,
- ,
- ,
allowed), 200–400 words, includes benefits, key specs, dimensions/materials, one CTA, use natural language, no promotional superlatives like “best” unless supported by facts. - features_list: JSON array of 5–10 concise feature objects: { "name": "...", "value": "..." }. - tags: up to 10 comma-separated tags. - image_alt_texts: array aligned with images, 6–12 words, include primary keyword when appropriate. - Do not invent horsepower/voltages/materials unless provided. - If data is insufficient, output a JSON with key "missing_fields": [ ... ] and do not generate content. Output format: Return only valid JSON. No additional commentary. Example output schema: { "name": "...", "meta_title": "...", "meta_description": "...", "friendly_url": "...", "short_description": "...", "description_html": "...", "features_list": [{"name":"","value":""}, ...], "tags": "...", "image_alt_texts": ["...","..."] } Category prompt template System role: You are a conversion and SEO content specialist for PrestaShop categories. Context: Store, brand voice, language, audience, primary category keyword(s). Input: { "category_raw_name": "{RAW_CATEGORY_NAME}", "products_count": {N}, "main_attributes": ["attr1","attr2"], "purpose": "{category purpose or users}" } Task: - Generate: category_name, meta_title(50–60 chars), meta_description(140–160 chars), friendly_url (slug), short_intro (1-2 short sentences for listing), description_html (150–300 words with H2/H3, benefits + CTA), suggested subcategory names (3–6), suggested CMS pages to link (e.g., size guide). - Provide JSON only. Do not invent product-level info. CMS page prompt template (About, Shipping, Returns, Size Guide) System role: You are a technical content writer specialized in e-commerce policy & help pages. Context: Store name, tone, country legal context (e.g., EU consumer rights), language. Input: { "page_type": "about|shipping|returns|size_guide|privacy", "key_points": ["point1","point2"] } Task: - Produce: title, meta_title (50–60 chars), meta_description (140–160 chars), friendly_url, body_html (clear headings H2/H3, bullet points, step-by-step instructions if relevant), FAQ section (3–7 Q&As). - For legal topics, include suggestion: “Consult local law—do not rely solely on this text.” - JSON-only output. Example filled product prompt + expected output (short) Prompt (inputs replaced): - Store: “GreenTrail Outdoor” - Brand voice: friendly, informative, slightly informal - Audience: hikers and campers in EU, language: English - Keywords: primary “lightweight camping stove”, secondary “portable backpacking stove, compact stove” Input product: { "sku":"GT-PS100", "name":"GT Ultralight Stove 2026", "manufacturer":"GreenTrail", "features":["Weight: 210 g", "Fuel: canister", "Boil time: 3 min 30 s", "Pack size: 10 x 6 cm", "Material: stainless steel"], "images":["stove1.jpg","stove2.jpg"], "intended_use":"backpack camping, ultralight hikers", "notes":"CE compliant" } Task: (use the product template above) Expected output (JSON keys only — truncated example): { "name":"GreenTrail Ultralight Camping Stove — 210 g", "meta_title":"GreenTrail Ultralight Camping Stove | Lightweight Stove", "meta_description":"Lightweight camping stove for backpacking — 3.5 min boil time, 210 g. Compact, CE-compliant. Shop now.", "friendly_url":"greentrail-ultralight-camping-stove-210g", "short_description":"210 g ultralight camping stove with fast boil time and compact pack size — ideal for backpackers.", "description_html":"
The GreenTrail Ultralight Camping Stove weighs just 210 g and delivers a 3 min 30 s boil time with standard canister fuel...
- Weight: 210 g ...
Buy now to cut pack weight without sacrificing performance.
", "features_list":[{"name":"Weight","value":"210 g"},{"name":"Boil time","value":"3 min 30 s"},...], "tags":"camping stove,lightweight,backpacking,portable", "image_alt_texts":["GreenTrail lightweight camping stove top view","Close-up stove burner and pot supports"] } Bulk generation considerations (CSV/JSON import) - Provide a template CSV with columns matching the JSON keys defined above. - Use the same prompt but iterate rows programmatically; set temperature low (0–0.2) for consistent outputs. - Use a unique identifier column like sku or id to map outputs back. - Validate outputs with a script: check meta lengths, friendly_url format, no missing required fields, no HTML in fields that shouldn’t contain it. Few-shot examples and safety - Provide 1–3 examples in the prompt if you want a consistent s. - Tell the model NOT to fabricate specs; if missing, it should return missing_fields. - Ask for alt_texts and slugs that are unique and keyword-aware. Testing and QA checklist - Meta title <= 60 chars, meta_description 140–160 chars. - Friendly URL: no spaces, lowercase, hyphens, no special chars. - Short description: 20–40 words, no HTML. - Description_html uses allowed tags only and is user-friendly. - No hallucinated specs (verify feature values against source). - Unique content across products (use plagiarism/duplication checks). - Check for required legal lines on returns/shipping pages (depending on country). Prompt tuning tips - Use system message: “You are an e-commerce copywriter...” - Keep temperature low for bulk/deterministic output; raise slightly for creative product names. - Use function calling or explicit JSON schema if your LLM supports it to enforce structure. - For translations: include “Translate outputs into X language while preserving SEO keywords and match local idioms” and ask for both original and translated fields in the JSON. Quick ready-to-use one-liners (for quick generation) - Product rewrite: “Rewrite the following product description for SEO and conversion, keep primary keyword {keyword}, produce JSON fields: name, meta_title, meta_description, friendly_url, short_description, description_html, features_list. Do not invent specs.” - Category page: “Write a category page for {category} targeting keyword {keyword}. Produce JSON: category_name, meta_title, meta_description, friendly_url, short_intro, description_html, subcategories.” Final checklist to save time - Always include primary keyword(s), brand voice, language. - Provide as many factual inputs as possible (weights, materials, model numbers). - Specify output format and constraints. - Validate outputs automatically before import. - Iterate: ask the model to score the output on SEO, readability, and conversion and then refine. If you want, I can: - Produce a copy-ready prompt for a specific product or category if you paste the raw product data. - Generate a CSV template for bulk generation. Which would you like me to do next?