Your first automation: form → AI summary → draft → spreadsheet, step by step (Day 22 of the 30-Day Challenge)

The task

Build the challenge's first automation end to end in Zapier: a form submission triggers an AI summary, drafts a reply email, and logs a spreadsheet row. This is the Day 22 build from the 30-Day AI-Native Challenge, fully worked — both lanes do this one.

Before AI

Every inbound request (demo form, support ask, intake) means: read it, summarize it for the tracker, draft an acknowledgment, copy details into a sheet. Four manual steps per submission, none of them judgment.

What you'll need

• A Zapier account (free tier runs a 2-step Zap; this recipe uses 4 steps, which needs a trial or starter plan — Make is the free-tier alternative, same shape) — VERIFY current limits
• A form (Google Forms or Typeform) and a Google Sheet
• 30–45 minutes the first time

First, pick the right task. The Day 22 rule: high-volume, clear-rules, low-stakes. A demo-request form qualifies. Anything involving money, commitments, or upset customers does not — that judgment call is the exercise.

The workflow

1. Trigger — New form submission. Zapier → Create Zap → Trigger: Google Forms "New Form Response." Connect the account, pick the form, pull in a test row.

2. Action — AI summary. Add step: "AI by Zapier" (or ChatGPT/Anthropic step). Paste this prompt, mapping the form fields into the brackets:

Summarize this inbound request in 2 lines for a tracking sheet:
Line 1: who + company + what they want, in plain words.
Line 2: urgency (low/medium/high) + why, based only on what they wrote.

Do not invent details that aren't in the submission. If the request is unclear,
urgency is "unclear" — not a guess.

Name: {{Name}}
Company: {{Company}}
Message: {{Message}}

3. Action — Draft the acknowledgment (as a DRAFT, not a send). Add step: Gmail → "Create Draft":

Write a 60-word acknowledgment to {{Name}}:
- Thank them, reference their specific ask in one clause (from: {{Message}})
- Say a human will reply within [your real SLA]
- No promises about outcomes, pricing, or dates beyond that.
Sign-off: [your name]

Creating a draft instead of sending IS your approval gate — the same pattern your Day 26 agent will formalize. You read, then you send.

4. Action — Log the row. Add step: Google Sheets → "Create Spreadsheet Row." Map: timestamp, name, company, the 2-line AI summary, urgency, status = "new."

5. Test with three real-ish submissions — one clear, one vague, one weird (paste an off-topic rant). Check: does the vague one get "unclear" instead of a confident guess? Does the weird one produce a sane draft? Edge cases are the test, not the happy path.

6. Turn it on. Run it for a week in draft-gate mode before you even consider auto-send (and for anything client-facing: don't).

Verify it worked

After a week: every submission has a sheet row, every draft needed ≤1 edit, and zero drafts promised something you didn't authorize. That last check is the one that matters — reread step 3's constraint if it fails.

Troubleshooting

• AI step returns fluff? Your form fields aren't mapped — check the {{brackets}} actually contain data in the test view.
• Drafts overpromise? Tighten the "no promises" constraint; list the specific things it must never say ("free trial," "custom pricing," dates).
• Zapier step limits? Make's free tier allows longer scenarios; the recipe translates 1:1 (Webhook → OpenAI/Anthropic module → Gmail draft → Sheets row).

Reality check

This saves 15 minutes a day, not your career. Its real value is the pattern: trigger → AI transform → human gate → action. Day 26 upgrades exactly this shape into a real agent with tools and a loop.

Data & security

Form submissions carry PII (names, emails). Check where your AI step processes data and confirm it's within your Day 6 boundary; don't route sensitive intake (health, finance, legal) through consumer AI steps.

Going further

Upgrade path: the capstone agent with an approval gate (Day 26) makes the gate explicit and adds a decision loop; the eval harness (Day 27) tests it on known answers before you trust it.

Your takeaway

One recurring chore fully automated behind a human gate — challenge artifact "First Automation," and the trigger→transform→gate→action pattern you'll reuse for every automation after it.