Skip to content

AI for catalog content

| Article: 4 min read | AI in e-commerce

AI for catalog content means using language and image models to prepare product descriptions, attributes, translations and images for an online store. A model drafts quickly, but it does not know the facts about your product: you have to give them to it and check the result.

In plain words

In a catalog of a thousand products, manual work on descriptions, translations into English or Polish, and photos takes weeks. AI can help with drafts. But a mistake in an attribute (wrong material, wrong size, an invented certification) becomes your store's mistake: the buyer gets something other than what they expected, and the return or complaint is yours.

Google says this directly: generative models predict likely sequences of words rather than retrieve facts, so their text may contain inaccuracies, and "it is critical to manually fact-check and review all AI content" before publishing (Google Search Central). This applies to the title, meta description, structured data and alt text too.

What you can hand to AI

Task

How it works

What to check

Description from supplied attributes

The model turns a table of attributes into readable text

Whether it added extras: materials, certificates, validity periods

Structuring attributes

Bring "хб", "хлопок", "cotton" to one value

Spot checks against the source (supplier card)

Translation

A draft in English, Russian or Polish

Terms, sizes, units, legal texts

Meta tags, alt texts

Short drafts

Match with the page, no invented facts

Photos

Remove the background, prepare image variants

Whether the image shows exactly the product you sell

Good practice: give the model structured data (PIM or a table) and ask it to write only from that, and to mark anything missing from the data as "unknown" rather than invent it.

Platform rules

  • Google Search. AI content is not prohibited. The risk is mass-producing pages with no value: Google treats this as a violation of its spam policy on scaled content abuse (Google Search Central).
  • Google Merchant Center. For images created by AI, you must keep the IPTC DigitalSourceType metadata with the value `trainedAlgorithmicMedia`; the requirement has applied since February 2024 and covers the main, additional and lifestyle images (Google Merchant Center, specification changes). For AI-generated text, Google added separate `structured_title` and `structured_description` attributes in April 2024 (same source). Check the current specification before setting up the feed (product feed).

Risks

Invented facts. This is not only about certificates. It is size, compatibility ("fits iPhone 17"), cosmetics composition, allergens. For household chemicals, food and children's products, a mistake can have consequences beyond money.

Influence on the buyer's choice. A UC San Diego study (presented in December 2025) on AI summaries of reviews: people who read an AI summary were 32% more likely to buy than those who read the original reviews, and the sentiment of the reviews changed in 26.5% of cases. The sample is small (70 people), and 18 tested ways of reducing the bias did not work across all models (UC San Diego, 2026). Takeaway: AI-written review summaries and comparisons can shift perception, so do not publish them without review.

Identical texts. When a thousand descriptions are written from one template, the pages are no different from competitors'. Add what only you know: usage nuances, sizes on the model, answers to real buyer questions.

There is no fresh data on how often AI makes mistakes in product descriptions. There are no measurements on Ukrainian stores either, so run your own sample: check how many of 50 descriptions contain an error.

How to organize quality control

  1. Give the model only verified input: supplier card, attribute table.
  2. Forbid inventing: "if there is no data, write 'not specified'".
  3. Check the first 50 results in full. Count errors by type: size, material, compatibility, translation.
  4. After that, spot checks (for example, every tenth) and full checks for sensitive categories: food, cosmetics, children's products, electrical goods.
  5. Record who checked and when, and mark AI images in the metadata.
  6. Watch the consequences: the share of returns "not as described" by category (returns).

Example (illustrative numbers)

The numbers are made up for the example. A store checked 50 sneaker descriptions created from an attribute table. In 6 descriptions, AI added "waterproof membrane", which is not in the table. Four descriptions had a wrong translation of the size chart. After changing the prompt ("write only from the table data"), errors of the first type disappeared, while the translation of sizes still had to be checked by hand.

Common mistakes

  • Generating descriptions with no input data ("write a description of Nike sneakers").
  • Publishing photos and translations without review.
  • Mass-producing pages for the sake of traffic.
  • Not marking AI images when Google Merchant Center requires it.

FAQ

Can I write product descriptions with AI? Yes, as a draft. Give the model structured attributes and check the result. Google does not prohibit AI content but requires value for the user.

Will Google penalize descriptions written by AI? Not for the mere use of AI. The problem is mass pages with no value.

Do I need to mark AI images in Google Merchant Center? Yes: images created by AI must keep the IPTC tag `trainedAlgorithmicMedia` (a requirement since 2024).

Is AI translation of a catalog reliable? Fine for drafts, but have a person check sizes, units, legal and safety texts. There are no separate fresh measurements of translation quality for product descriptions.

What should I check first in an AI-written description? Materials, sizes, compatibility, composition, certificates, shelf life: everything that can be checked against the source.

Sources

Want to check this on your store?

AI automation

An AI assistant checks the store every day, suggests fixes and makes them after your «Yes».

About the service →

Share

Related

  • Agentic commerce

    | Article: 5 min read

    Agentic commerce is shopping in which an AI agent acts on a person's behalf: it searches for and compares products, and within the permission it has been given it can also place the order and pay. For a store, this means part of the "visitors" will be…

  • Technical debt

    | Article: 6 min read

    Technical debt is the future cost of fixing and reworking that arises when software was built fast and without room for change. Like financial debt, it lets you move faster today, but with interest: every next change costs more and takes longer.

  • Vibe coding

    | Article: 4 min read

    Vibe coding is building programs where a person describes the task in ordinary language, AI writes the code, and the person accepts the result almost without checking it and goes by whether it "looks like it works", adding new wishes in the chat. For quickly…

  • How an online store team uses AI every day

    | Article: 2 min read

    AI in a store team's daily work is a set of small, repeating tasks in which a language model prepares a draft and a person checks it and decides. The value comes not from separate "smart" features but from simple habits assigned to specific people.