What is catalog enrichment?

A product page with a name, a price and a photo can be bought from, but it does not answer what the shopper is asking: what it is made of, what size fits, who it is for, what code identifies it. Catalog enrichment is the work that closes that gap, product by product. It is not the customer data enrichment of a CRM, which adds information about people, and it does not mean making descriptions longer: it is completing and correcting the data of each product and knowing where each value comes from. It matters to you because whatever you do not write, every reader has to guess, whether a person, Google or an AI assistant, and each one guesses in its own way.

What it covers

Enriching a catalog touches four layers of data, and each one has its own readers.

• Attributes. Color, material, size, pattern, measurements: the product attributes that let people filter, compare and choose. The product data specification of Google Merchant Center organizes them by section and, for clothing and accessories, requires color, size, gender and age group in free listings.⁠^{[1]Google Merchant Center Help. Product data specification}
• Identifiers. The GTIN, the brand, the MPN and the SKU of each item, which say exactly which product it is and set it apart from similar ones. When a product is sold in several sizes or colors, each combination is one of its product variants and carries its own identifier.
• Text. The product description, the frequently asked questions, the title and the meta description, written from the data above and not in place of it.
• Markup. The product page's structured data, usually in JSON-LD, and the product feed sent to each platform. Google explains that providing both structured data on the page and a Merchant Center feed helps it understand your data and verify it, and that in some experiences it combines the two sources.⁠^{[2]Google Search Central. Intro to Product structured data on Google} That is why the page, the markup and the feed have to say the same thing.

Original data, inferred data and the human decision

An enriched catalog is not measured only by how many fields are filled in, but by whether you can say where each value came from. There are three possible origins, and it pays not to mix them.

Original data is what only the person who makes or sells the product can provide: the barcode, the composition on the label, the measured weight, the warranty. It cannot be deduced from anything. Google is blunt about this for the GTIN: submit it only if you are sure it is correct, do not guess it or make up a value, and do not submit one for a product that does not have it. If you are the only seller or the product is your own brand, it will normally have no GTIN, and then the brand and the MPN are enough.⁠^{[3]Google Merchant Center Help. GTIN [gtin]}

Inferred data is obtained from other data. It can be a simple deduction (the color, taken from the variant name "Navy / M"; the category, taken from the description) or a text a language model writes from the product page. Google reminds you that what an AI generates can contain inaccuracies, known as hallucinations, and asks you to check it by hand before publishing it, including in the elements that appear in results: the title, the meta description, the structured data and the images' alt text.⁠^{[4]Google Search Central. Google Search's guidance on using generative AI content on your website} Merchant Center goes a step further: titles and descriptions written with AI are submitted separately, in the structured_title and structured_description attributes, with the digital_source_type sub-attribute set to trained_algorithmic_media.⁠^{[5]Google Merchant Center Help. AI-generated content} The platform itself makes you separate the inferred from the original.

The human decision is not a third kind of data, but the step that decides what gets published. Approving an inferred value does not turn it into an original one: it is still inferred, but now it has an owner and a date. That record is what lets you go back when a value turns out to be false, and that is why enrichment goes hand in hand with content versioning.

In a store

Fictitious example. A mountain clothing store enriches the product page of a waterproof jacket, navy variant in size M. This is the worksheet for a single product, with the origin of each field in plain view:

Look at the third row: an empty field is better than a made-up value, and a frequently asked question that depends on it waits too. Each tool handles the origin column in its own way; Clione, for example, marks as AI-inferred the quality scores it calculates for each product and publishes nothing to the store until you press Push.

How to check it

1. Pick the ten products you sell the most and open them next to Merchant Center's product data specification. Note which required and recommended attributes they are missing.⁠^{[1]Google Merchant Center Help. Product data specification}
2. Mark the origin of each filled field: original, inferred or "I don't know." The "I don't know" ones are the finding, because nobody can defend a value whose origin they do not know.
3. Look for suspicious GTINs: the same code on two different products, or a code on an own-brand product. If the product has no GTIN, Google asks you not to submit one.⁠^{[3]Google Merchant Center Help. GTIN [gtin]}
4. Compare the product page, its markup and the feed for a single product: color, size and price and availability have to match in all three.⁠^{[2]Google Search Central. Intro to Product structured data on Google}
5. Review the AI-generated text: who approved it, when, and whether in Merchant Center it goes in structured_title and structured_description.⁠^{[5]Google Merchant Center Help. AI-generated content}

Related concept: PIM

A PIM (product information management) is the system where product data is stored and managed; enrichment is the work done on that data, inside a PIM, in the platform's admin or in a spreadsheet. Having a PIM does not enrich anything by itself: it organizes the fields, but someone has to fill them in and decide where each value comes from. Nor is it the same as a catalog audit, which is the diagnosis beforehand: it says what is missing, and enrichment completes it.

To keep going, the catalog and entities reading path puts the rest of the terms in order, from the attribute to the product graph.