Day 3

A multi-step content pipeline in Make.com: raw notes submitted via Google Form → AI creates outline → AI writes draft → AI edits for clarity → finished article saved to Google Docs.

Why Chain Instead of One Big Prompt?

The temptation is to write one massive prompt: "Take these notes and write me a polished, well-structured, SEO-optimized article with strong headlines." The result is usually mediocre — it does everything passably and nothing well.

The better approach is the same one publishing houses have used for centuries: specialize each step.

# Single prompt (mediocre)
"Write a great article from these notes."

# Chained prompts (better quality)
Step 1: "Convert notes → structured outline (JSON)"
         Focus: structure only, not writing quality

Step 2: "Write first draft from outline"
         Focus: prose quality, not structure decisions

Step 3: "Edit draft for clarity and conciseness"
         Focus: cutting and tightening, not new content

Step 4: "Generate metadata: title, tags, meta description"
         Focus: discoverability, not content quality

Each step does one job well. The AI isn't trying to make structure decisions while writing prose while optimizing for SEO. The output quality is noticeably higher, and each step is independently debuggable — if the draft is bad, you fix step 2 without touching step 1 or 3.

Passing Data Between AI Steps

The key to reliable chaining is being deliberate about what you pass between steps. There are two approaches:

Pass the full text. Simple — just map the output of step 1 as input to step 2. Works well when the output is clean prose that the next step can work with directly.

Pass structured JSON. More reliable when the next step needs to use specific pieces of the output. Ask step 1 to output JSON, parse it with Make's JSON parser module, then pass individual fields into step 2's prompt.

In Make.com, add a "JSON — Parse JSON" module between AI steps when you want structured output. Map the JSON fields by name in subsequent modules — no guessing, no fragile text parsing.

Build: Notes → Outline → Draft → Edit → Doc

Create a Google Form with a single textarea field: "Paste your raw notes." This is the starting point.

Module 1: Google Forms — Watch Responses. Triggers when someone submits the form.

Module 2: OpenAI — Create Outline (JSON output).

Convert these raw notes into a structured article outline.
Return JSON only:
{
  "title": "compelling article title",
  "intro_hook": "1 sentence opening",
  "sections": [
    {"heading": "...", "points": ["..."]}
  ]
}
Notes: {{form_response}}

Module 3: JSON Parse. Parse the JSON output. This gives you clean title, intro_hook, and sections fields.

Module 4: OpenAI — Write Draft. Pass the parsed outline into a writing prompt:

Write a clear, engaging article based on this outline.
Style: conversational but professional. No filler. ~500 words.
Start with this hook: {{intro_hook}}

Outline: {{sections}}

Module 5: OpenAI — Edit. Pass the draft to a final editing step: "Edit this article for clarity and conciseness. Remove filler phrases. Shorten weak sentences. Return the edited article only."

Module 6: Google Docs — Create Document. Use the title from the parsed JSON as the document name. Paste the edited article as the body. The document lands in a "Published" folder automatically.

Go Further on Your Own

• Add a step 6 that generates SEO metadata: title tag (under 60 chars), meta description (under 160 chars), and 5 target keywords
• Add a conditional branch after the edit step: if the article is over 700 words, also generate a 50-word TL;DR summary
• Build a Python version of this pipeline using the Anthropic SDK — no automation tools, just direct API calls in sequence

Nice work. Keep going.

Day 4 is ready when you are.

Want live instruction and hands-on projects? Join the AI bootcamp — 3 days, 5 cities.

What's Next

The foundations from today carry directly into Day 4. In the next session the focus shifts to Day 4 — building directly on everything covered here.