(C) BoingBoing
This story was originally published by BoingBoing and is unaltered.
 . . . . . . . . . .



A $5 AI book draft that fools writing detectors [1]

['Ellsworth Toohey']

Date: 2026-03-21  

Andrew Wheeler's previous book took a year to write. His new one, LLMs for Mortals, took two months and cost about $5 in API fees, he writes on his blog. About half of the final text is original prose from Anthropic's Sonnet 4.1, lightly copy-edited over 20 to 30 hours per chapter. When he ran the finished book through Pangram, an AI writing detector, none of the passages were flagged.

Wheeler's technique relies on Claude Code's 500,000-word context window. He feeds his prior writing — blog posts, papers, notes — as k-shot examples so the model mimics his style rather than defaulting to its own. Then he gives it an outline and lets it draft from there. He works in plain text and Markdown, not Word, because formatting-heavy documents confuse the model.

He splits LLM-assisted writing into two use cases. Copy editing is easy — LLMs catch grammar and phrasing issues that spell checkers miss. Writing new content is harder because the model can be wrong, and it has a recognizable voice. Wheeler's fix for the voice problem is a Claude.md instructions file that says things like "do not include emojis" and "avoid excessive lists and verbosity."

For academic papers, he puts citations in a BibTeX file and tells the model to reference them inline.

"LLMs are a tool — they can be used poorly, but I think they can be used quite well," Wheeler said.

Previously:

[END]
---
[1] Url: https://boingboing.net/2026/03/21/a-5-ai-book-draft-that-fools-writing-detectors.html

Published and (C) by BoingBoing
Content appears here under this condition or license: Creative Commons BY-NC-SA 3.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/boingboing/