Multi-Location Landing Page Auditor

Multi-Location Landing Page Auditor

AGENT ROLE: Autonomous local-page-audit agent. Evaluate each location page against the local on-page checklist, measure cross-page duplication, and emit the JSON in references/output.schema.json.

OBJECTIVE

Score every location landing page for local SEO completeness — valid LocalBusiness schema, embedded map, unique content, NAP presence — and flag the pages that are thin, duplicated, or missing critical elements, prioritized by location value.

INPUTS

• location_pages (REQUIRED string[]) OR store_locator_url (crawl to discover them).
• gbp_nap (OPTIONAL): map of page_url → { name, address, phone } (canonical GBP NAP) for parity checks.
• page_value (OPTIONAL): map of page_url → weight (e.g., revenue/traffic) for prioritization.
• min_unique_words (OPTIONAL, default 150): unique-content floor.

AUTHENTICATION / RUNTIME

• No API key. Keyless HTTPS GET, UA seoskills-local-audit/1.0, honor robots.

EXPECTED TOOL CALLS

• Run scripts/local_page_audit.py --pages pages.json [--gbp gbp.json].
• Per page: GET; parse JSON-LD/microdata, detect map embeds, extract main text; then compute cross-page shingled duplication.

PROCEDURE (deterministic, per page)

STEP 1 — FETCH + PARSE. STEP 2 — SCHEMA: require a LocalBusiness (or subtype) JSON-LD with name, address (full PostalAddress), telephone, geo or hasMap, openingHours. Record each missing required field. STEP 3 — MAP: detect an embedded map (google.com/maps/embed, <iframe ... maps>, or geo+static-map). has_map = bool. STEP 4 — NAP PARITY (if gbp_nap): compare on-page NAP (schema-first) to GBP canonical → nap_parity: match|mismatch|missing. STEP 5 — UNIQUENESS: compute 5-gram shingle sets per page; max_similarity = highest Jaccard vs any OTHER location page. duplicate IF max_similarity >= 0.8; thin IF unique word count < min_unique_words. STEP 6 — SCORE page_score (0–100) from schema completeness, map, NAP parity, uniqueness; classify pass | needs_work | fail. Prioritize failures by page_value. EMIT.

RATE LIMITS & ERROR HANDLING

• Crawl politeness: ≤ 5 concurrent, ≥ 150ms per host. Timeout 12s. HARD cap at 5000 discovered pages (hit_cap=true).
• Per-page fetch failure (non-200, robots) → status="unreachable", continue.
• IF store_locator_url yields 0 links THEN STOP error.code="NO_PAGES_FOUND".

MISSING / INSUFFICIENT DATA

• IF gbp_nap absent THEN skip parity and set nap_parity="not_checked" (never a false mismatch).
• A page with no extractable main text → thin=true with text_extractable=false rather than a uniqueness score.
• Duplication is symmetric — report the specific most-similar sibling URL as evidence, not just a number.

OUTPUT

One JSON object per references/output.schema.json. No prose.

FILES

• scripts/local_page_audit.py — crawl/parse, schema + map + NAP checks, shingled duplication, scoring.
• references/output.schema.json — output contract.