Medium: The AI-assisted workflow I use as a writer

Published in March 2026, Darren Moss writes from his position as an editor at a major UK publisher and describes a workflow built around a clear division: AI handles the cognitive load of research, SEO logistics, and initial editorial feedback; Moss handles the actual writing. The article is unusual in that it is specific about which tools do which tasks, rather than making a general case for AI-assisted writing.

The three-tool setup

Moss uses NotebookLM as a research compression tool. He feeds documentation into it and uses the podcast-style output to get across a topic faster than reading everything in full. The purpose is not to replace reading but to prioritize it — to know which sources are worth reading closely before he commits the time.

For content planning, he uses a customized Gemini Gem configured as a content advisor. Given a brand’s guidelines and keyword strategy, the Gem generates SEO content briefs at scale. The point is to move brief creation from a task that requires writing attention to a task that requires editorial judgment. A writer receives a brief that already reflects brand voice and search intent, and their job is to produce content that serves the brief — not to figure out what the brief should be.

The third tool is a second Gemini Gem configured as a sub-editor. Moss trains it on brand examples and style guides so it can provide feedback that is specific to the publication’s standards rather than generic. He describes it as a first-pass editor that can be consulted before the work goes to a human review cycle.

The privacy constraint

Moss uses the Google Workspace Enterprise tier for both Gemini tools. This is not incidental — he notes that the Enterprise tier prevents proprietary content from being used in model training, which matters when the content being run through the system includes unpublished editorial material, client briefs, and style guides. For writers working in professional publishing contexts, the distinction between consumer and enterprise AI tiers has practical implications for content security.

Why the framework holds

The article’s central statement is that the workflow preserves what he values about writing: “I don’t use AI to write for me; I use it to handle the cognitive load of research, SEO logistics, and initial sub-editing so I can focus on the ‘human’ part of the story.” The division is not ideological but practical. Research compression and brief generation have identifiable inputs and outputs; AI can handle them reliably. The judgment about what a story should say and how it should say it is harder to specify, and that is what he keeps.

Who it is useful for

Editors and content leads working in publishing or content marketing who are trying to reduce process overhead without compromising editorial quality. The specific tool choices — NotebookLM for research, Gemini Gems for briefs and feedback — are practical enough to evaluate and adapt for different organizational contexts.