I Trained AI to Write Like Me - And It Actually Worked

A repeatable AI writing system can produce 90% of a creator’s long-form drafts in their own voice—if the model is trained on a dedicated style guide plus the creator’s best writing samples. The payoff is time: the workflow turns what used to take days of drafting and editing into a process that often finishes in 10–15 minutes, leaving only a light human pass for accuracy and personal details.

The method starts with selecting a broad set of recent long-form essays—20 samples—so the system can learn the range of writing patterns the author uses. Rather than relying on a one-off prompt, the AI (Claude, in the example) proposes two core “tools” it needs to replicate that writing: a style guide and an implementation guide. The style guide captures the voice and preferences—frameworks, dos and don’ts, and examples of what the writing should sound like and what it should avoid. The implementation guide turns those rules into an operational process the AI can follow every time.

In practice, the workflow is built inside Claude “Projects,” where a project acts like a container for documents. The author uploads a master document—effectively the style guide merged with the implementation guide—so every new draft automatically draws on roughly 20,000+ words of writing rules. When generating a new piece, the user answers a short set of questions: topic, target format (a long-form thought piece for a blog), specific emphasis (direct, concrete impacts already happening in everyday life), and length (about 2,500–3,000 words). The system also supports using the author’s own notes to add specificity.

For a climate-change essay, the author supplies excerpts from a book (“The Heat Will Kill You,” finished recently) rather than asking the model to rely on general knowledge. Those highlights are synced from Kindle to Evernote via Readwise, producing a large context bundle (over 5,000 words, about 10 pages). Claude then generates a draft that the author typically finds “90–95%” complete. The author does one or two editing passes to ensure the piece is truly personal and publishable.

Editing is handled through versioning and targeted instructions. If the opening anecdote isn’t genuinely personal, the author requests a replacement with a real story from their own life. If a paragraph feels unclear, they can highlight it and ask for an explanation. If the tone drifts into overly academic language, they can request a less formal rewrite. The author also verifies structure and content against the highlights they provided, noting that the AI organizes impacts into logical sections—social, environmental, economic, and immediate effects on daily life—matching the emphasis requested.

The broader business lesson is that writing value is uneven. The author describes producing many pieces where only part of the work requires their unique perspective; other parts—like turning outlines into prose or transforming book takeaways into publishable drafts—can be delegated. The result is a leverage shift: more time for the “meaningful half” of writing and less time on the mechanical half. A 15-minute prep checklist is offered as the next step: collect best writing, spot patterns, capture the unique style, and assemble the materials needed to build the system.