AI in e-commerce: product texts, customer service, mandatory information
Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.
If you sell online, you need a text for every product, often in several languages and for several marketplaces. On top of that come e-mails about delivery times and returns. An AI takes a lot of writing off your hands. But what is true about the goods and what is mandatory comes from you and from the manufacturer.

Key points
- An AI writes many product texts from a spreadsheet if you forbid it to add details.
- Translations and marketplace texts are done quickly. Still check dimensions, units and technical terms.
- Product photos show the real goods. AI images that look real can mislead customers.
- Manufacturer details, warnings and legal texts come from your documents, not from the chat.
- Orders, addresses and customer names don’t belong in a private AI chat.
Which tasks in an online shop suit AI
An AI is good at turning bullet points into text, translating texts and writing templates. But it has never seen your goods. Everything it writes about dimensions, material or what’s in the box has to come from your details.
| Task | You provide | You check |
|---|---|---|
| Product texts in bulk | A spreadsheet with all details | Every detail against the spreadsheet |
| Texts for marketplaces | The finished text, the platform’s rules | Length, banned words, required fields |
| Translations | Text, target language, technical terms | Dimensions, units, tone |
| Reply templates for customer service | Occasion, your rules, your legal texts | Deadlines, promises, form of address |
| Checklist of mandatory information | What your shop shows today | Every detail with the manufacturer |
In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson
Many product texts from one spreadsheet
Elias sells coffee equipment in a small online shop: grinders, filters, pots. New goods arrive with a list from the supplier. He adds his own notes to the list and has the AI write texts from it, always in batches of ten products:
For each row of the table below, write a product description for my online shop for coffee equipment.
Structure per product: one sentence on what it’s good for; then three bullet points with the most important details; then one sentence on care or cleaning, only if it is in the table.
Rules:
- Use only details from the table. Don’t add any material, dimensions or features.
- No claims about the environment or health, no superlatives.
- If a cell is empty, write [missing: column name].
Return the result as a table: item number, text.
[paste the table with item number, name, material, dimensions, contents, note here]The item number in every row helps with checking: Elias puts the result next to the supplier list and ticks off every detail. With long lists, an AI sometimes swaps values between rows. That’s why ten products at a time are better than a hundred.
Marketplaces and other languages
Ines sells handmade jewelry in her own shop and on a marketplace. Every platform has its own rules: how long a title may be, which words are banned, which fields are required. She copies the rules from the platform’s help pages into the instruction and has the text adapted to them. What applies is in the help pages, not in the chat.
For customers abroad, she has texts translated:
Translate this product description into French for customers in France and Belgium.
Keep all dimensions and material details exactly as they are. Don’t convert any units.
Jewelry terms: sterling silver = argent sterling, drop earrings = boucles d'oreilles pendantes.
Tone: friendly, formal.
Below the text, list every place where you are unsure about a technical term.Before listing, Ines compares every number with the original text. For her most important products, she asks someone who speaks the language well to read the texts once.
Keeping product photos honest
With AI you can swap backgrounds, improve the lighting or generate whole scenes. But a product photo should show what arrives. If the AI changes the goods themselves, such as their color, shape or size, that misleads people and leads to returns.
- A new, calm background behind the real photo: generally unproblematic.
- Goods that look nicer, bigger or in a different color than in reality: misleads customers.
- A whole scene from the AI that looks real: the EU AI Act may require a notice. A line such as “AI-generated image” is honest in any case.
Customer service: templates for delivery time, returns, defects
Many e-mails are about the same things. With templates you reply faster and equally kindly every time. Malik sells spare parts for garden equipment. He has templates written in which the legal sentences come from his own legal texts:
Write three reply templates for my online shop for spare parts, formal, at most six sentences per template:
1. Delivery is delayed. Rule: insert the new date as [date], offer free cancellation.
2. Customer wants to return something. Don’t answer the legal situation yourself. Insert [text from our withdrawal policy] and describe our process: return label by e-mail, refund as soon as the goods are back with us or the customer proves they were sent.
3. Part doesn’t fit or is defective. Ask for a photo and the device number. No promise of replacement or refund.
Placeholders instead of names and order numbers.If you have the AI adapt a reply, paste the real customer e-mail without name, address and order number. You insert the real details afterwards in your shop system.
If an AI on your website answers customers directly, it generally has to be clear that an AI is writing there, at the latest with the first message. Check this with ready-made chat services.
Replying to reviews
An AI helps you reply to bad reviews calmly and objectively: thank the person, address the point, offer a way forward. Give it the review without names and add what really happened.
Write a public reply to this review of my shop. At most four sentences, formal, calm, no excuses.
What happened: the parcel arrived four days late because a part was missing at the supplier. We didn’t let the customer know.
Offer that the person contacts us by e-mail. No vouchers, no promises.
Review: [paste text without names]Read the reply before it goes online: is every sentence true, and would you say it that way? Writing reviews yourself, having them written or buying them is prohibited in the EU, even if an AI phrases them.
Mandatory information comes from your documents
The EU General Product Safety Regulation (GPSR) has applied since December 13, 2024. Anyone offering consumer products online in the EU must show, among other things, these details clearly and visibly in the offer:
- Name and postal address of the manufacturer and an electronic address, such as an e-mail.
- If the manufacturer isn’t based in the EU: name, address and electronic address of the responsible person in the EU.
- Details that make the product clearly identifiable, such as a picture, its type and model.
- Warnings and safety information, where there are any, in understandable language.
These details also apply to offers on marketplaces. An AI can build you a checklist to go through your listings with. The content itself comes from the manufacturer or supplier. Don’t let the AI add an address, a warning or a model designation.
Since June 28, 2025, many online shops for consumers also have to be accessible, under the German Accessibility Strengthening Act (Barrierefreiheitsstärkungsgesetz). Microenterprises are exempt: fewer than ten employees and at most two million euros in annual turnover or annual balance sheet total. An AI can help with the image descriptions for your shop, the alt texts that screen readers read aloud. Check each one to see whether it describes what is in the picture.
Customer data and shop extensions
Orders, addresses, phone numbers and customer accounts don’t belong in a private AI chat. Some shop systems offer their own AI functions or extensions that access orders. Before you switch them on, check which data they see and whether there is a data processing agreement (DPA). More on this in the guide Data protection and AI tools.
How to tell an AI exactly what to make from your details, and how to check the result, is what you practice in the learning app “KI können”, in short lessons with examples from retail. You do the actual work in your own AI. The trial lesson is open without an account.
Sources
- Regulation (EU) 2023/988 on general product safety (EUR-Lex) (opens in a new tab)
- German Accessibility Strengthening Act (Barrierefreiheitsstärkungsgesetz), Section 3 Accessibility requirements (in German) (opens in a new tab)
- German Accessibility Strengthening Act (Barrierefreiheitsstärkungsgesetz), Section 2 Definitions (in German) (opens in a new tab)
- German Act against Unfair Competition (UWG), Annex (prohibited commercial practices; in German) (opens in a new tab)
- Regulation (EU) 2024/1689 on artificial intelligence (EUR-Lex) (opens in a new tab)
- Federal Network Agency (Bundesnetzagentur), Transparency obligations (in German) (opens in a new tab)
Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.
Read more
Try it now
The trial lesson with examples for Online retailer. No account, a few minutes.
Start the trial lesson