Product descriptions and translationswith AI, done properly.

Why product descriptions drive revenue

In a shop, customers can pick up a product. Online, the product page has to replace that experience, and the studies are clear about it. In Syndigo’s «State of Product Experience 2026», with 8,736 respondents, 83 % say they are likely to leave a product page if they cannot find the information they are looking for. 26 % returned an item in the last six months because the content did not match their expectations. Detailed descriptions (52 %) matter more to the purchase decision than discounts (49 %).^[1]

A study by Salsify reaches similar results: 54 % of respondents abandoned a purchase because of contradictory information, and 71 % returned items because of inaccurate descriptions.^[2] Both studies mainly surveyed people in the US and Europe. The figures cannot be transferred one to one, but the direction is clear.

For Switzerland, the 2025 online retailer survey by FHNW provides the practical context. Two thirds of the 643 shops surveyed already create product descriptions with AI, and almost all use it for texts and translations. At the same time, nine out of ten retailers struggle with the introduction, and 75 % have concerns about data security.^[3] The market is large enough to take the topic seriously: Swiss online consumption reached CHF 15.8 billion in 2025, 6 % more than in the previous year.^[4]

Multilingual content: not an extra in Switzerland

According to the Federal Statistical Office, 62 % of the population name German as a main language, 23 % French and 8 % Italian. Multiple answers were possible.^[5] A shop that only sells in German does not address a large part of its home market in its own language.

The CSA Research study «Can’t Read, Won’t Buy», with 8,709 respondents in 29 countries, shows why this matters: 76 % prefer to buy when product information is in their language, and 40 % never buy from websites in another language.^[6] The data dates from 2020 and does not mention Switzerland explicitly. It is still a useful guide.

In a shop, translation means more than the description. It also covers titles, variants, attributes and filter values, categories, SEO titles, meta descriptions and alt texts. If one of these is missing, the page looks half translated, and filters return different results in each language.

What Shopify and WooCommerce offer

Both systems support multiple languages, but they take different routes. The following overview summarises the key points from the vendors’ documentation.

Vendor documentation as of September 2026. Features and limits change regularly.

For automated workflows, the API is the decisive point. In Shopify, translations can be written specifically per product, metafield and metaobject. A system can therefore detect new or changed products and translate only those, instead of reprocessing the whole range each time.

What Google says about AI-generated text

Google does not prohibit AI content. Its guidance on generative AI asks for a focus on accuracy, quality and relevance, including titles, meta descriptions and alt texts.^[15] Under «scaled content abuse», however, its spam policies explicitly name the use of generative AI to create many pages without adding value for users. The same applies to automated rewording or translation with little benefit.^[14]

Shopping ads come with an additional rule: in Merchant Center, AI-generated titles and descriptions must be submitted in separate attributes and labelled as AI-generated.^[16] If you generate product data automatically, build your feed accordingly from the start.

Copying manufacturer descriptions word for word brings little benefit. When content is identical, Google selects one canonical page and shows the others less often.^[20] In addition, longer product descriptions can be protected by copyright; a German higher regional court ruled this in 2014 (OLG Düsseldorf, I-20 U 174/12).^[19] Your own, product-specific texts are therefore the safer choice.

Special case food: mandatory information belongs on the product page

Anyone selling food or food supplements online has to provide more than a good description. Under Art. 44 of the Swiss Foodstuffs and Utility Articles Ordinance (LGV), all information required by food law must be available when goods are offered by distance selling, except for the date of minimum durability and the lot number. All information must be available on delivery.^{[17, 21]}

According to the Federal Food Safety and Veterinary Office (FSVO), mandatory information includes the name of the food, ingredients, date marking, storage, address, country of production and nutrition declaration, in at least one official language.^[18] For the EU market, Art. 14 of the Food Information Regulation (FIC) follows the same principle.^[22]

What a reliable workflow looks like

The requirements above lead to a workflow that works well in practice. It is not a standard, but a sensible sequence for shops with many products.

1. Collect source data: manufacturer information, data sheets, images and existing attributes in one place, for example in metafields or a PIM.
2. Define rules: tone, structure, length, forbidden claims and a glossary per language.
3. Generate text from facts only: the AI writes, but does not invent attributes. If information is missing, the field stays empty and is flagged.
4. Check automatically: numbers, dimensions, mandatory fields and filter values are compared with the source data.
5. Translate with terminology: technical terms and brand names stay consistent; filter values are maintained per language.
6. Review sensitive content: food, health claims and legal texts are checked by a person; spot checks by native speakers secure the other languages.
7. Process changes only: new or changed products run through automatically, everything else stays untouched.

With this set-up, AI becomes a controlled production step rather than a text generator. Quality then depends less on the individual prompt and more on clean data and clear checking rules.

Frequently asked questions