Google product categories at scale: Our approach to AI category matching

Picking the right Google product category sounds like a small task. You have one product, and Google has a list of categories to choose from. How hard could it be?

Pretty hard, it turns out. Google's taxonomy has thousands of categories arranged in a deep hierarchy, and plenty of products could technically fit under two or three different branches. Pick the wrong one and your product might miss category-specific requirements, get grouped with the wrong competitors, or simply match less accurately with what shoppers are searching for. Now multiply that decision by every product in your catalog, and you can see why category selection quietly becomes one of the most overlooked parts of feed management.

We've been working on a way to fix this for a while, and it's now live inside Productrise. Using language models to understand what a product actually is, rather than just matching keywords, we built a category mapping tool that maps any product to its best-fitting Google category in seconds. It's also become part of how we audit product feeds. Here's how it works.

What is a Google product category

The google_product_category attribute is how you tell Google what kind of product you're selling. Google actually assigns a category to your products automatically based on your titles, descriptions, and other feed data, but you can override that assignment with your own value when needed, for example to remove category-specific requirements or to organize Shopping campaigns.

To choose that value, you have to pick from Google's own list of predefined categories, available as a full taxonomy file. This file lists thousands of categories, each with a numeric ID and a full path, like "Animals & Pet Supplies > Pet Supplies > Bird Supplies > Bird Toys." Google recommends using the most specific category that still accurately describes the product, and using only one category per product.

That last part is where things get tricky. The taxonomy is built as a hierarchy, so a single product often fits under several valid branches at once, and only one of them is truly the best match.

Why the right category isn't always obvious

Take a bird toy as an example. You could categorize it under Pet Supplies, under Bird Supplies, or under the more specific Bird Toys category. All three are technically correct, since a bird toy is indeed a pet supply and a bird supply. But only one of them, Bird Toys, actually describes the product with any precision.

Now imagine making that same judgment call across a catalog with hundreds or thousands of products, each with its own set of overlapping category options. Doing this by hand isn't just slow, it's also easy to get wrong, especially for anyone who isn't deeply familiar with the shape of Google's taxonomy.

Most existing category finder tools don't help much either. Many of them just match the exact words in your product title against category names. That approach breaks down the moment your title uses different phrasing than Google's taxonomy does. Search for "dog food" and you might get a decent match. Search for "food for dogs" on some tools, and the match disappears entirely, even though it's the same product.

Meet the Google product category mapper

This is the problem our Google product category mapper is built to solve, and it's free to use for anyone, not just Productrise customers.

Instead of matching words, it matches meaning. Every one of Google's roughly 5,595 product categories gets converted into an embedding, a numerical representation that captures what a category actually means rather than the specific words used to describe it. Your product title goes through the same process, and the tool then finds which categories sit closest to it in that meaning-based space using cosine similarity. To make the result easier to explore, the tool also visualizes this space as an interactive two-dimensional map, so you can see where your product lands relative to thousands of Google's categories at once.

Because it's working with meaning rather than exact phrasing, it can handle the "food for dogs" versus "dog food" problem that trips up keyword-based tools. It can also surface the more specific option, like Bird Toys over the broader Pet Supplies, when that's genuinely the better match.

Category matching is now part of your feed audit

This same intelligence is now built directly into Productrise's feed audits. Every audit checks whether each product has a category defined at all, since a missing google_product_category is a critical error that can keep a product out of search results entirely. But it goes a step further than that. For every product that does have a category, Productrise now compares it against what our category mapper considers the best match, and flags the products where a more accurate category is likely available.

In one feed we audited, out of roughly 200 products, 84 had no category defined at all, an easy fix once you know about it. Of the remaining products, 81 had a category that we flagged as a possible mismatch. One example: a handmade wool cat bed that had been filed under Pet Doors. It's a pet product, so the general area wasn't wrong, but a cat bed clearly isn't a door. Running that same title through the category mapper immediately surfaces the correct Cat Beds category instead.

You can export these findings to a spreadsheet and work through them at your own pace, rather than manually checking every product against a taxonomy file with thousands of entries.

Here's what that looks like inside a real audit.

What's next

Category matching is the newest addition to Productrise's list of feed audit checks, and it won't be the last. The next step is bringing this same intelligence into supplemental feeds, so that alongside AI-generated titles, descriptions, and missing attributes, you'll also get suggested categories for products that need them, without having to look anything up yourself.

Try it yourself

You don't need an account to try the category mapper. Head over to the tool, type in a product, and see where it lands. If you want to see how your entire feed is doing, including category matches, missing images, duplicate titles, and everything else that affects your organic visibility, you can get started with Productrise for free, no credit card required.