Skip to content
P2Lab AI

Tailor prompts to your shop

The default prompts shipped with P2Lab AI are intentionally generic — they have to work for anyone. Five minutes spent refining them against one of your real products before you run anything in bulk pays back on every subsequent generation: headings start the way you want, descriptions land at the length you want, the tone matches the copy you already have on the shop.

This page walks through that process end to end, using the generate_content.product template as the example. The same flow applies to every other template.

What you need before starting

Section titled “What you need before starting”
  • You’ve finished Your first prompt so you’ve seen what a default generation looks like
  • One product in your catalogue whose existing description represents the style you want AI to match — tone, headings, length, level of detail. This is your reference product

Step 1 — Open the prompt template

Section titled “Step 1 — Open the prompt template”
  1. Go to Catalogue → AI Assistant → Prompt templates.
  2. Open Generate product content (template type generate_content.product).

You land on the template detail page. Scroll to the System prompt card.

Step 2 — Open “Edit with AI”

Section titled “Step 2 — Open “Edit with AI””

In the top-right corner of the System prompt card click Edit with AI (wand icon).

The Adjust prompt with AI screen opens — a wide modal with chat on the left and a tabbed editor on the right. The right pane has three tabs: System prompt, User prompt and Test on product.

Step 3 — Pick a Reference product

Section titled “Step 3 — Pick a Reference product”

In the modal header use Reference product to pick the product whose existing copy reflects your shop’s standard. The chat will render the prompt against this product’s real data, so the model sees what the prompt actually produces, not just placeholders.

Step 4 — Run a baseline test

Section titled “Step 4 — Run a baseline test”

In the right pane switch to the Test on product tab and click Run test.

The plugin executes the current default prompt against your reference product and shows two columns per field:

  • ORIGINAL — what the product has today
  • AI-IMPROVED — what the default prompt produced

This is your before/after baseline. Read it. The gap between what the default produced and how you’d write it yourself is exactly what you’re about to close in chat.

Step 5 — Discuss the results in chat

Section titled “Step 5 — Discuss the results in chat”

On the left, type what you want changed in plain language. Be concrete:

  • “Always start the description with an <h2> containing the product name”
  • “Keep the description under 2500 characters”
  • “For supplements always include a section about macronutrients with a bulleted list”
  • “Don’t generate emojis”
  • “Use a formal, technical tone — no marketing fluff”

Press Enter to send. The assistant responds with a proposed updated System prompt (and possibly User prompt), a short summary of what changed, and optional warnings.

The right pane updates automatically with the new draft — you can edit it directly if you want to nudge wording.

Step 6 — Apply Suggested adjustments

Section titled “Step 6 — Apply Suggested adjustments”

When the model thinks a runtime parameter would help (e.g. Description: Max length 1500 → 4000), it surfaces a Suggested adjustments panel under its reply. Click Apply per row to take a single suggestion, or Apply all in the panel header to take everything.

These adjust the Test on product parameters used by Run test, not the prompt text — handy for quickly verifying “would a longer/shorter limit fix this?” before committing to changes in the prompt itself.

Step 7 — Iterate

Section titled “Step 7 — Iterate”

Switch back to Test on product and click Run test again. The new result is added to Previous runs below; the badge shows whether it’s still Current (same prompt + params as right now) or Archival (the prompt has drifted since).

Each result row has a Share with AI toggle. Leave it on for the run you want the next chat message to react to — the model will see that output as context. Turn it off for older runs you don’t want polluting the conversation.

Repeat Steps 5–7 until the AI-IMPROVED column reads how you’d write it yourself.

Step 8 — Apply + Save

Section titled “Step 8 — Apply + Save”

When you’re happy:

  1. Click Apply in the bottom-right of the modal — this copies the draft into the template detail page (the modal closes).
  2. Save the template using Shopware’s standard save button at the top of the page.

Until you save, nothing is persisted. The chat session itself is saved as you go (you can reopen it from the Chats sidebar), but the prompt change requires Apply + Save.

Repeat for the templates you actually use

Section titled “Repeat for the templates you actually use”

Walk the same loop for every template you plan to run in production. The most common ones:

  • generate_content.product — product descriptions
  • generate_seo.product — meta title / meta description / keywords
  • translate.product — translations

You don’t need to touch templates you won’t use — they keep working on defaults.

Image-edit presets (edit_image.*) are tailored differently — they are short single-prompt style presets you create or duplicate rather than refine in chat. See Add a template and Edit image.

Where to go next

Section titled “Where to go next”