Why this matters
The freelancer who built the booking automation has moved on. It ran 212 times without a problem, and this week four group inquiries got no reply, each worth $4,000 to $9,000. The run history shows two error messages that mean nothing to the front desk.
ChatGPT can read those messages next to what changed last week and tell you, in plain words, what most likely happened. The habit that makes it safe: ask what it is sure of and what it is guessing before you ask for the fix.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
What you’ll build
You run the front desk at The Alder House, a 22-room hotel (made up). You attach what the automation does, what changed last week and the failed runs, and get each error explained with a "sure" list and a "guessing" list, plus whether Thursday's quick fix helped.
Then you get the fix in a safe order: pause, check the spreadsheet step, add a check for a missing email, test with your own address, turn it back on, and follow up with the four groups by hand. And a short note to the colleagues whose changes broke it.
What you need
- Nothing to start the practiceThe lesson is a practice copy of ChatGPT. The hotel, guests and log are made up.
- A ChatGPT account, to do it for realAny plan. The answers shown came from a Plus account.
- The run history of the failed runsYour tool's executions or run history page: which step failed and the exact error text. Copy it as text.
- What changed recentlyAsk around: renamed tabs or columns, changed form fields, new passwords or permissions.
Concepts
What the two errors meant
"Unable to parse range: Inquiries!A:F" came from the Google Sheets step. The automation was writing to a tab called Inquiries, which no longer existed after someone renamed it on Monday.
"Invalid To header" came from the Gmail step. The guest left the email field empty, which marketing had just made optional, so there was no address to send to.
Sure versus guessing
ChatGPT was sure the spreadsheet step failed four times and the email steps never ran. It was guessing, strongly, that the rename caused it, because the file shows the timing but no technical proof. That split tells you what to check before you change anything.
It also judged Thursday's quick fix correctly: better, not fixed. The spreadsheet step worked; the email step then failed for a different reason.
Fix safely, and ask one more question
Pause it, fix one step at a time, test with a made-up guest and your own email address, and never replay failed runs blindly: that can send duplicate emails. The four groups get a personal reply today.
Thursday's fix pointed the step at the new, empty tab "Inquiries 2027". ChatGPT accepted that without asking. Before you turn it back on, ask the events manager which tab new inquiries belong in.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- failed-runs-this-week.txtThe made-up run log used in the lesson.
Quick check
3 questions to check you’ve got it.
1. Why ask what ChatGPT is sure of and what it is guessing?
- To make the answer longer
- So you know what to check before you change anything
- Because it is required by the tool
Show the answer
So you know what to check before you change anything. A strong guess is still a guess. The "sure" list is what you can act on; the "guessing" list is what you confirm first.
2. What does "Invalid To header" from Gmail usually mean?
- The subject line is too long
- The recipient address is empty or not a valid address
- Gmail is down
Show the answer
The recipient address is empty or not a valid address. Here the guest left the email field blank, so there was no address to send to.
3. Four runs failed. Should you replay them all once it is fixed?
- Yes, right away
- Not blindly: check what already happened, or you may send duplicates
- Only the oldest one
Show the answer
Not blindly: check what already happened, or you may send duplicates. Some steps may have half-run. Check the sheet and the sent mail first, then reply to the people by hand.
You reached the end of Plan, Prompt and Fix Automations with ChatGPT: Before and After You Build in n8n, Make or Zapier
You can now plan an automation, prompt its AI step and fix it when it breaks. Next, build one: n8n for Beginners, AI Automation with Make or AI Automation with Zapier.
Pick your next course →