seo-page Checked

3. **Ranking keywords** DATA_getDomainKeywords (use the url param for exact match — not filter_url)

• Pull every keyword the URL ranks for in the target country, with positions.
• Sort by traffic-weighted score: volume × CTR-by-position (use a standard CTR curve: 1=28%, 2=15%, 3=11%, 4=8%, 5=7%, 6=5%, 7=4%, 8=3%, 9=2%, 10=2%, 11+=1%).
• Take the top 3–5 as the URL's "primary keywords" for SERP work in step 5.

4. **Page authority** DATA_getPageAuthority + DATA_getPageAuthorityHistory

• Current PA score and its 12-month trajectory.
• Flag any drop > 5 points in the last 90 days.

4b. **GSC URL performance + URL Inspection** *(only if google-api.json is present, tier ≥ 1)*

• Replaces the directional traffic estimate from step 2 with first-party Google data for the exact URL.
• Pull GSC search analytics for the URL (last 28 days, by query and page):

python3 scripts/gsc_query.py --property "{config.default_property}" --url "{target_url}" --days 28 --json

• Pull URL Inspection (real indexation status, canonical Google sees, last crawl date):

python3 scripts/gsc_inspect.py "{target_url}" --site-url "{config.default_property}" --json

• If the URL's domain isn't a verified GSC property: surface "GSC: {target_domain} not verified — add it in Search Console" and continue with SE Ranking data only.
• Surface in PAGE.md "## Snapshot": GSC last 28d: {clicks}/{impressions}/{ctr}% CTR / pos {position} and Google sees: {INDEXED|EXCLUDED} · canonical {userCanonical} → {googleCanonical} · last crawled {date}.
• **Feed into the verdict heuristic:** INDEXED + impressions > 100 + position 4–10 → harden REFRESH (clear quick-win). EXCLUDED (any reason) → harden KILL or CONSOLIDATE. userCanonical ≠ googleCanonical → flag as critical issue regardless of verdict.
• See skills/seo-google/references/cross-skill-integration.md § "seo-page" for the full recipe.

5. **SERP context** DATA_getSerpResults and DATA_getAiOverview

• For each of the 3–5 primary keywords:
• Top 10 organic results (URL, title, snippet).
• SERP features present (PAA, image carousel, video, shopping, etc.).
• AIO presence and citations — is this URL cited in the AIO?

6. **HTML sense-check** WebFetch (always) + mcp__firecrawl-mcp__firecrawl_scrape (when available)

• **WebFetch first** (free, instant): extract <title>, meta description, all <h1..h6>, lang, word count, internal-link count, image count. WebFetch returns markdown so anything in <head> beyond <title> is lost — the next bullet recovers it.
• **Firecrawl second** (1 Firecrawl credit; pass formats: ["rawHtml"]) — recovers what WebFetch can't see. Pin the format to rawHtml; the default html is post-processed and silently strips canonical, hreflang, and <script type="application/ld+json"> blocks on many sites. If those head fields come back zero on a site that obviously has them (any major SaaS homepage), you forgot the rawHtml flag — re-run.
• From metadata: og:title, og:description, og:image, twitter:card, viewport, robots meta, canonical URL.
• From the returned rawHtml: every <script type="application/ld+json"> block. Parse each as JSON, list detected @types (Article, BreadcrumbList, Organization, Product, etc.). Note any block that fails to parse.
• hreflang: count <link rel="alternate" hreflang="…"> occurrences in rawHtml.
• **If Firecrawl unavailable:** the WebFetch portion still runs; populate Page basics' OG / Twitter / canonical / robots / JSON-LD / hreflang lines as (skipped — Firecrawl not installed). Don't infer from markdown.
• **Sense-check:** does the page actually talk about its top-ranking keyword in title and H1? If a page ranks for keywords it doesn't address textually, that's a strong consolidation signal.

7. **Domain context** DATA_getDomainOverviewWorldwide (parent domain)

• Light-weight: parent DA, total keywords, total traffic. Used to contextualise the page's PA against its domain.

8. **Cannibalization check** DATA_getDomainPages

• Pull the parent domain's pages ranked by organic traffic. Cap at the top 50 — that's where the cannibalization risk concentrates.
• For the candidate URL's top-3 traffic-weighted keywords (from step 3), scan the peer-page list: does any other URL on the same domain rank in the top 20 for any of those keywords?
• **Cannibalization signal:** any peer URL ranking ≤ 20 for the candidate's top-3 keywords. Record peer URL, keyword, peer position vs candidate position, peer traffic.
• **Cheaper than the old approach.** Previous versions cross-checked via DATA_getDomainKeywords on the parent domain — that endpoint can return tens of thousands of rows on large sites and is unnecessarily heavy for this question. DATA_getDomainPages ranked by traffic surfaces the high-impact peers directly.
• If no peer URL competes, this signal is "no cannibalization detected" — pass through to step 9.

9. **Synthesise verdict**

• Apply the verdict heuristic (see Tips) to produce KEEP / REFRESH / CONSOLIDATE / KILL.
• For REFRESH, identify the top 3 specific changes (e.g., "add an H2 on 'alternatives' which 7 of 10 SERP winners use").
• Write PAGE.md (output spec below).

Output format

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

seo-page-{target-slug}-{YYYYMMDD}/
├── PAGE.md                       (synthesised verdict + plan — primary deliverable)
├── keywords.csv                  (full keyword list with positions — load-bearing CSV the auditor walks through row-by-row)
├── 04-serp-context.md            (top 10 + AIO for top 3–5 keywords — load-bearing reference for SERP-driven REFRESH discussions)
└── evidence/
    ├── 01-url-overv