Merchant Feed

Multi-Channel Merchant Feed Pipeline

You build a feed-as-product-data discipline. In 2026, feed management is less about producing "a feed" and more about running a data product: continuously modifying attributes, enriching missing fields, validating quality, and tying changes to performance.

============================================================ === PRE-FLIGHT ===

• Product source of truth: Shopify, BigCommerce, WooCommerce, Magento, custom DB, ERP (NetSuite, SAP).
• Target channels: which of GMC / Meta / TikTok Shop / Pinterest / Bing / Snap?
• Catalog size: < 500 SKUs → single feed file is fine. 5k+ SKUs → use Content API / Commerce API for incremental updates.
• Product types: physical goods, digital, services, software, age-restricted (alcohol/tobacco/cannabis — extra scrutiny). Subscription products require additional Google attributes (subscription_cost).
• Brand assets: high-res images, GTINs, MPNs. Without GTIN, disapprovals climb 30-50%.
• Localization: feed per country / language? Currency? Local tax/shipping?

Recovery:

• Missing GTINs: flag SKUs with identifier_exists=false (Google allows this for niche brands) but expect lower performance.
• Image quality issues: route through a transform CDN (Cloudinary, imgix, Vercel Image Optimization) to auto-meet 500×500 minimum + AVIF/WebP output.

============================================================ === PHASE 1: CANONICAL PRODUCT SCHEMA ===

Build the single source of truth. Every channel feed projects from this:

{
  "id": "SKU-12345",
  "title": "...", // primary attribute for Google match
  "description": "...",
  "link": "https://...",
  "mobile_link": "https://...", // optional but recommended
  "image_link": "https://.../primary.jpg", // ≥ 500×500 (2026 requirement)
  "additional_image_link": ["...", "..."], // up to 10
  "availability": "in_stock|out_of_stock|preorder|backorder",
  "availability_date": "2026-08-15", // if preorder
  "price": "29.99 USD",
  "sale_price": "24.99 USD",
  "sale_price_effective_date": "2026-05-23T00:00-07:00/2026-06-30T23:59-07:00",
  "brand": "...",
  "gtin": "00012345678905",
  "mpn": "...",
  "identifier_exists": true,
  "condition": "new|refurbished|used",
  "google_product_category": "Apparel & Accessories > Clothing > Shirts & Tops",
  "product_type": "Mens > Shirts > T-Shirts",
  "color": "navy",
  "size": "L",
  "material": "100% cotton",
  "gender": "male|female|unisex",
  "age_group": "adult|kids|toddler|infant|newborn",
  "item_group_id": "TSHIRT-001", // for variant grouping
  "shipping": [{ "country": "US", "service": "Standard", "price": "4.99 USD" }],
  "tax": [{ "country": "US", "region": "CA", "rate": 9.0, "tax_ship": true }],
  "custom_label_0": "high_margin", // for ad bidding strategy
  "custom_label_1": "summer_collection",
  "custom_label_2": "best_seller",
  "custom_label_3": "new_arrival",
  "custom_label_4": "low_inventory",
  "video": "https://.../product.mp4" // critical for TikTok Shop
}

Persist as products.json with versioning (so feed history is auditable).

VALIDATION: 100% of products have id, title, description, link, image_link, availability, price, brand. ≥ 90% have GTIN.

============================================================ === PHASE 2: TITLE OPTIMIZATION ===

Title is the highest-leverage attribute for Google matching. Use this structure:

[Brand] [Product Type] [Key Attribute] [Variant] [Size/Quantity]

Examples:

• ❌ "Awesome T-Shirt"
• ✅ "Nike Dri-FIT Men's Running T-Shirt — Navy, Large"
• ✅ "Apple AirPods Pro (2nd Gen) — USB-C, Active Noise Cancellation"

Title cap: 150 chars (Google), 65 chars displayed in shopping results. Front-load the most important info.

Maintain a per-product title A/B test queue: rotate one new variant per category per month, measure CTR + ROAS at the keyword level via Search Terms Report.

VALIDATION: Median title length 60-120 chars. No "AAA-promo" stuffing or all-caps.

============================================================ === PHASE 3: PER-CHANNEL FEED PROJECTION ===

Project the canonical product into each channel's format:

Google Merchant Center (XML or TSV):

• Full canonical schema. Required: id, title, description, link, image_link, availability, price, brand, gtin (if available), condition, google_product_category.
• Sale price requires sale_price_effective_date to qualify as a deal.
• Feeds via Content API for Shopping (real-time) or scheduled fetch (8-24 hour cadence).

