Audit your product feed

How the Marpipe Feed Auditor works

The Marpipe Feed Auditor scores your product feed across the seven largest advertising channels — Meta, Google, Snap, Pinterest, TikTok, Axon, and Reddit — and turns opaque catalog data into a defensible plan you can act on. Drop a CSV or paste a hosted feed URL and the auditor returns a 0–100 score per channel, a per-field breakdown of what's strong and what's dragging the score, and a ranked list of fixes with the projected lift in points for each one.

What gets scored

Each channel score is a weighted blend of three independent components. A feed can be 100% required-field-complete and still score in the 60s because of shallow taxonomy, missing labels, or junk values in populated fields — so we score each component separately and surface where the actual drag is.

Required coverage (40% of channel score)

For every field the channel requires at ingestion, what percent of SKUs have a non-null, format-valid value? Format validity is doing real work here — a price of 19.99 without a currency code doesn't count as present for Google, and an availability value of Available doesn't map to any channel's standard enum. Required failures are the most urgent because SKUs that fail required validation don't ship at all.

Recommended coverage (30% of channel score)

Optional fields that affect ranking, faceted browsing, and audience targeting. Pinterest weighs GTIN and brand for product matching; Meta and Google use gender and age_group for apparel categorization; channel-specific custom labels unlock bid strategy flexibility. Recommended fields are weighted independently per channel — the same field can be weight 1.4 on Pinterest and 0.8 on Snap.

Field utilization quality (30% of channel score)

For high-leverage fields like google_product_category, GTIN, and title, we score not just whether the field is populated but how well. Google's product taxonomy rewards depth — level 4 categories get full credit, level 2 only 40%. GTINs are checked against the GS1 check-digit algorithm. Titles are scored on length, brand prefix presence, and front-loaded keyword density. A populated field still costs you score points if it's a sentinel value, a literal non-GTIN string, or only one level deep.

Channels we audit

Each channel is scored against its own published spec. The full field list per channel — required, recommended, and quality-scored — is available on the channel reference page.

• Meta — DPA-style catalog with brand, condition, item_group_id, and apparel attributes.
• Google — the most prescriptive spec; GTIN required, deep GPC heavily rewarded.
• Snap — accepts Google-format feeds with a slimmer required field list.
• Pinterest — long descriptions and brand/GTIN drive product matching.
• TikTok — title quality is front-loaded; first 70 chars carry the weight.
• Axon — scored against Marpipe's canonical schema with unique variant identifiers required.
• Reddit — Google-format compatible with extra weight on title length.

Recommendations engine

Beyond the score, the auditor produces a ranked list of fixes — what to change, the specific lift in points each fix would unlock, the tier of effort required, and which channels benefit. Lift numbers come from the same scoring math used for the channel scores, so applying a fix and re-auditing actually moves the number. Tiers map to implementation strategy:

• Tier 1 — deterministic rules. Whitespace, casing, currency formatting, removing known-bad sentinel values. Auto-applies cleanly.
• Tier 2 — rules with AI assist. Mapping rich tag/collection data into custom labels, deriving gender/age_group from category signals. Proposed rules reviewed before applying.
• Tier 3 — preview-then-apply with agentic AI. Deepening Google product categories from product_type + title, extracting color from titles and SKU codes. 20-SKU preview before bulk apply.
• Tier 4 — content generation. Video, channel-specific aspect ratios, product highlight bullets. Net-new creative produced in Marpipe's creative library.

Privacy and data handling

Uploaded CSVs are parsed in your browser. When you audit a URL, the file is fetched once through our proxy, scored, and discarded — nothing about your catalog is stored on our servers. The audit results live only in your current browser session.