Product feed

What a product feed contains

At minimum, shopping feeds need a stable product ID, title, description, link, image link, price, availability, and brand. Exact requirements vary by channel. Variants usually need item group IDs so color and size sit under one parent. Identifiers such as GTIN, MPN, and your internal SKU help platforms match products and reduce "product not found" issues.

Category taxonomy (Google product category, Meta category) and attributes like color, size, material, gender, and age group improve eligibility and relevance. Optional fields still matter commercially: sale price and effective dates, shipping dimensions, condition, custom labels for campaign structure, and multipack or bundle flags. Each channel documents its schema; Google's product data specification is the usual reference for Shopping. Do not invent fields the channel ignores while leaving required ones empty.

A thin feed that validates beats a rich feed that fails diagnostics. Map every column to a source of truth in your PIM, Shopify fields, or spreadsheet, and name the owner of that source.

How feeds move: files, APIs, and platform sync

Feeds ship as scheduled file fetches (TSV, CSV, XML), API pushes, or native connectors (Shopify to Google, Meta pixel catalog sync, and similar). File feeds are transparent and easy to debug; they go stale if the schedule is slow or the job fails silently. API and app syncs feel automatic until a scope change, rate limit, or mapping error freezes attributes for days.

Whatever the pipe, you need a last-success timestamp and a visible error log. Multi-channel stores often maintain one canonical catalog and channel-specific rules: title length, banned words, image requirements, and category mapping. Do not hand-edit five divergent spreadsheets. Use rules in a feed tool or PIM to transform once from master data. Version or changelog major mapping changes so you can explain a disapproval spike after a bulk title rewrite.

The transport layer is boring infrastructure; the mapping layer is where most money is lost.

Disapprovals, diagnostics, and why spend stalls

Channels reject or limit products when required attributes are missing, identifiers conflict, landing pages do not match price or availability, images break policies, or claims look deceptive. Merchant Center diagnostics and Meta Commerce Manager surface item-level issues; ignoring them is how "the ads stopped working" becomes a data problem misdiagnosed as creative fatigue.

Common fixes: align sale price with the landing page, fix GTIN mismatches, replace placeholder images, and correct availability when inventory is zero. Work the error list by revenue impact, not alphabetically. Prioritize bestsellers and high-spend SKUs. Schedule a weekly disapproval scrub the same way you scrub stockouts. Policy updates happen; assign someone to read channel announcements for your category. Feed health is a leading indicator for Shopping and social catalog performance.

Creative tests cannot compensate for a catalog that platforms will not serve.

Freshness: price, stock, and landing URL truth

A feed that updates nightly may be fine for slow catalogs and deadly for flash sales or thin inventory. Price and availability should refresh on a cadence that matches how fast those facts change in checkout. Landing page URLs must resolve to the advertised product with consistent pricing. Message match for shopping ads is largely a feed-plus-PDP problem.

Out-of-stock items left "in stock" in the feed buy clicks that bounce and train the auction that your products disappoint. Define SLAs by field class: availability and price near real time or frequent; titles and taxonomy on editorial change; seasonal copy on campaign calendar. Monitor delta volume after each sync so a zero-row update or a mass wipe is obvious.

When using multiple warehouses, feed availability should reflect what the shopper can actually buy online, not theoretical warehouse totals that never ship to their region. Freshness is operational discipline, not a vendor checkbox.

AI content risk in titles and attributes

Teams use AI to bulk-write titles, descriptions, and attribute fills. Speed is real; so is fabrication. Models invent materials, certifications, dimensions, and "compatible with" claims that never appear on the physical product. That is the same hallucination problem as support bots, except the audience is both shoppers and channel policy systems. Incorrect attributes drive returns, disapprovals, and legal exposure in regulated categories. Constrain generation to approved fields and source text.

Prefer extraction from tech sheets over free invention. Require human review for safety, claims, and category-sensitive attributes. Keep a diff of AI bulk edits so you can roll back. Do not let a model overwrite GTINs, prices, or availability. Feed copy should help discovery without lying. If you would not put the sentence on the physical hangtag, do not put it in the feed.

Operating the feed as catalog truth

Name owners: catalog/PIM for attributes, ops for availability, merchandising for titles and custom labels, performance for channel diagnostics. Put feed health on the same weekly meeting as merchandising pins and paid search structure. Custom labels (margin tier, bestseller, clearance, season) turn the feed into a campaign control plane. Keep them updated or media structure rots. Treat first-party catalog accuracy as an asset.

Thread discussions about owned data apply here: platforms and AI tools amplify whatever product truth you publish. Audit a sample of top SKUs monthly: ID stability, image quality, category, and price parity with the live PDP. When you expand to a new channel, start with a clean subset of SKUs rather than dumping every draft product. A smaller accurate feed outperforms a full inaccurate one.

The product feed is how the outside world reads your store; keep it as honest as your checkout.