Beginning
Think about the last time you left feedback for a business, a review, a comment, a short message after a purchase. Did anyone answer? Most of the time nobody does, and that silence quietly tells the customer that their opinion didn't matter.
Small businesses don't ignore feedback on purpose. They read it late, or they read it, mean to reply, and then the day gets busy. This lesson builds a system that answers every piece of feedback within minutes, and, just as importantly, knows when it should not answer by itself.
This is the second project of Phase 4, and it brings together a lot of what you've learned: Google Sheets, OpenAI, filters, and Gmail. It also introduces one new idea that matters far more than any single module, which is letting AI draft a reply while a human stays in charge of the risky cases.
In this lesson, you'll learn:
- How to collect feedback with a free Google Form that fills a Google Sheet automatically
- How to use the Watch New Rows trigger so your scenario reacts to every new response
- How to make AI label a message as Positive, Negative or Neutral, and why the answer must be a single word
- How to write a reply prompt that sounds human and never promises refunds or discounts
- How to use a filter so that only happy customers get an automatic reply
Core Concept: The Reply Desk With a Safety Gate
Picture the customer-service desk of a small shop. A trainee reads every note that comes in, sorts it into a pile, and writes a draft reply. The thank-you notes go out straight away. The complaints, though, go to the manager first, because a wrong word in a reply to an unhappy customer can make things worse.
That is exactly the system you're building. Google Forms collects the notes, the AI is the trainee, and a filter is the rule that decides which pile a note belongs to.
The pattern should feel familiar. In Lesson 11 you learned Read → Think → Respond. This project adds one step in the middle: Read → Think → Decide → Respond. The AI labels the feedback (that's the thinking), and the filter acts on that label (that's the deciding). Only then does anything get sent.
There's a name for this in the automation world: human in the loop. It simply means the automation handles the easy cases, and a person handles the risky ones. It's one of the most valuable habits you can build as an automator, and it's the reason clients trust a system like this with their real customers.
One honest note about timing. In Lesson 13, the chatbot answered instantly because a webhook fires the moment something arrives. The Watch New Rows trigger works differently. It checks the sheet on a schedule, which is every 15 minutes on Make's free plan. For feedback replies that's perfectly fine, since a customer who gets an answer within a quarter of an hour is still impressed.
Tool Walkthrough: Building the Feedback Reply System
You'll need your Google account and the OpenAI connection from Lesson 10. Nothing else.
Step 1 — Create the Google Form
Open Google Forms and create a new blank form. Name it "Customer Feedback" and add three questions:
- Name (Short answer)
- Email (Short answer)
- How was your experience? (Paragraph)
Mark all three as required. Then open the Responses tab and click Link to Sheets to create a new spreadsheet that fills up automatically with every submission.
Step 2 — Prepare the Sheet and Send a Test Response
Open the linked sheet. You'll see columns for Timestamp, Name, Email and the feedback answer. Add two new column headers right after them: Sentiment and AI Reply. Leave them empty, because Make will fill them in.
Now submit one test response through your form. Use a second email address that you own, so that you can actually receive the reply later, and write something clearly positive, for example: "The service was wonderful and the staff were so friendly."
Step 3 — Add the Watch New Rows Trigger
Create a new scenario, search for Google Sheets, and choose Watch New Rows. Connect your Google account, select your feedback spreadsheet and the correct sheet tab, and make sure the table contains headers in row 1. Set the limit to 1 so each run handles one response at a time.
Click Run once. The first time, Make asks Choose where to start. Select All, so the module also reads the test row that is already sitting in your sheet, then click Save. Make remembers how far it has read, so after this first run it only picks up rows that are new.
Step 4 — Add the Sentiment Check
Click the + after the trigger, search for OpenAI, and choose Generate a response. Select your existing connection and the GPT-4o model.
This first AI step has one small job: read the feedback and answer with a single word. Paste the prompt below, then replace [Feedback] at the end with the feedback chip from your trigger module.
Scroll down and set Max Output Tokens to `16`. That is the lowest value OpenAI accepts, and it's still far more than a one-word answer needs. The small limit also stops the AI from wandering into a long sentence.
Step 5 — Add the Reply Writer
Add a second OpenAI → Generate a response module after the first one. This one writes the actual reply. Paste the prompt below, then replace [Business Name] with your own business name, and map the Name and Feedback chips from your trigger module.
Set Max Output Tokens to `200` here, so replies stay short. The prompt also tells the AI to write only the email body, because the subject line is added later by the Gmail module.
Why two separate AI steps instead of one big prompt? Because small, single-purpose jobs are far more reliable. One step only judges the mood. The other only writes. When something goes wrong, you immediately know which step to fix.
 |
Step 6 — Save the Results to the Sheet
Add Google Sheets → Update a Row. Select the same spreadsheet and sheet, then fill in:
- Row number: map the Row number from the Watch New Rows module (never type a number by hand)
- Sentiment: map the Result from the first OpenAI module, the sentiment check (shown as `2. Result` in our build)
- AI Reply: map the Result from the second OpenAI module, the reply writer (shown as `3. Result` in our build)
This step matters even for customers who never get an automatic email. It is where the owner finds the draft reply for every review.
