Watch the video with chapters & transcript →
AI TOOLS
In short: Gemini 3 Pro preview generated an importable n8n workflow JSON from a single prompt: run daily, read an RSS feed, limit to ten items, append to Google Sheets.
This experiment tests whether Google's new Gemini 3 Pro preview, a free model released about a week before the recording, can build a working n8n workflow on its own. The use case is content research: a daily workflow that pulls the most viewed recent YouTube videos on n8n, AI tools, and software topics into a Google Sheet, so a creator can see which subjects are getting traction.
You ask for the workflow explicitly in JSON. The prompt used here says: create a complete n8n workflow in JSON, run daily with a schedule trigger, read an RSS feed from a given URL, limit the results to ten, and append the items to a Google Sheet. Gemini returns the JSON along with an explanation of the nodes, and a download button saves the file, which is then imported in n8n via the three-dot menu and "Import from file".
Three of the building blocks arrived correctly configured, but the workflow did not run as imported. The RSS Read node existed only as a placeholder and had to be added by hand, and an extra Set Columns node was auto-connected in the wrong place, which blocked execution until it was removed.
| Component | Gemini's output | Manual work needed |
|---|---|---|
| Schedule trigger | Created and configured | None |
| RSS Read node | Referenced but not created | Add node, paste feed URL, connect to trigger |
| Limit node | Correct: max items 10, keep first items | Reconnect after RSS fix |
| Google Sheets node | Created | Connect account and map all fields |
| Set Columns node | Added unnecessarily | Delete it |
The RSS feed is completely free and needs no API key, yet it still surfaces the most hyped recent videos. The feed URL used here searches for keywords such as n8n, AI tools, and software, with the location set to US; you can swap that country code, for example SE for Sweden, and even change the language. A spot check confirmed the data was live: the top result had been published 13 hours earlier and showed 495 views on YouTube.
Not yet, at least with a short prompt. Ani Björkström rates the generated workflow around 80 percent accurate: it clearly made the build faster, but the missing RSS node, the stray Set Columns node, and the Google Sheets field mapping, where title, link, publish date, content, content snippet, and ISO date were dragged in by hand, all required a person in the loop. A more detailed prompt, for example one drafted with ChatGPT, might raise that score.
No. The model was free at the time of recording, and the RSS-based workflow it generates also avoids any paid API.
In this demonstration n8n runs locally, which is free. The generated JSON is imported through the "Import from file" option.
Create a sheet in Google Drive first, then in the Google Sheets node choose that document, here named Gemini 3.0, select Sheet1, and drag each field from the Limit node into its column.
Read the full video transcript, with chapters, on the video page →
Want this working inside your finance team?