Coding & Automation prompt
ChatGPT Prompt to Design a Zapier, Make or n8n Automation
ChatGPTClaudeGemini
The short answer
Turns a tedious manual process into a step-by-step automation blueprint, with triggers, field mappings, error handling and a test plan. Copy the prompt below, replace the 6 [BRACKETED] variables with your own, and paste it into ChatGPT, Claude or Gemini.
You are an automation architect experienced with Zapier, Make and n8n. You design workflows that are reliable, cheap to run, and easy for someone else to maintain.
CONTEXT
- Platform I will use: [PLATFORM] (Zapier, Make, n8n, or "recommend one")
- The manual process today, step by step: [CURRENT_PROCESS]
- Apps and accounts involved (with plan level if known): [APPS_INVOLVED]
- How often it happens and typical volume: [FREQUENCY_AND_VOLUME]
- What a successful run produces: [DESIRED_OUTCOME]
- My technical comfort (no-code only, can write a little JavaScript/Python, developer): [TECH_LEVEL]
TASK
1. If I asked for a recommendation, compare the platforms for this specific workflow (fit, complexity, likely cost drivers such as tasks or operations per run) and pick one.
2. Design the workflow as numbered steps: trigger, each action, filters, routers or branches, data transformations, and the final output.
3. For each step, list the exact fields to map from earlier steps.
4. Add error handling: what happens if an app is down, data is missing, or a duplicate arrives.
5. Write a test plan with 3 to 5 test cases, including at least one bad-data case.
RULES
- Use only triggers and actions you are confident exist for these apps. If unsure whether a native integration supports something, say so and offer a fallback (webhook, HTTP request, or code step).
- Do not quote prices or plan limits as fact; tell me what to check on the vendor's pricing page.
- Keep the number of steps as low as possible. Explain any step that is not obviously necessary.
- Flag anything involving personal data or credentials and how to handle it safely.
- If the process has a human judgment step that should not be automated, keep a human in the loop and say where.
OUTPUT FORMAT
Platform choice (if requested) / Workflow outline as an indented list / Step details with field mappings / Error handling / Test plan / Maintenance notes (what is likely to break first).What to fill in
Everything in [GOLD CAPS] is yours to replace. Leave one blank and the machine will guess — and the machine guesses like a meatbag on a Monday.
[PLATFORM][CURRENT_PROCESS][APPS_INVOLVED][FREQUENCY_AND_VOLUME][DESIRED_OUTCOME][TECH_LEVEL]
How to use it
- Write out the manual process in boring detail, including the step where you squint at a spreadsheet. That squint is usually a filter the automation needs.
- Check every integration it names against the platform's app directory before you build. Models invent triggers with the confidence of a junior consultant.
- Run the bad-data test case first. Automations fail on the weird row, never on the clean one you demoed to your boss.
Example
PLATFORM: recommend one
CURRENT_PROCESS: new Typeform lead -> I copy it into HubSpot -> check if company size > 50 -> if yes, post in #sales Slack and email the rep
FREQUENCY_AND_VOLUME: around 30 leads a day
TECH_LEVEL: no-code onlyWorks with
Tested for ChatGPT, Claude, Gemini. Any modern chat model that follows long instructions will cope; the weaker ones will need you to restate the output format if they drift.
Weapons from the Arsenal that pair well with this prompt: