Test on product
Testing on a product runs the current draft of a prompt against real product data and shows you the response. Nothing is persisted — the result is a one-shot preview you can throw away. Use it to validate changes before saving the template.
How to run a test
Section titled “How to run a test”- On the prompt detail page, scroll to the Test panel (or open the Test tab in the AI Assistant chat).
- Click Pick product and select a product. The picker is the standard Shopware product search.
- Click Run test.
- The plugin builds the request the same way it would in production — applying the current draft, resolving placeholders, attaching the connection model — and sends it to OpenAI.
- The response renders in a panel tailored to the prompt type.
Per-type result viewers
Section titled “Per-type result viewers”The viewer depends on what the prompt produces:
- SEO — three-field layout: meta title, meta description, keywords.
- Measures — grid of weight, dimensions, unit, pack unit labels, with source quotes.
- Properties — two columns: matched property groups (with confidence per row) and proposed new groups, both with source highlights.
- Alt text — image thumbnail next to the proposed alt text.
- Content / Improve / Unify — diff viewer comparing source vs proposed for each field.
- Translate — per-language tabs, each showing the translated value.
- Image edit — before / after image pair.
Every viewer also surfaces the model’s reasoning and per-field confidence so you can spot uncertain proposals.
Test runs are real calls and incur the same per-token cost as production runs. The result panel includes the estimated cost so you know what each iteration costs.