Product taxonomy: help shoppers and shopping platforms understand what you sell

A product taxonomy answers “what kind of item is this?” and helps decide which attributes are relevant to it. Used well, it helps your team keep the right attributes on each product and map items into the categories each channel supports. The practical task is to connect a product's actual identity with each system's structure without stripping away the language that makes your store distinctive.

Store navigation, merchandising collections and standard classifications do related but different jobs. A collection can group items around a season or a campaign. A standard category describes the type of product. A product can belong in “Gifts under £50” without “gift” being an adequate technical classification.

So avoid forcing one field to do all three jobs. Keep your commercial presentation, your product type and your destination mappings, each understandable in its own terms.

The cost of mixing them is visible wherever feeds are read at scale. SO-ME, the consumer shopping platform Sighted powers, receives feeds from ASOS, REVOLVE, Vinted, M&S, New Look, Debenhams and EGO, and the same kind of product arrives under a different category from one retailer to the next, with a different set of attributes beside it. That is why Sighted's own categorisation, described in the closing section, checks the category a product arrives with against its own title, type and tags.

Google assigns every Merchant Center product a category from its own taxonomy automatically. The google_product_category attribute is optional, and Google's documentation (as of October 2026) uses a submitted value to override its own choice only in set cases:

• a product wrongly placed in a category with extra requirements, such as apparel, mobile phones or software;
• Google Ads campaigns structured by category;
• alcoholic drinks.

The attribute takes one value, the ID or the full path, and asks for the most specific category that matches the product's main function. Look up the current value before sending a replacement.

On Shopify, each product can have one category from its Standard Product Taxonomy, alongside a product type you define yourself. The category brings its own category metafields (a shirt can carry size, neckline and sleeve length type), and with Shopify Tax it sets the tax rate.

The two lists name and nest things differently, so map them deliberately, and never paste a collection name into a standard category field and assume it will be accepted. The Google product category and Shopify taxonomy glossary entries give the short definitions.

Imagine a retailer selling a long-sleeved linen shirt. Its storefront includes the item in “Travel essentials” and “Summer edit”, and the approved specification confirms that it is a shirt, with its composition, colour and sizes. This is an illustrative mapping exercise, not a live catalogue example or a client result.

Keep the two collections if they help customers browse. For the standard classification, choose the supported category that describes the shirt: in Shopify's taxonomy that is Apparel & Accessories > Clothing > Clothing Tops > Shirts, while Google's UK list files shirts under Clothing & Accessories > Clothing > Shirts & Tops. Then review the fields the category raises, such as fabric, size and colour, against the source evidence.

Do not reclassify the shirt as travel equipment to match the campaign. Equally, do not add “wrinkle-free”, “UV protective” or other travel performance attributes because the item appears in a travel collection. The campaign gives a shopping context; the classification and the verified attributes describe the item, and both can stand without contradicting each other.

1. Export the classification fields separately

Include the internal product reference, title, current store type, collections, standard category and destination category mappings, with their original field names. Merging everything into one “category” column hides the source of the confusion.

Sample across your main product types, not only your best-maintained range, and include accessories, bundles and the items staff regularly disagree about: those show where your rules need more precision.

2. Agree what the product actually is

Use the approved record and the sellable offer. A replacement coffee-machine filter is not the machine itself (Google's taxonomy files filters under coffee maker accessories and machines under kitchen appliances), and a set of several products may need a different decision from one component sold on its own. Where the decision is ambiguous, write a short rationale that names the item and the evidence, rather than “closest match suggested by AI”.

3. Select the supported category

Use the most appropriate supported value for that destination rather than an invented label that sounds more specific. Record the taxonomy version or the date you checked, and where the value came from: Google's taxonomy file carries a version date in its first line, and Shopify names its taxonomy releases by date.

Never publish a numeric category code from memory. If your integration transforms categories automatically, inspect what it sends as well as what you selected.

4. Review the relevant attributes

A category decision should lead to useful data questions. For clothing that may be fabric or size; for an accessory, compatibility. The category can organise the questions, but it cannot supply the missing answers, so leave unknown values unresolved until they are verified. The product data optimisation guide shows how to record sources and approvals for the changes you propose.

5. Verify the published mapping

Check the store and each intended destination. Confirm the item kept the right category and that no rule substituted a parent value or a legacy label during export.

Test representative products before a broad mapping change, keep the previous values, and check whether filters, collections or operational processes depend on the field you are changing. On Shopify, a category change keeps only the category metafields that have values or are linked to variant options.

AI can propose categories and group uncertain records for review. Give the model the destination's permitted values, the product evidence and an instruction to flag ambiguity instead of inventing a category, then review its explanation as well as its suggestion. A model that files a filter as a coffee machine because the title names the machine's brand has made a predictable error with real consequences.

Start with a labelled sample your team has checked. Compare the model's suggestions with those decisions, inspect the disagreements and refine the rules. A high overall match rate is no reason to ignore errors in an important product group.

Consistent classification and relevant attributes make product records easier for any system to interpret. They are part of the information foundation rather than a guaranteed ranking mechanism for every assistant.

Classification can also be a gate. Shopify Catalog, which carries a Shopify store's products to ChatGPT, leaves out a product Shopify itself classifies as Uncategorized, and Shopify derives that classification from the product's own information; the ChatGPT troubleshooting guide covers the check.

A shopper's detailed question still needs evidence that the product suits it. A correctly classified shirt does not establish its fit, its care requirements or whether it is in stock. The guide to how AI shopping systems choose products covers the wider picture.

Is product taxonomy the same as website navigation?

No. Navigation helps shoppers browse your store; a standard taxonomy classifies product types for a particular system. Keep the relationship clear, but do not assume the labels must be identical.

Should you replace your creative category names?

Not automatically. Keep useful customer-facing merchandising language, and keep accurate standard classifications in the fields meant for them.

Will fixing your taxonomy guarantee AI recommendations?

No. Verify the classification improvement first. Suitability, the evidence available, the offer and the shopper's request still decide whether a product is relevant.

1. Export the classification fields separately, with their original field names, and sample across your main product types, accessories and bundles included.
2. Agree what each product actually is from the approved record and the sellable offer, and write a short rationale where the decision is ambiguous.
3. Select the most appropriate supported category for each destination, and record the taxonomy version or the date you checked.
4. Review the attributes the category raises, and leave unknown values unresolved until they are verified.
5. Verify the published mapping in the store and at each destination, testing representative products before a broad change.
6. If AI proposes categories, give it the destination's permitted values and compare its suggestions with a labelled sample your team has checked.

Sighted places each product in a category before it enriches it: a category family, which decides the attributes that matter, and a Google product category, which goes into the feeds Sighted exports when you have not set one yourself. It starts from the Shopify category already set and checks it against the product's own title, type and tags: where the words name a different kind of product, the words decide instead.

A product it cannot settle is shown as a best guess or uncategorised for you to review, and you can change any category, one product or many at once. On Shopify, each attribute value is mapped to the taxonomy's closest standard value, and a fact Shopify has no field for goes into a Sighted field instead. The free AI visibility audit shows what up to 50 of your product pages tell a platform about what you sell, and the categorisation and enrichment behind it are the technology that powers SO-ME.