seo-geo Checked

5. **Page passage-level audit** WebFetch

• Pull the page HTML.
• Identify "passages" — paragraphs that could be extracted standalone (TL;DR boxes, definition paragraphs, summary sentences after H2s).
• For each passage, score citability:
• Has it a complete thought in 1–3 sentences?
• Does it answer a specific question (i.e., the question its parent H2 implies)?
• Has it a stat / number / named entity?
• Has it a clear timestamp or freshness signal?
• This is the citability layer.

6. **Compare candidate to cited sources**

• For each AIO query where candidate is NOT cited, identify the cited sources.
• WebFetch 2–3 of them.
• Extract the cited passage (often a snippet from the AIO answer).
• Compare passage shape: candidate vs cited. Surface specific structural / content / freshness gaps.

7. **Schema check** mcp__firecrawl-mcp__firecrawl_scrape

• WebFetch in step 5 returned markdown — JSON-LD blocks were stripped before parsing. The schema check requires Firecrawl to recover them.
• **If Firecrawl available:** scrape the target URL once (1 Firecrawl credit), parse the returned html for every <script type="application/ld+json"> block. Specifically check for: Article/BlogPosting with valid author + datePublished + dateModified; FAQPage if Q&A blocks present; BreadcrumbList; mainEntityOfPage self-canonical.
• **If Firecrawl unavailable:** write Schema check: skipped — Firecrawl required to parse JSON-LD blocks (WebFetch returns markdown only). into evidence/06-schema-check.md, mirror the same line in the GEO.md "Schema check" section. Don't infer from markdown — that's the bug this section closes.
• Schema isn't a direct citation signal but it correlates strongly with citation rates in Google's AIO.

8. **AI-protocol files** mcp__firecrawl-mcp__firecrawl_scrape

• **If Firecrawl available:** scrape https://{domain}/llms.txt and https://{domain}/.well-known/rsl.json (and the legacy /RSL.txt location as a fallback). Cost: 2 Firecrawl credits (one per file).
• For each file: capture HTTP status (200 / 404 / other), full body if present, and a parsed summary (declared content categories, allow/deny scope, attribution requirements).
• Surface in evidence/07-ai-protocol-files.md and in GEO.md as a new "AI-protocol files" section. These signal the domain's stance on LLM training and citation — present-and-permissive correlates with higher AIO citation rates.
• **If Firecrawl unavailable:** write AI-protocol files: skipped — Firecrawl not installed. Don't fall back to WebFetch (it would work for plain text but the integration stays uniform; runtime savings are negligible).

9. **Synthesise** GEO.md

Output format

Create a folder seo-geo-{target-slug}-{YYYYMMDD}/ with:

seo-geo-{target-slug}-{YYYYMMDD}/
├── GEO.md                            (synthesised report + recommendations — primary deliverable)
├── 04-page-passages.md               (extracted passages + citability scores — load-bearing reference editors consult)
├── 05-cited-source-comparison.md     (gap vs cited sources — load-bearing reference)
└── evidence/
    ├── 01-url-keyword-footprint.md   (URL overview + top keywords — raw step output)
    ├── 02-aio-by-keyword.md          (AIO presence + citation per keyword)
    ├── 03-leaderboards.md            (full leaderboards per keyword)
    ├── 06-schema-check.md            (JSON-LD audit for GEO-relevant types — requires Firecrawl)
    └── 07-ai-protocol-files.md       (llms.txt + RSL status and content — requires Firecrawl)

Top-level: GEO.md + 04-page-passages.md + 05-cited-source-comparison.md. The other step files preserve raw API/scrape outputs in evidence/ for reproducibility — editors and writers don't open them in the normal flow.

GEO.md follows this shape:

# GEO Analysis: {URL}

> Snapshot dated {YYYY-MM-DD} · Country: {country} · Keywords analysed: {n}

## Citation footprint

| Keyword | AIO present | Candidate cited | Citers |
|---|---|---|---|
| {keyword 1} | ✓ | ✗ | {3 cited sources} |
| {keyword 2} | ✓ | ✓ | {includes candidate + 2 others} |
| ... |

**Citation rate: {n}/{checked} ({%}) of AIOs where candidate could appear actually cite it.**

## Where the candidate IS cited
- {keyword X} — passage cited: "{passage text}"
- ...

## Where the candidate is NOT cited (and AIO is present)
- {keyword Y} — cited sources tend to share these patterns:
  - {pattern 1: short definitive answer in first 100 words}
  - {pattern 2: numbered stat with date}
  - {pattern 3: schema-marked Article with author bio}
- The candidate is missing: {specific gap}.

## Page passage-level audit

Top-scoring passages on the candidate (by citability score):
1. {passage at H2 "X" — score 8/10. Strong: definitive sentence, named stat. Weak: no date.}
2. ...

Lowest-scoring passages (refresh candidates):
1. {passage at H2 "Y" — score 3/10. Weak: vague generalities, no specific data.}
2. ...

## Schema check
- `Article` (or sub-type) present and valid: {✓/✗ | skipped — Firecrawl required}
- `author` populated with `@type: Person` and `url`: {✓/✗}
- `datePublished` + `dateModified` ISO 8601: {✓/✗}
- `FAQPage` for visible Q&A: {✓/✗/N-A}
- `BreadcrumbList`: {✓/✗}

## AI-protocol files
- `/llms.txt` present: {✓ status 200 / ✗ status {n} / skipped — Firecrawl required}
- `/.well-known/rsl.json` (or `/RSL.txt`) present: {✓ / ✗ / skipped}
- Stance summary: {permissive / restrictive / mixed / unknown — based on declared categories and allow/deny scope}

## Recommendations (top 5 to improve citability)

1. {Specific change — e.g., "A