how: Zapier → ChatGPT for e‑commerce product descriptions

Recommendation (short): Use a Zap that triggers on "New Product" (or "Product Updated") in Shopify, packages only the essential fields into a concise JSON prompt, sends that to the ChatGPT API via Zapier’s Webhooks/Code step, parses a structured JSON response, and then updates the Shopify product (or writes to a review queue). Keep temperature low, enforce token/output length limits, and add a human QA step for the first 100 products.

Field mappings (what to send to ChatGPT and suggested response schema)
- Send from Shopify to ChatGPT: product_id, title, vendor, product_type, tags, raw_description (if any), images (first image URL), variants (list of SKU, price, option values), materials, dimensions, target_audience, 3 brand-tone keywords (e.g., "friendly, premium, concise"), seed_SEO_keywords (comma list).
- Ask ChatGPT to return JSON only with fields: seo_title, meta_description (max 160 chars), body_html (product description using simple HTML paragraphs and

/
• for bullets), short_description (1–2 lines), bullets (array of 3–6 benefit bullets), suggested_tags (array), safety_claims (explicit; must match input specs or be empty).

  Prompt constraints (practical template & limits)
  - System: "You are a copywriter for [brand], tone: {brand-tone}. Do NOT invent product specs. If a spec is missing, leave the field blank or request human review."
  - User prompt (concise): provide the JSON product payload. Instruction bullets:
  - Produce SEO title (max 60 chars) including primary keyword.
  - Produce meta description (max 155 chars).
  - Produce body_html: 120–220 words, 3 benefit bullets, one short technical-specs list. Use

  and

  . No promotional claims beyond provided specs.
  - Provide suggested_tags (max 8).
  - Output only valid JSON.
  - API params: temperature 0.0–0.3 (consistency), max_tokens for output 400–600.

  Batching and rate-limit advice
  - For single new-product triggers: use immediate zap. For many imports or CSV uploads: collect new/updated product IDs into a staging sheet (Google Sheets) or use Zapier’s Scheduler to batch 10–25 products per run. Batching reduces API calls and is cheaper per-run but requires your prompt to handle arrays and return arrayed outputs.
  - Recommended batch size: 5–20 products per ChatGPT call for balanced cost and manageability. >50 increases chance of partial failures and longer latency.

  Best-for / Avoid-if
  - Best-for: stores with structured product data, consistent brand voice, moderate SKU churn ( smaller batch sizes or keep Zapier single-product triggers with manual QA. Larger budgets -> larger batches, more auto-updates, A/B variations.
  - Skill level: low -> use Zapier Webhooks with a simple template and manual approval step. High -> use Zapier Code step for parsing and error handling.

  Practical checklist
  1) Map Shopify fields to the JSON payload (title, vendor, tags, images[0], variants, specs).
  2) Create prompt template with system + user messages and example output.
  3) Build Zap: Shopify trigger -> Formatter/Code (clean fields) -> Webhooks/ChatGPT call -> Formatter/Code (parse JSON) -> (optional) Slack/Google Sheets approval -> Shopify Update Product.
  4) Test with 10 varied products.
  5) Turn on logging, set alerts for parsing errors, and implement manual QA for the first 100 descriptions.
  6) Monitor token usage and adjust batch size.

  If you want, I can draft a ready-to-copy prompt + example JSON response and the exact Zap step list for Zapier->ChatGPT->Shopify.

  Compare Zapier and Make