Product Feed SEO Optimizer

Product Feed SEO Optimizer

AGENT ROLE: Autonomous product-feed agent. Validate identifiers and attributes, score and rewrite titles against a demand-informed pattern, generate Product schema, join Merchant Center disapprovals, and emit the JSON in references/output.schema.json. One diff that improves organic and Shopping together.

OBJECTIVE

For every product in a feed, return: identifier validity (real GTIN check digit), attribute completeness, a title score with a rewritten demand-informed title (diff), recommended Product JSON-LD, and any Merchant Center disapproval — so the feed ranks in Shopping and the landing pages win rich results.

INPUTS

• feed (REQUIRED via --feed): a Google Merchant product feed as TSV/CSV (id, title, description, link, image_link, availability, price, brand, gtin, mpn, condition, color, size, ...) or JSON array of product objects. Format is auto-detected.
• volumes (OPTIONAL via --volumes): JSON map of term -> monthly volume; drives demand-informed title front-loading.
• title_pattern (OPTIONAL, default "{brand} {title} {color} {size}"): tokens {brand} {title} {color} {size} {material} {gender} {category}.
• title_max (OPTIONAL, default 150): Merchant title limit; title_optimal (OPTIONAL, default 70): organic visible length.
• merchant_id (OPTIONAL via --merchant-id): when set, disapprovals are pulled from the Content API for Shopping.
• max_products (OPTIONAL, default 100000) and max_status_pages (OPTIONAL, default 40): cost guards.

AUTHENTICATION (Content API for Shopping)

1. The feed file is local and needs no credentials.
2. IF --merchant-id is passed THEN REQUIRE env CONTENT_API_ACCESS_TOKEN (OAuth bearer, content scope). IF unset THEN STOP error.code="AUTH_MISSING_CONTENT_API". IF the token is rejected THEN STOP error.code="AUTH_CONTENT_API_FORBIDDEN". Calls productstatuses.list.

EXPECTED TOOL CALLS

• Run scripts/product_feed_optimizer.py --feed feed.tsv [--volumes volumes.json] [--merchant-id 1234567] [--title-pattern "{brand} {title} {color} {size}"].
• Zero network calls in the offline path; paginated productstatuses.list (250/page) only when --merchant-id is set.

PROCEDURE (deterministic, per product)

STEP 1 — IDENTIFIER: strip non-digits from gtin; validate length in {8,12,13,14} and the GS1 mod-10 check digit. Status = gtin_valid / invalid_gtin / brand_mpn / identifier_exempt (when identifier_exists=false) / missing_identifier. STEP 2 — ATTRIBUTES: check the required Merchant set (id, title, description, link, image_link, availability, price, condition, gtin-or-mpn); add color/size/gender/age_group for apparel categories; compute coverage and the missing list. STEP 3 — TITLE AUDIT: penalize over-limit length, over-optimal length (info), promotional ALL-CAPS, promotional phrases, brand not front-loaded, and keyword stuffing; produce a 0-100 title score. STEP 4 — REWRITE: compose a title from title_pattern; IF volumes given THEN front-load the highest-volume phrase the product actually matches; truncate at a word boundary to title_max. Emit old vs new and title_changed. STEP 5 — SCHEMA: generate recommended Product JSON-LD (name, brand, sku, mpn, valid gtin, image, description, offers{price, priceCurrency, availability, itemCondition, url}). STEP 6 — DISAPPROVALS: IF --merchant-id THEN map each product id to its disapproved item-level issues from the Content API. STEP 7 — EMIT products sorted worst-coverage-first with feed-level rollups.

RATE LIMITS & ERROR HANDLING

• Content API 429 -> backoff 2^attempt (max 5) then STOP RATE_LIMITED; 5xx retry <=3 then STOP REQUEST_FAILED; pages are fetched sequentially with pacing (concurrency 1) and capped at --max-status-pages.
• 401/403 from the Content API STOP AUTH_CONTENT_API_FORBIDDEN.
• An empty or unparseable feed STOPs error.code="EMPTY_FEED".

MISSING / INSUFFICIENT DATA

• WITHOUT --volumes titles are rewritten structurally only (demand_informed=false, demand_lead_keyword=null); no volume is invented and no keyword is front-loaded on a guess.
• WITHOUT --merchant-id disapprovals are not fetched (disapprovals_source="none"); the audit never claims a product is approved or disapproved.
• schema_status="recommended" reflects that JSON-LD is generated from feed data; the skill does not assert the markup is live on the page unless that is separately verified.
• A GTIN that is absent is missing_identifier (or brand_mpn/identifier_exempt), never silently treated as valid; a present-but-wrong GTIN is invalid_gtin.

OUTPUT

One JSON object per references/output.schema.json.

FILES

• scripts/product_feed_optimizer.py — feed parsing, GS1 GTIN validation, attribute + title audit, demand-informed title rewrite, Product JSON-LD generation, Content API disapproval join.
• references/output.schema.json — output contract.