How I use AI to run 4 publishing projects simultaneously

I run four publishing projects at the same time. Two newsletters, two blogs. Each has a distinct voice, a publishing schedule, and a specific audience. None of them have a team behind them.

That setup was not practically possible five years ago. It is now, because AI has changed what one person can maintain without the work becoming unmanageable. This post is about exactly how — the specific tools, the daily workflow, and the places where AI helps versus the places where it still can't.

This isn't a pitch for AI as a productivity revolution. It's a description of what actually works when you're running multiple publications with no support staff and finite time.

The four projects

I'll keep this vague on specifics — the projects are separate from this blog — but the structure is relevant. Two of the publications are in different niches and have distinct editorial voices. One is long-form and research-heavy. One is shorter and more opinion-led. The newsletters associated with each have different frequencies and formats.

The constraint that matters: each project needs original content produced regularly, published on schedule, and written in a voice specific to that publication. AI has to serve four different editorial identities, not one.

The core workflow

Monday: planning and briefing

Monday is the day I interact most with AI. I don't write on Mondays — I brief. For each project, I produce a detailed brief for the week's content: topic, angle, target reader, tone notes, structure suggestions, things to avoid, word count. The better the brief, the better the output.

These briefs go into Claude. I have a separate conversation thread per publication — this is important. Claude holds the context of previous conversations within a session but not across them, so I keep each publication's work in its own thread and front-load the relevant voice notes at the start of each session.

Monday also covers editorial planning: what's coming up, what needs to be commissioned, what the internal linking opportunities are across the four sites.

Tuesday and Wednesday: drafting

Claude produces first drafts based on Monday's briefs. I don't use it to write everything from scratch in one pass — I use it iteratively. A structural draft first, then a detailed pass on each section, then a final read where I make the edits that only a human familiar with the publication can make.

The ratio of Claude output to my editing varies by project. The opinion-led newsletter is maybe 60% Claude, 40% me. The research-heavy blog is closer to 40% Claude, 60% me — the factual density requires more hands-on work. Neither is fully automated and neither should be.

Research: before the draft

For any post where search traffic matters, I run a Frase brief before Claude touches it. Frase tells me what the top-ranking pages cover, how long they are, and what questions they answer. That brief becomes part of the prompt I give Claude — so the draft is structurally informed by what actually ranks, not just what seems logical.

For anything requiring current information, Perplexity runs before Claude. I pull the relevant facts, paste them into the Claude prompt as context, and Claude works with that material rather than relying on its training data.

Organisation: keeping four projects straight

The operational risk of running multiple publications isn't the writing — it's the organisation. Missing a deadline, publishing the wrong piece to the wrong outlet, losing track of what's been commissioned versus what's been published.

I use Notion for project management across all four. Each publication has its own database with a content calendar, draft statuses, and publishing notes. Notion AI earns its keep here not as a writing tool but as a summarisation and organisation tool — I'll ask it to pull a summary of what's outstanding across projects, or to reorganise a content calendar when something shifts.

Publishing and scheduling

Each platform has its own scheduling tool — Ghost, beehiiv, Substack each handle this natively. I don't use AI at the publishing stage. The scheduling, meta descriptions, and SEO titles are written by hand because they're short enough that it's faster than prompting.

Where AI doesn't help

It's worth being direct about this, because workflow posts tend to oversell the automation and undersell the work that remains.

Editorial judgment

Deciding what to publish, what angle to take, what the publication's position on something should be — none of this is delegatable to AI. Claude can generate five angles on a topic. Choosing the right one for a specific publication and audience requires understanding that publication and audience. That understanding doesn't live in a prompt.

Voice development

Claude can match an existing voice reasonably well. It cannot develop one. The voice of each of my publications emerged from years of writing in that niche, for that reader, with that editorial stance. AI maintains and extends that voice. It didn't create it and it couldn't.

Relationship with readers

Newsletters in particular live or die on the sense that there's a real person behind them. The replies, the occasional personal note, the adjustments that come from reading reader feedback — none of that is automatable. The human layer at the end of the process is what makes the work worth receiving.

What this actually costs

The AI stack that makes this possible runs to £87/month. Against four revenue-generating publications, that's straightforward to justify. The more useful framing is time: without this stack, running four publications at a sustainable quality level would require either hiring help or working unsustainable hours. Neither was the plan.

My full AI stack with costs and verdicts: https://thepracticalai.digitalpress.blog/my-ai-stack-2026/

Claude for Writers: the full review: https://thepracticalai.digitalpress.blog/claude-for-writers-review/

How to trial AI tools before committing to a subscription: https://thepracticalai.digitalpress.blog/how-to-trial-ai-tools/

— Ellis

This post contains affiliate links to Claude Pro, Frase, and Notion AI. I pay for all subscriptions mentioned. Full disclosure at disclosure.