Grounding Audit

Grounding Audit answers the question: "Would an LLM even pull this page as a source — and if not, why?" You drop in a URL, a list of URLs, or a sitemap.xml. Rankion fetches the pages, runs them through a rule-based check pipeline, and returns two scores plus structured findings with a concrete fix per issue — everything async via REST.

rankion.aiTwo scores, not one (since 2026-05-16)

A single audit now produces three numbers:

• technical_score — Technical Eligibility (0-100). What Google itself publicly confirms as a precondition for AI Overviews: indexable, crawlable, no noindex, clean HTTP status, language detectable, valid markup. Evidence Tier A — we say with high confidence what has to be true.
• content_score — Content/GEO Signals (0-100). Probabilistic signals that correlate with citations from the extractive engines (ChatGPT Search, Perplexity, Claude, Copilot, Grok): entity-first structures, atomic definitions, FAQ repetition, dated volatile facts, disambiguation. Tier B — hypothesis-grade, not engine-confirmed.
• score (Legacy 0-100). A reweighted compatibility number. Prefer the two new scores for new workflows.

rankion.aiEngine Capability Matrix — why two scores

The LLM engines work differently:

• Google AI Overviews / Gemini. Source = Google index. What Google wants for SEO applies here too — eligibility is high-confidence Google-confirmed. Google explicitly says llms.txt, AI-specific schemas, and semantic chunking do nothing for Google.
• ChatGPT Search / Perplexity / Claude / Copilot / Grok. Extractive engines with a browser tool. The same "GEO hacks" are plausible for them, but not publicly confirmed — they are Tier B (hypothesis).

We don't sell hypothesis as fact. Every finding carries an evidence_tier (A or B). Backlog priority: Tier-A findings with severity in ['critical','high'] first, then Tier B.

rankion.aiWhat it can do

• Single-URL audit — submit one URL, get 202 + audit_id back, poll for status.
• Bulk audit — dispatch up to several hundred URLs in one batch, shared batch_id, shared webhook.
• Sitemap audit — drop in a sitemap.xml, Rankion parses it (including sitemap-index recursion), filters, dispatches the batch.
• NDJSON stream — stream batch results line by line.
• HMAC-signed webhooks — batch finishes → POST to callback_url with X-Rankion-Signature (SHA-256, 4 retries).
• Structured findings — per issue: id, severity, framework, dimension (technical|content), evidence_tier (A|B), signal, title, description, fix.type, fix.action, spec URL.

rankion.aiWorkflow

1. Start a single audit — POST /v1/grounding/analyze with {url, frameworks?:["v1.5"]}. Returns 202 + audit_id + poll_endpoint.
2. Poll — GET /v1/grounding/audits/{id} until status=completed. Returns technical_score, content_score, score (legacy), tier, findings[], raw_text.
3. Bulk — POST /v1/grounding/batch with urls[] + optional callback_url. Returns batch_id + callback_secret (one-time — store it!).
4. Sitemap — POST /v1/grounding/sitemap-audit with sitemap_url + filters. Credits are charged AFTER parse.
5. Fetch results — GET /v1/grounding/batches/{id} (summary) or …/results.ndjson (line-by-line stream).
6. Prioritize — filter findings by evidence_tier (A first), then by severity.

rankion.aiAPI

Throttling: analyze 30/min, batch + sitemap 5/min, polling 120/min. Auth: Sanctum token.

rankion.aiKnown limits — honest

• eeat + people-first frameworks are stubs in the REST pipeline today — v1.5 is the framework that runs end-to-end.
• Tier-B signals are hypotheses. Anyone selling Tier-B findings as "proven" is lying. They are plausible and consistent with what extractive engines seem to prefer — but nobody outside the engine vendor knows for sure.
• Per-batch concurrency is not guaranteed. Effective concurrency = 5 workers server-wide.
• No auto-refund on audit failures. Sitemap audits debit credits AFTER parse.
• TTL — audit results soft-expire after 30 days, hard-delete after 90 days.

rankion.aiRelated modules

• Engine Capability Matrix — the SSOT that the evidence tiers come from. Explains what Google confirms vs. what we hypothesize for the extractive engines.
• Grounding Check Validation — empirically measures which gp.* checks actually correlate with citations on YOUR pages.
• AI Visibility Tracking — measures whether your pages actually land in LLM answers. Grounding Audit explains why (or why not).
• Content Audit — classic SEO issues.
• Page Deep Audit — deeper SEO/performance check per URL.
• Agentic Chat — the master agent can dispatch grounding audits directly from chat.