workflow: Zapier → ChatGPT for automated content briefs

Recommendation
Use a Zap with: Trigger = Google Sheets (New Row), small data-cleaning steps (Filter/Formatter), Action = ChatGPT (Create Chat Completion), then parse/save the JSON returned. Use low temperature (0–0.2) for consistent briefs and gpt-4 only when you need higher quality and can afford it.

Exact Zap setup (step‑by‑step)
1) Trigger — Google Sheets: New Spreadsheet Row
- App: Google Sheets → Trigger Event: New Spreadsheet Row (or New/Updated Spreadsheet Row if you edit rows)
- Connect Google account. Pick Drive → Spreadsheet → Worksheet. Make sure the sheet has a header row with columns like: Keyword, Intent, WordCount, Audience, Tone, ExampleURLs.

2) Step — Filter (Zapier)
- Only continue if Keyword is present: “Keyword” (Text) → Exists / Is Not Empty. Prevents blank-row noise.

3) Optional — Formatter (Zapier) → Utilities / Text
- Trim whitespace, normalize casing, or join multiple example URL cells into one field.

4) Action — ChatGPT (OpenAI in Zapier)
- App: ChatGPT or OpenAI (Create Chat Completion)
- Model: gpt-4 (best quality) or gpt-3.5-turbo (cost saving)
- Temperature: 0.0–0.2
- Max Tokens: 800–1500 (adjust by output size)
- Input / Prompt: use the template below (replace placeholders with sheet field tags).

Prompt template (use system + user messages if available)
System: You are an expert SEO content strategist. Output JSON only, no explanation. Use keys exactly: title, meta_description, primary_keyword, search_intent, recommended_word_count, headings (array of {heading, depth, suggested_word_count, notes}), internal_links (array of {url, anchor}), external_refs (array of urls), CTA, notes.

User: Create a content brief for the keyword: "{{Keyword}}". Search intent: "{{Intent}}". Audience: "{{Audience}}". Tone: "{{Tone}}". Target word count suggestion: "{{WordCount}}". Example referenced URLs: "{{ExampleURLs}}". Keep result concise, action-ready, and in valid JSON only.

Example expected output (for testing): {"title":"...","meta_description":"...","primary_keyword":"...","search_intent":"...","recommended_word_count":1200,"headings":[{"heading":"H1 - ...","depth":1,"suggested_word_count":150,"notes":"..."}],"internal_links":[],"external_refs":[],"CTA":"...","notes":"..."}

5) Action — Formatter / Code (JS) to parse JSON
- Use “Formatter → Text → Extract Pattern” or “Code by Zapier (JS)” to validate/parse the JSON string. If invalid, route to error handling.

6) Action — Update Google Sheets / Create Row
- Save parsed fields back to the spreadsheet (new sheet called Briefs) or push into CMS via API.

Error handling & reliability tips
- Validate incoming rows: Filter and a Lookup step to prevent duplicates (use a unique ID column).
- Rate limiting: if you push many rows, batch them or add a Delay step; OpenAI rate limits and cost can spike.
- Parsing failures: If ChatGPT returns non‑JSON, add a “Re-prompt” (send back: “Output must be valid JSON only — reformat now.”) or log the row to a “Failed” sheet and notify Slack/email.
- Retries & alerts: Enable Zapier’s auto-retry for temporary API errors and add a Slack/Email action for permanent failures.
- Backoff: On repeated 429s, add a Delay and exponential backoff (Delay for 30s, 2m, etc.) via Paths.

Decision criteria (quick)
- Choose gpt-4 when brief quality and nuance matter (higher cost).
- Choose gpt-3.5-turbo if volume and budget matter more than tiny qualitative gains.
- Use single-turn prompts for speed; multi-turn if you need iterative refinement (extra cost & complexity).

Best‑for / Avoid‑if
- Best for: small-to-medium teams automating 10–200 briefs/week; quick standardized briefs.
- Avoid if: you need fully human-edited, publication-ready content (use brief as draft only) or you must strictly guarantee factual citations without human review.

Practical checklist before turning Zap on
- [ ] Sheet headers: Keyword, Intent, Audience, WordCount, Tone, ExampleURLs
- [ ] Connect Google & OpenAI API keys in Zapier
- [ ] Add Filter for non-empty Keyword
- [ ] Choose model & set temperature
- [ ] Add JSON parse step and a fallback “Failed” sheet
- [ ] Add notifications for failures (Slack/email)
- [ ] Run 10–20 test rows, validate outputs

When the right setup varies
- Budget: high volume → cheaper model + batch processing.
- Skill level: non-dev → stick to Formatter and Filter; devs can add Code steps to robustly parse and transform.
- Workflow stage/team size: small solo teams can append brief JSON to sheet; larger teams should push into a CMS with review steps.

If you want, I can convert the above prompt to the exact Zapier field values (with sample field tags from your sheet) or provide a ready-to-import JSON example for the ChatGPT action.

Compare Zapier and Make