Claude · Anthropic
Strong on longer, nuanced texts
Own API key, own model choice, own tone default, completely independent.
Models: the latest Claude models
ChatGPT · OpenAI
Fast, cost-effective generation
Proven for large volumes. Own key/model/tone configuration here too. Connection test directly in the admin.
Models: the latest GPT models
Mistral AI
Strong European models
Good value, own key/model/tone default, connection test in the admin.
Models: the latest Large, Medium and Small models
Gemini · Google
Very fast flash models
Large context windows, own key/model/tone default, connection test in the admin.
Models: the latest Gemini models
Provider switch
All configurable in parallel
When several providers are active, pick the right one per call in the generator modal, Claude for the demanding main description, a flash model for fast meta data, for example.
Per language
Individual texts per channel & language
Each channel-language combination gets unique texts. Or generate once and write the same text to all selected combinations, you decide per run.
Tone
Five style templates
Professional, casual-modern, premium/luxury, technical-factual or emotional-promotional. Set a default per provider, override per generation.
Context chips
You control what the AI sees
Which product data flows into the prompt? Name, manufacturer, properties, category, price, selectable via chip. Plus own keywords for SEO focus.
Bulk-edit
AI in bulk too
In bulk edit, ChatGPT and Claude generate texts for many products at once and DeepL translates, each with its own switch in the config matrix. The number of parallel requests is adjustable for ChatGPT and Claude. Mistral and Gemini are not available there.
Safety
Review first, then apply
In the detail view, generated texts appear in a preview first and are only saved with "Apply texts". In bulk edit the AI writes directly into the selected fields; a snapshot saves the previous state beforehand.
Central instruction
Master prompt for every provider
Set salutation, brand spelling, mandatory information or forbidden phrasings once, the instruction travels with every request, in the detail view as well as in bulk edit and regardless of the provider.
Lengths
Targets per field in words and characters
Product name 120 characters, SEO title 60, meta description 155, description 250 words, the lengths search engines display in full. The values also drive the token budget, so the response no longer breaks off mid-JSON.
Starting text
Copy from the default language
A channel with its own language but no own texts? Name, description, meta fields and keywords can be copied over from the default language, only then does the AI have something to rewrite in that language. By default only empty fields are filled.
New field
The product name as well
Besides description, meta title, meta description and keywords, the AI now also writes the product name, for ChatGPT as well as Claude. Not preselected, so existing names are not overwritten by accident.
Large runs
Saved in batches instead of product by product
Generated texts are collected and saved in blocks: for 2000 products the number of save operations drops from over 2000 to around 40. On request limits the run waits and retries automatically.