The Zero-Dollar Blogging Automation Stack: What I Built in Make After Burning Through My First 941 Operations
It’s 2:14 AM and my Make scenario is stuck in a loop, pinging the same Google Sheet row 38 times because I forgot to add a filter after a JSON array. My phone is buzzing with error emails. I’m not writing this from a place of “here’s a shiny system that changed my life.” I’m writing it from a place of having broken this exact stack twice, fixed it, and now actually trusting it enough to let it touch my blog while I sleep.
If you’ve been eyeing Zapier’s pricing page and closing the tab in horror, this is for you. You can build a genuinely useful zero-dollar blogging automation setup with Make.com’s free tier, ChatGPT, Claude, and a Google Sheet that looks ugly but works. No credit card. No trial period that quietly bills you in 14 days. Here’s what’s actually worth your time and what isn’t.
Why I Even Bothered With This
I run a one-person blog. No team, no VA, no budget for tools that charge per “task” like it’s a taxi meter. My problem wasn’t ideas — I had a spreadsheet with 63 topic ideas sitting untouched for four months. My problem was the boring connective tissue: pulling research, turning it into a draft, formatting it, and getting it into WordPress without doing the same five manual steps every single time.
Make’s free plan gives you 1,000 operations a month and 2 active scenarios. That sounds small until you realize a single blog post pipeline — trigger, fetch, AI call, format, push to WordPress — eats maybe 6 to 9 operations per run. I ran the math wrong the first week and thought I could publish daily. I couldn’t. More on that mistake below.
The Actual Stack (No Fluff, Just What’s in It)
1. Make.com — the glue, not the brain
This is the free orchestrator. It watches for triggers, moves data between apps, and calls APIs. It is not smart. Treat it like plumbing, not a strategist.
2. Google Sheets — your dumb, reliable database
I use one sheet with columns for Topic, Status, Research Notes, Draft, and Publish Date. It’s not elegant. It’s free, and Make reads and writes to it without hiccups, which is more than I can say for the Airtable free tier once you hit its row cap.
3. ChatGPT — for research, not for thinking
This is the part most people get backwards. ChatGPT (free tier, or via low-cost API calls) is genuinely strong at pulling together facts, comparing angles, and summarizing what’s already out there on a topic. Where it falls apart is when you ask it to actually reason through something — give it an open-ended “what do you think” prompt and it hands you safe, generic, dummy-feeling output. So I stopped asking it to think. I only ask it to fetch, list, and summarize. Research assistant, not strategist.
4. Claude — for writing, but only inside a role
Claude writes noticeably better prose than ChatGPT out of the box, but it’s cautious by default — it hedges, it over-explains, it sounds like it’s afraid of committing to an opinion. The fix that actually worked for me: force it into a specific persona before it writes anything. “You are a blunt, experienced blogger who’s tested this personally” produces a completely different draft than a bare prompt. Skip the role-setup and you get competent-but-forgettable copy. Include it and the writing has an actual voice. This one change did more for my draft quality than any prompt-engineering guide I paid attention to.
5. A free publishing target
WordPress.com’s free tier or Blogger both have APIs Make can hit without a paid connector. I push finished drafts here as unpublished posts, never auto-published — more on why in the limits section.
(Internal link opportunity: link to your own “Make.com vs Zapier free tier comparison” post here if you have one.) (External reference: link Make’s official [operations and scenario limits documentation] so readers can verify current free-tier numbers, since Make updates these occasionally.)
How I Actually Wired This in Make
- Trigger: A new row added to the Google Sheet with Status = “Ready for Research.”
- Step 1 — Research pull: Make calls the ChatGPT module with a strict, narrow prompt: “List 8 factual points, no opinions, on [topic].” It writes the raw output back into the Research Notes column.
- Step 2 — Writing pass: Make calls Claude with the research notes plus a role-based system prompt defining tone, audience, and one required personal-experience angle. Output goes into the Draft column.
- Step 3 — Formatting cleanup: A basic Make text parser strips stray markdown symbols ChatGPT sometimes leaves behind (asterisks in weird places, mostly).
- Step 4 — Push to WordPress: Draft posts as unpublished, Status updates to “Needs Human Eyes.”
That’s it. Five steps, roughly 7 operations per full run depending on how chatty the responses are.
The Mistake That Cost Me 3 Days and 941 Operations
I set the trigger to check the sheet every 5 minutes instead of every hour, “just to be safe.” I also hadn’t added a filter to stop the scenario from re-processing rows that already said “Done.” Result: it re-ran the ChatGPT research step and the Claude writing step on the same three rows, over and over, for about 6 hours before I checked my phone and saw the operations counter sitting at 941 out of 1,000 — on day 9 of the month. I spent the next 47 minutes hunting through the scenario history trying to figure out why, and the answer was one missing filter block that took 90 seconds to add once I found it. Cheap lesson, expensive way to learn it.
If you build this, add a filter checking Status ≠ “Done” before every AI-calling step. It’s the single most important safeguard in the whole thing.
Where This Stack Actually Breaks Down
Being honest about the limits matters more than selling you on the dream:
- Long-form depth suffers. Neither model, run through a narrow automated prompt, will produce a genuinely deep 2,500-word guide unsupervised. You’ll get a competent skeleton, not a finished authority piece.
- Fact-checking is still on you. ChatGPT’s research pass is a starting point, not a source. I’ve caught it confidently citing statistics that don’t check out on the actual source page. Verify before publishing.
- The free operations ceiling is real. 1,000 operations a month sounds like a lot until one bad loop eats 90% of it in an afternoon. Budget for maybe 15–20 full article pipelines a month if each run is clean, fewer if you’re iterating on drafts.
- Never auto-publish. I keep every post landing as a draft, not live. Automation gets you 70% of the way; the last 30% — actual editing, fact-checks, and making it sound like you — still needs a human sitting down and doing the work.
FAQ
Is Make’s free plan actually enough for regular blogging, or is this just a demo toy? It’s enough for a solo blogger publishing a few times a month, as long as you’re disciplined about filters and don’t let scenarios re-trigger on already-processed rows. Push past 4–5 posts a week and you’ll likely need the paid tier.
Why use both ChatGPT and Claude instead of just one? Because they’re genuinely different tools with different strengths. ChatGPT is faster and more compliant for narrow, factual research tasks. Claude, once you give it a defined role, writes with more personality and commits to an actual point of view instead of hedging everything. Using one for research and the other for writing plays to what each is actually good at.
Can I automate publishing directly instead of pushing drafts? Technically yes, Make supports it. I don’t recommend it. AI-generated drafts still need a human fact-check and voice-edit pass before they go live — auto-publishing skips the step that actually protects your credibility.
The Takeaway
The real win here isn’t that this stack is free — it’s that it forces you to stop treating AI tools as one interchangeable blob. ChatGPT pulls the raw material. Claude, given an actual role to inhabit, turns it into something readable. Make just makes sure those two handoffs happen without you copy-pasting between six browser tabs at midnight. Build the guardrails first, the filters, the “never auto-publish” rule, and the free version of this stack will do real work instead of just burning through your operations count by lunchtime.