Meta Catalog (CSV or Catalog Batch API):

• Similar fields; uses availability (lowercase: in stock / out of stock).
• inventory (numeric count) for ads with low-stock urgency.
• Additional Meta-specific: applinks for app-deep-link, rich_text_description for Shops.
• home_listing for real estate variant, flight for travel variant.

TikTok Shop (TSV or Commerce API):

• Video required for top placements. Map to tiktok_video_url field.
• Stricter content moderation: no medical claims, no age-restricted products in many regions.
• Stricter image requirements (white background preferred for product imagery).

Pinterest Catalogs:

• Image quality matters most (Pinterest is image-first).
• additional_image_link (5+) outperforms single-image listings.
• Pin link landing must match feed link exactly.

Bing Shopping (close to GMC format):

• Reuses Google product feed mostly; small differences in attribute names.

Snap Catalogs:

• Use Snap's tags (age_group, gender) for AR try-on eligibility.

Generate per-channel projection in feeds/{channel}/{date}.{ext}.

VALIDATION: Each channel feed validates against the platform's spec (use vendor schema validators / spec linters).

============================================================ === PHASE 4: VALIDATION & DISAPPROVAL PREVENTION ===

Before submission, run pre-flight checks:

Output feed_validation_{date}.md with per-SKU issues.

VALIDATION: Pre-submission validation catches > 95% of issues that would otherwise be caught downstream.

============================================================ === PHASE 5: WEEKLY FEED AUDIT CRON ===

Schedule a weekly task (Mondays, post-weekend sales data):

1. Fetch disapproval reports from GMC, Meta Commerce Manager, TikTok Shop, Pinterest.
2. Map each disapproval to the underlying canonical product issue.
3. Auto-fix the deterministic ones (image upscale, title tightening, category mapping).
4. Surface non-deterministic ones (policy violations, restricted content claims) for human review.
5. Track 48-hour SLA: every new disapproval must be addressed within 48h.
6. Track custom_label rotation: quarterly review of which custom labels are working (e.g., high-margin label producing top ROAS).

Generate weekly_audit_{date}.md digest.

VALIDATION: SLA met for ≥ 95% of disapprovals.

============================================================ === PHASE 6: PERFORMANCE FEEDBACK LOOP ===

Tie feed changes back to performance:

• Pull impressions, clicks, conversions, cost, revenue per SKU per channel.
• Compute: CTR, conversion rate, ROAS, CPA.
• Flag underperformers (bottom 20% by ROAS) for title test or image refresh.
• Flag overperformers (top 5%) for custom_label tagging → push to higher-priority campaigns.

VALIDATION: Performance attribution joins canonical product ID to channel SKU ID without drift.

============================================================ === PHASE 7: OUTPUT PACKAGE ===

merchant-feed/
├── README.md
├── products.json                # canonical
├── feeds/
│   ├── google/products.xml
│   ├── meta/products.csv
│   ├── tiktok/products.tsv
│   ├── pinterest/products.csv
│   ├── bing/products.xml
│   └── snap/products.csv
├── validation/
│   └── feed_validation_{date}.md
├── audit/
│   └── weekly_audit_{date}.md
├── tests/
│   └── title_ab_tests.csv       # active title variants per SKU
└── perf/
    └── sku_performance_{date}.csv

VALIDATION: Every channel feed validates and submits without disapproval.

============================================================ === SELF-REVIEW ===

• Complete: All 7 phases run. Canonical schema → 6 channel projections + validation + audit + perf loop.
• Robust: Handles missing GTINs, image upscaling, channel-specific edge cases?
• Clean: Disapproval SLA tracking, no feed drift between channels?
• Ecommerce-credible: Would a feed manager at a $100k+/month Shopping spend accept the system?

Common gap: image upscale missing → silent quality degradation. Verify CDN transform pipeline.

============================================================ === LEARNINGS CAPTURE ===

~/.claude/skills/merchant-feed/LEARNINGS.md.

============================================================ === STRICT RULES ===

• Never submit images below 500×500 for Google after Jan 2027. Auto-upscale or omit.
• Never silently drop attributes a channel requires. Surface in validation report.
• Never let disapprovals sit > 48 hours. Compound spend loss.
• Never reuse one feed across channels without per-channel projection. Each platform has unique requirements.
• Always preserve canonical → channel ID mapping for performance attribution.