Why the outline should be generated before the article
Writing the article is not what slows this work. Switching between tabs for twenty minutes is. In that time you work out what already ranks, what those pages contain, and which angle is still open. Editors do not enjoy assembling that brief forty times.
Producing the article in one pass avoids the problem instead of solving it. You end up with 1,200 words on a structure you would have refused. Rejecting that structure means rejecting the article as well.
Automate the task by splitting it. The first step creates headings only. You then approve or replace them. The second step fills the approved structure. Costly generation never runs against a structure you did not want.
Title options and the meta description come last. They are written beside the final text, not the brief. That is why they describe the article that exists, not the one you intended.
How the ActionFlow runs on the canvas
A Start node takes the keyword, the audience in a sentence, and an optional word count. Start Scheduled replaces Start when you work through a keyword list on a cadence.
The first Generate Object node returns the outline as a structure, not prose: an array of headings, each with the question it answers. Returning the outline as an object matters because the next steps loop over that array.
The outline is held for approval by a Human in the Loop node using Slack, Teams, or Email. This is the cheapest check, because rejecting five headings costs much less than rejecting a complete draft.
After approval, the Loop node walks the heading array and makes one Generate Text call per section. Writing one section at a time keeps each call short and focused, so you can regenerate a poor section on its own.
A Merge node assembles the sections into one document. A final Generate Object produces title options and the meta description from that text, returned as fields you can write straight into a CMS.
Google Docs or Notion receives the finished draft. Google Sheets logs the keyword, slug, and run so an editor has a queue rather than a folder.
Each AI node has its own model service. Section writing can run on a frontier model and the metadata step on a faster one. The Flow Requirements panel displays an estimate of the total cost before the flow runs.
Nodes this ActionFlow uses
- Start: receives the keyword and audience. Start Scheduled runs a keyword list instead.
- Generate Object (AI Core): returns the outline as a structured array of headings rather than prose.
- Slack (Human in the Loop): holds the outline for approval before any long-form generation runs.
- Loop (Control): walks the approved headings and generates one section at a time.
- Generate Text (AI Core): writes each section against its own heading and question.
- Merge (Control): assembles the sections into a single document.
- Google Docs (Integration): receives the draft. Notion works the same way.
- Google Sheets (Integration): logs keyword, slug, and run so the editor has a queue.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.