google_product_category

Google Product Category — GPC — punches well above its documentation weight. It's the field that determines policy treatment (restricted categories), tax handling (per-region tax rules), and ad delivery (Advantage+ and PMax both use it as the primary segmentation signal when product set rules don't specify otherwise). Get it wrong and the consequences are spread across systems in ways that are hard to trace back to category mapping.

The taxonomy itself is published by Google and updates roughly twice a year. About 6,000 nodes total, hierarchical from broad ("Apparel & Accessories") to specific ("Apparel & Accessories > Clothing > Dresses > Maxi Dresses"). The right move is always to map to the deepest applicable node — broader nodes work but cost you segmentation precision. A maxi dress mapped to "Apparel & Accessories > Clothing" competes against every clothing item; mapped to the maxi-specific node, it competes against maxi dresses. The CPC difference between broad and deep mapping for the same product can be 30-50% in competitive categories.

GPC has three forms you can use in the feed: the full path ("Apparel & Accessories > Clothing > Dresses"), the numeric ID ("2271"), or the deprecated short path ("Dresses"). Always use the numeric ID. The IDs persist across taxonomy revisions; the paths don't. When Google rewords "Health & Beauty > Health Care" to "Health & Beauty > Personal Care", every feed using the old path breaks overnight. Feeds using IDs continue working — Google maps the ID to whatever the current path is. The short path is deprecated and unreliable; Google's docs still mention it but the auto-reviewer increasingly rejects items using it.

Policy enforcement keyed on GPC is the biggest hidden consequence of mis-categorisation. Restricted categories — alcohol, supplements, weapons, financial products, gambling, healthcare — face stricter rules and additional review. The classifier looks at the GPC, not the title or description, to decide which policy applies. A supplement listed under "Health & Beauty > Health Care > Vitamins & Supplements" gets the restricted-category review; the same supplement misclassified as "Beauty > Personal Care" doesn't trigger restricted treatment initially but gets re-flagged when Google's content classifier catches the mismatch. The downstream consequence is harder than just being in the right category from the start.

For multi-region catalogs, GPC can differ per region because tax categorisation differs. Children's clothing is zero-VAT in the UK but standard-VAT in the US; the GPC mapping should reflect the regional treatment. Most teams pick one canonical mapping globally and accept the tax-treatment imperfection, which is fine for most cases but worth knowing about for high-volume categories where the difference matters.

For variant groups, all variants under one item_group_id must share the same GPC. Mixed GPCs within a variant group get rejected as a categorisation inconsistency. This catches catalogs where seasonal cross-overs happen — a "swim trunks" item that's also marketed as casual shorts ends up with variants categorised differently because of the marketing dual-use. Pick one canonical category per item_group; the cross-promotion happens at the marketing layer, not the catalog layer.

The Optimiser approach to GPC mapping at scale: read each product's title, description, brand, and image, classify against Google's 6,000-node taxonomy, and pick the deepest applicable match. Confidence scoring on each match means low-confidence items get flagged for human review rather than guessed. Manual mapping at this taxonomy depth is impossible for catalogs over ~500 items; automation with human review is the only sustainable approach.