Playbooks, not prompts: the right way to write a book with AI

I used to keep libraries of prompts and subscribe to every tool. Most authors writing with AI start the same way. Two weeks later you have forty fragments in nine voices, none of them connected, none of them sounding like you.

The problem is not the prompts. A prompt is a one-off request, and a book is four hundred connected decisions about structure, argument, evidence, and voice.

Prompts do not carry decisions forward. Playbooks do.

The research backs the process people. In a controlled MIT experiment, professionals using AI on writing tasks finished about 40% faster with an 18% jump in rated quality, and the biggest gains went to writers who edited and restructured the output inside their own process (Noy and Zhang, Science, 2023). Then the warning label: a Harvard field experiment with 758 consultants found AI lifted quality dramatically on tasks inside its range, and made users 19 points more likely to get it wrong beyond it (Dell’Acqua et al., 2023). Same tool, opposite results. The difference was structure.

A playbook is that structure written down. Ours runs from discovery interviews through a book strategy brief, framework mapping, chapter drafting, editorial passes, and activation. Every stage has a defined input, a defined output, and a human checkpoint. Here is where the checkpoints sit.

Before drafting, a human locks the strategy. What the book claims, who it is for, and the job it does in your business. No model touches those calls.
Inside every chapter, a human reads every pass. Are the claims supported, are the examples true, do the sentences sound like you.
At the end, a human runs the logic audit. The hunt for claims asserted but never supported and ideas introduced but never developed. AI-heavy drafts fail here more than anywhere else.

A blog post from a lucky prompt costs you an afternoon. A manuscript assembled from three hundred lucky prompts costs you a year, because the failure only shows when you read the whole thing and it does not hold together.

And voice? Voice consistency is a process property, not a prompt property. You cannot paste “write in a warm, direct tone” into a chat window and get your rhythm back. You can build documented voice rules from interviews and writing samples, apply them at every drafting pass, and check them at every edit. Do that across a manuscript and the book sounds like one person wrote it. You.

The goal is not to have AI write for you. It is to train AI to think with you.

Now I ask one question before reaching for any tool: what problem am I solving today? Tools multiply. Problems stay countable, and a playbook is the written record of how you solve the ones that repeat.

If you have a body of work and no book, or a stack of AI fragments that refuses to become one, this is the problem our playbook exists to solve. Start a Conversation.

Sources

Noy, S. and Zhang, W., Science, 2023. A randomized experiment on professional writing tasks found generative AI cut completion time by about 40% and raised rated quality roughly 18%, with the strongest gains among writers who edited and restructured the output. https://www.science.org/doi/10.1126/science.adh2586
Dell’Acqua, F. et al., Harvard Business School working paper, 2023. A field experiment with 758 BCG consultants found large gains on tasks inside AI’s capability frontier and a 19-point rise in wrong answers beyond it, the case for structured human oversight. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321