The challenge
Most AI writing sounds alike. It leans on the same inflated vocabulary, the same tidy three-part lists, and the same confident conclusions that are more settled than the thinking behind them. If I was going to use AI for real correspondence, proposals, and teaching material, the drafts had to sound like a person I’d recognize.
The approach
I wrote my voice down as a reference library and released it as version 1.0 on August 1, 2026. A controlling style guide sits at the top. Beneath it are a drafting and revision workflow, a set of voice profiles for different relationships, and focused guides for sales writing, technical writing, executive work, presentations, correspondence, and teaching.
Each file carries its version number in its name. When I revise a standard, the new version goes beside the old one instead of replacing it, so I can always see what changed and why.
The development
Two minor releases followed the first. Version 1.1, in August 2026, added guidance on nominalizations, long sentences that pile up claims, and prose that sounds neater or more certain than the underlying thought. Version 1.2, in September 2026, added a short set of release-blocking safeguards against the habits I found most often in AI-assisted drafts.
One principle runs through all of it. Characteristics that look like AI writing are editing clues, never proof of who wrote something. The question is always whether a sentence serves the reader.
The deliverables
- A style guide and a drafting workflow
- Voice profiles for executive, sales engineering, teaching, advising, family, and informal writing
- Discipline guides for the kinds of writing I do most
- Standards for reasoning, sentence rhythm, and final review
- A changelog covering three releases