Where it fits my work
My local n8n prototype used Gemini to help prepare copy from collected research. The workflow connected sources, structured records, and content branches. Its retained diagram shows the setup; it has not been retested as a current production service.
How I would improve it
I would keep the original link, publication date, extracted facts, and draft text in separate fields. I would give Gemini a fixed output format and require it to flag missing evidence.
What I would check
I would check dates and figures before approving a draft, detect duplicate stories, and stop records with missing sources from moving forward. An automated draft still needs a human review.
The next measure
I would compare a manual run with an assisted run on the same articles, including correction time. I have not yet measured a reliable time saving for this workflow.
