QWE AI Academy You want ChatGPT to read your Google Sheet, generate personalized emails, and send them via Gmail – automatically, every time a new row appears. Make.com promises exactly that: visual automation, no code required, AI baked in.
The setup takes 20 minutes. The workflow runs beautifully. Then, three days later, you’ve burned through 900 of your 1,000 free operations and the automation stops.
What happened? You set the trigger to check for new rows every 5 minutes. Make charged you 288 operations per day just to ask “anything new?” – even when the answer was “no.”
That’s the gap between Make tutorials and Make in production.
Most guides skip the part where your workflow dies from operation starvation. Or where OpenAI calls timeout at exactly 5 minutes because you tried Deep Research. Or where Google Drive file transfers randomly fail with cryptic errors.
What You’re Actually Building
A Google Sheets row triggers an OpenAI API call. AI generates content based on the row data. Make writes the result back. Real use: product descriptions, email replies, research summaries at scale.
End result: a spreadsheet control panel for AI-generated content. Processes dozens or hundreds of items, zero manual copy-paste.
The Parts You Need Before You Start
Make orchestrates. It’s not the AI. Three accounts:
- Make.com – Free plan: 1,000 operations/month (as of 2026). Testing only.
- OpenAI API account – New accounts get $5 free credits, valid 3 months (as of 2026). Not ChatGPT Plus.
- Google Sheets – Your data source. Input column, output column.
Go to Make’s pricing page and sign up. Then OpenAI’s API dashboard to generate your key. Paste it into Make later.
One catch: as of November 2025, all paid Make plans let you use your own AI provider connection (your OpenAI key) instead of Make’s. Free plan? You’re stuck with Make’s provider.
Building the Workflow
Log into Make. Click Create a new scenario. Blank canvas, plus button.
Step 1: Set the trigger (and avoid the polling trap)
Search Google Sheets, pick Watch Rows. Connect Google account, select spreadsheet and sheet.
The trap: Make defaults to polling – checks your sheet on a schedule. Set it to 5 minutes? 12 checks/hour × 24 hours = 288 operations/day = 8,640/month. 90% of your free plan gone checking a folder that gets new data once a week.
Fix: webhooks (if supported), or accept the cost. For this tutorial: 15-minute polling (free plan minimum).
Step 2: Connect OpenAI
Click the plus after Google Sheets. Search OpenAI, select Create a Completion.
Paste API key. Model: gpt-4o-mini (cheapest, fast). In Messages, click icon to map spreadsheet data.
Prompt structure matters. Not “write a description.” Try: “Write a 50-word product description for: [Product Name]. Target audience: [Audience]. Tone: professional.” Map [Product Name] and [Audience] to sheet columns.
Watch out:Make’s OpenAI module timeout: hardcoded at 5 minutes (as of 2026). High-reasoning models (GPT-5, o1) or complex prompts? Call fails mid-generation. Workaround: break long tasks into smaller API calls or use a different layer for heavy reasoning.
Step 3: Write the result back
Add Google Sheets > Update a Row. Same spreadsheet. Map Row ID from trigger. Output Column: paste OpenAI response field. Run once manually. Check sheet. AI response in right column? Done.
The Hidden Costs (That Add Up Fast)
Scale this to 100 rows and here’s what breaks:
Operations aren’t tasks.One Zapier task = 3-8 Make operations. This workflow: 3 operations per row (read, call OpenAI, write). Free 1,000-operation plan = ~330 real executions.
Data transfer limits.Free: 100MB/month. Core: 1GB/month (as of 2026). Images or video? You’ll hit this first. Also, Google Drive file downloads to OpenAI Vision fail randomly with timeout errors – remember that 100MB ceiling from earlier? Pass URLs instead of transferring files through Make.
The custom field problem.Destination app has custom fields? Make’s module often can’t map them. HTTP module for raw API calls. Adds complexity.
| Cost Factor | Free Plan | Core Plan ($10.59/mo) |
|---|---|---|
| Operations | 1,000/month | 10,000/month |
| Real workflow runs | ~330 (3 ops each) | ~3,300 |
| Data transfer | 100MB | 1GB |
| Min interval | 15 minutes | 1 minute |
When Make Isn’t the Right Tool
Sub-5-minute AI responses at high volume? Make’s pricing model breaks. Per-operation billing works for predictable workflows; fails when demand spikes.
Regulated industry? Free plan doesn’t let you bring your own AI provider. Custom provider connections (your OpenAI key) require paid plans. Matters for data residency, compliance logging.
Alternatives exist. Zapier: simpler, pricier. n8n: powerful, self-hosted. Make: visual enough for non-coders, flexible for complex logic, but cost traps if you don’t plan operations.
FAQ
Can I use Make.com for free long-term?
Free plan works for testing – 1,000 operations/month with scenario run limits (as of 2026). Most production workflows exceed this fast. Especially with polling.
Why does my OpenAI module keep timing out?
All OpenAI calls from Make timeout at exactly 5 minutes (as of 2026). Deep Research? GPT-5 with high reasoning? Large contexts? Call fails. I tried running a multi-step research workflow once – burned 400 operations before I realized the 5-minute wall kept killing it. Workaround: use streaming endpoints via HTTP module, or chunk the task into smaller pieces.
What’s the difference between Make operations and Zapier tasks?
One Zapier task can equal 3-8 Make operations. Even at 2x operation usage, Make is 2.5x cheaper than Zapier (as of 2026) for most workflows. Savings are real, just not as dramatic as raw pricing suggests.