoring (Perf/A11y/BP/SEO), AI analysis of results, proxy and auth.
What the Site Audit module does
The module performs a complete site audit: it crawls pages with a headless Chromium that runs JS, runs the official Lighthouse suite (Performance / Accessibility / Best Practices / SEO), and optionally pipes the results through an LLM for a thorough breakdown. Works for both single-page audits and full-site crawls.
What you get:
- Full JS rendering — the browser executes scripts, so SPAs and lazy-loaded sites are analyzed correctly.
- Real Lighthouse — Google's official Lighthouse package, not a homegrown checker.
- AI analysis — collects page metadata and Lighthouse JSON into one prompt and asks the LLM for human-readable recommendations.
- Parallel crawling — up to 10 threads, up to 500 pages per job.
- Proxy and auth — headers, cookies, query parameters apply to both the browser and Lighthouse.
- Local reports — results live in the YoBench database, available offline.
- Infinite scroll on long reports — large crawls keep scrolling smoothly without pagination clicks.
Template parameters
A template is a reusable audit configuration. You set:
- Name — template label.
- URL — start page.
- Mode —
single(one page) orall(follow internal links, capped at 500). - Parallel threads — 1–10.
- Lighthouse —
onoroff(without it the audit is structural-only and faster). - Page timeout — 5–180 seconds.
- AI provider (optional) — which provider from AI Chat to use for analysis.
- Proxy (optional) — route traffic through a proxy profile.
- Auth profile (optional) — headers, cookies, query.
- AI prompt — a large default (~216 lines) that's easy to adapt to your checklist.
What is checked
Structural audit (cheerio)
- Title — length and presence.
- Meta — description, canonical, viewport, charset, lang, robots.
- Open Graph / Twitter Cards / Schema.org JSON-LD — presence and validity.
- Heading hierarchy — H1–H6, H1 count.
- Images —
altattributes. - Links — internal and external, deduplicated,
wwwnormalized.
Lighthouse (when enabled)
- Performance.
- Accessibility.
- Best Practices.
- SEO.
Lighthouse takes ~10–30 seconds per page. Run timeout: 60 seconds.
AI analysis
When a provider is configured, for each page (or the whole site) the module collects data and sends it to AI:
- URL, title, meta, headings, images, links, Schema.org.
- Lighthouse scores plus the list of failed audits (score < 50).
- Base prompt + your data.
The output is a detailed Markdown report in the UI language (RU/EN/DE/FR/ES/PT/ZH-CN) with sections for SEO, content structure, images, links, technical setup, Lighthouse metrics and failed audits.
Global settings
There are no dedicated siteAudit* keys in the central settings. The module uses shared resources:
language— language of AI reports.- AI providers, proxies, auth profiles — app-wide registries.
Workflow
1. Create a template
- Open the Site Audit module from the left sidebar.
- On Templates click Create template.
- Fill in name, URL, mode, threads, timeout.
- Toggle Lighthouse on, pick an AI provider, proxy, auth profile.
- Optionally edit the prompt.
- Save.
2. Run an audit
The Run button next to the template creates a report with running status. In all mode the crawl proceeds within the configured thread limit. Only one audit can run at a time.
3. Review the report
On the Reports tab you see:
- Site-wide Lighthouse scores.
- Per-page list with metrics and status.
- Page detail panel: HTML breakdown, Lighthouse categories, AI analysis.
- A Retry analysis button reruns the analysis for a single page (handy after editing the prompt).
4. Use the results
- The AI breakdown gives concrete recommendations (headings, meta, scoring).
- Lighthouse failed audits highlight what's most critical.
- Lighthouse JSON and structural data are persisted — you can pull them into AI Chat for follow-up questions.
Next steps
- Connect AI providers — without them you get Lighthouse + structural report only.
- For deep parsing of non-standard pages, use the Web Parser.
- For ongoing uptime checks, run Health Check alongside.
Help and feedback
Want scheduled audits or run-to-run comparison? Contact us via the feedback form.