Step 7 — Add the Safety Filter
Click the line between Update a Row and the next module (you'll add Gmail in a moment), and choose Set up a filter. Name it "Positive only". For the condition, pick the Result from the first OpenAI module (the sentiment check, `2. Result` in our build), set the operator to Equal to, and type `Positive`. Don't pick the second OpenAI module here. Its Result is the long reply text, which will never equal a single word.
From now on, only feedback labeled Positive continues past this point. Negative and Neutral feedback stops here, with its drafted reply safely waiting in the sheet.
Step 8 — Send the Reply with Gmail
After the filter, add Gmail → Send an Email. Fill it in like this:
- To: map the Email chip from your trigger module
- Subject: type `Thank you for your feedback, ` and then add the Name chip. Keep the comma and the space before the chip, otherwise the subject reads "feedbackSara"
- Content: map the Result from the second OpenAI module (`3. Result` in our build)
As you saw in Lesson 14, the To field must always be mapped, never typed. If you type your own address while testing and forget to change it, every future customer's reply goes to you.
Step 9 — Test Both Paths, Then Activate
Click Run once. Your positive test row should pass through every module, and you should receive a reply email at your second address. Check your sheet too: the Sentiment and AI Reply columns should now be filled in for that row.
Then submit a second response through your form, this time clearly negative, such as: "I waited 45 minutes and nobody apologised." Run the scenario again. This time the run should stop at the filter, no email should arrive, and the sheet should show Negative with a drafted apology.
When both paths behave correctly, switch the Active toggle on.
Real Example: The Guesthouse That Answers Every Guest
A small guesthouse puts a QR code at the checkout desk, linked to the feedback form. One evening a guest writes that the breakfast was the best part of their stay. At the next scheduled check, the guest receives a warm email that mentions the breakfast by name and invites them back.
Another guest writes that the Wi-Fi never worked in their room. No email goes out. Instead, the owner sees "Negative" in the sheet next to a ready-made apology. She adjusts a couple of lines, calls the guest the next morning, and offers a small gesture that only she is in a position to offer.
The owner spends a few minutes a day instead of an hour, and no unhappy guest was ever answered by a machine. That is what makes this a service worth selling: it saves time without adding risk.
Common Mistakes
Mistake 1: Letting the sentiment step answer with a sentence
If the AI replies with "The mood seems positive," that text never equals "Positive," so the filter blocks every customer. The fix is the one-word rule in the prompt and the low token limit. After your first test, always check the Sentiment column and confirm it contains one clean word.
Mistake 2: Sending automatic replies to unhappy customers
A generic apology can make an angry customer angrier. The filter exists to prevent exactly that, so don't remove it just to make the scenario "reply to everyone."
Mistake 3: Mapping the wrong thing
Typing a fixed email address in the To field, or typing a row number by hand in Update a Row, works perfectly in your test and fails silently with real customers. Map both from the trigger.
Try This Yourself
Before moving to Lesson 16, build this end-to-end on your own, then try these extensions:
- Submit one clearly angry review and confirm that nothing is emailed but the draft appears in the sheet
- Add a Status column, then add a second Update a Row after Gmail that writes "Sent" once the email goes out
- Change the filter so Neutral reviews also receive an automatic reply, then compare the tone of the replies
Key Takeaways
- Google Forms plus Watch New Rows gives you a free, reliable way to turn any sheet into a trigger
- Two small AI jobs, one to classify and one to write, are more dependable than one big prompt
- A filter can work as a safety gate: automate the easy cases and keep a human on the risky ones
- This is a sellable service for restaurants, salons, shops and guesthouses: "customer feedback reply setup"
Frequently Asked Questions
Can this reply to Google Maps or Facebook reviews?
Not directly in this free setup, because those platforms need their own connections and approvals. The same classify-and-draft logic carries over, though, and you can always paste reviews into the sheet yourself and let the scenario draft the replies.
Can I send the replies through Telegram or WhatsApp instead of email?
Yes, once you have those accounts. Replace the Gmail module with the matching Send a Message module. The AI steps and the filter stay exactly the same.
Is it okay to store customers' emails and feedback in a sheet?
It is personal data, so keep the sheet private and collect only what you actually need. Lesson 16 covers this properly.
Next Lesson Preview
This project stores real customer names, emails and opinions, which makes the next lesson the natural one to take. Lesson 16 steps back from building and covers Data Privacy Basics + Common Automation Mistakes: what data you should and shouldn't collect, and how to avoid the pitfalls that trip up even experienced no-code builders.
Navigation:
← Previous: Lesson 14 — Project 1: Lead Capture → CRM → AI Follow-Up System
→ Next: Lesson 16 — Data Privacy Basics + Common Automation Mistakes